# Hypernatural

> Hypernatural makes AI video. Hypernatural Desktop is a free Mac app that connects to Claude Code, Codex and Claude Desktop and gives the agent every AI video, image, voice and music model, free editing tools, and workflows and skills for making videos.

The workflow and skill pages below return markdown, with instructions for agents, when requested with `Accept: text/markdown`.

## Hypernatural Desktop

- [Hypernatural Desktop](https://hypernatural.ai/): the app, what it does and how to install it
- [Workflows](https://hypernatural.ai/workflows): end-to-end video jobs an agent can run, from the brief to the finished cut
- [Skills](https://hypernatural.ai/skills): how-tos for each step of making a video, and techniques for each model and tool
- [Pricing](https://hypernatural.ai/pricing): free app and tools; pay only for AI generations

## Workflows

- [Script to video](https://hypernatural.ai/skill/hn/workflow-script-to-video): Make a finished narrative video from a written script: adapt it into script.json, cast and perform the voices, design the sets, stage the cast in stills, compose every shot with its dialogue spoken in the cast voices, composite titles over the footage, then make the cuts and deliver. For scripted shorts, skits, training dramas, quiz shows and multi-character scenes made from nothing but text.
- [Product explainer video](https://hypernatural.ai/skill/hn/workflow-product-explainer-video): Make a ~35 s faceless marketing explainer for an app or product from its own code repo or website: a voiceover over kinetic type in the brand's colours and the product's real screenshots in a phone frame, with a music bed, built as one hn4a-compositor spec cued from the voice take's own word times. Problem, brand, three how-it-works steps, one feature, social proof, end card. Covers mining the repo for copy, facts and images, a script and aspect gate, the opening scene in a few looks on estimated word times shown live minutes after the script, beside the voice and music auditions, then the whole layout in the picked look cued to the picked voice, shown live before the render. Costs well under $1. For an ad with a presenter avatar and rebuilt UI, use workflow-product-ui-ad.
- [Animated music video](https://hypernatural.ai/skill/hn/workflow-animated-music-video): Make a fully generated animated music video from a song: map the song to its bar grid, write a treatment, pick a drawn look from style frames, cast the characters on model sheets, write the shot list on the bars, make one still and one animated clip per shot, cut every shot exactly on its bars, then add kinetic lyric captions that land on the sung syllables and pulse on the beat. For cartoon and drawn-style music videos without lip sync; for a multi-look video with live-action scenes and singing on camera, use workflow-multi-style-music-video.
- [3D-animated product ad](https://hypernatural.ai/skill/hn/workflow-3d-animated-product-ad): Make a 20–30 s 3D-animated product ad from a brand's website and product photos: the product itself becomes the hero character (a can that walks, shops and reacts), with its label kept exact, in a stylized or real-looking world that acts out what the product promises, narrated, with a music bed and titles in the brand's font, colours, logo and CTA, delivered at 16:9 and 9:16. This product-character recipe owns its production sequence: brand and facts from the site, an ad script with a hook and a story world, a gate with a price, the product character's sheet, stills, one motion test, the batch on H3 Max, voice and bed, the base cut, the brand's titles shown live, and both aspects. For a drinks, snack, beauty or household product, or software with one clear object to animate; for a handmade clay look use workflow-website-to-claymation-ad.
- [Add captions and motion graphics to video](https://hypernatural.ai/skill/hn/workflow-add-captions-and-motion-graphics-to-video): Take footage someone already has (a webcam or talking-head video, a creator short, a cut of best takes) and make it more exciting in their brand: captions in a chosen style, kinetic titles, type behind the person, titles that react to their face and hands, graphics for the words, punch-ins and shakes on hits, and synthesized sound effects, at the intensity they pick from captions-only to go wild. Free and local; the first build takes about an hour. Use for 'add captions and motion graphics', 'make my video pop', 'go wild with graphics', augmenting shorts, reels and webcam videos.
- [Animated learning video](https://hypernatural.ai/skill/hn/workflow-animated-learning-video): Make a narrated animated learning video (1–3 min) from a lesson plan, a topic or a doc: one or two recurring guide characters teach one idea per scene in a drawn 2D, 3D or paper-cut look, with a recap and a question to check understanding, diagrams and key words composited on top, and calm captions. Covers the learning beats and their gate, the look and the characters' sheets, a set the scenes share, stills, one motion test then the batch, the narration with word times, the base cut, the diagrams and type cued on the words, the live gate and delivery. For kids' lessons, classroom and training explainers, onboarding and how-it-works lessons; for a 45 s social retelling of a post use workflow-blog-post-to-animated-short.
- [Blog post to animated short](https://hypernatural.ai/skill/hn/workflow-blog-post-to-animated-short): Turn a blog, Substack or newsletter post into a ~45 s vertical (9:16) animated short: retell the post as a voiceover shot list, cast a narrator, design one stylized set (paper-cut by default) and a fictional character, stage and approve a still per shot, animate on H3 Max at 1080p, cut to the voice, then add calm word-by-word captions and motion-graphic cards in the publication's brand with hn4a-compositor, shown live before the render. About $11–12 of model spend; the cut, captions and cards render locally for free. For a news story with sourced facts and a hook, use workflow-news-article-to-vertical-short.
- [Board to summary video](https://hypernatural.ai/skill/hn/workflow-board-to-summary-video): Turn a Figma or FigJam board (feedback, research, a retro, a dogfood wall of screenshots and stickies) into a narrated summary video: read the board, group its items by topic and rank them, write and voice a script, rebuild the board as one cluster per topic, and fly a camera across it in time with the voice. Uses the board's own images; no generated footage.
- [Feature guide video from how-to page](https://hypernatural.ai/skill/hn/workflow-feature-guide-video-from-how-to-page): Turn a feature's how-to page (a help article, a docs page, a skill's page) into a 60-75 s narrated video guide to embed on that page, for people who'd rather watch than read: a before/after demo shot up front (the same generated shot without the feature, then with it), then kinetic type and animated diagrams in the brand that show every thing the page describes, with each one actually played or shown as it is named, a music bed and an end card with the page's address. Covers checking every claim against the page, a script gate with what you see and hear per beat, the demo shot from a start and an end frame, the opening scene in three looks beside the voice and music, the demo's assets picked on the shot itself, one build script that cues every layer and sound from the take's word times, and the mix. Costs about $2. For a marketing explainer of a whole product, use workflow-product-explainer-video.
- [GitHub repo to explainer video](https://hypernatural.ai/skill/hn/workflow-github-repo-to-explainer-video): Turn a GitHub repo into a 30–60 s branded explainer: a stylized metaphor world that acts out what the project does, a narrator, and titles, chips, captions and an end card in the project's own brand that name the real things it ships (commands, modules, features), ending on its install command. Covers reading the repo itself rather than only the README and checking every claim against it, capturing the brand from the project's site, a script and shot list gate, three looks, a character turnaround, voice with word times from the TTS alignment, stills, H3 Max animation, the rough cut, and the compositor layout shown live before the render. For gift videos to open-source maintainers, repo launch videos and "what this project does" explainers; for a product's own app screens, use workflow-product-explainer-video.
- [Green screen news spot](https://hypernatural.ai/skill/hn/workflow-green-screen-news-spot): Put a person's own webcam read in a TV news studio: a sourced story and script, the read recorded and cut to its best takes, the person cut out with a free local matte (no green screen needed), a generated news-set backdrop with a city window and a video wall, the presenter scaled and lit into the set, and a broadcast chyron (kicker, station bug, name and headline straps, LIVE bug, a ticker already scrolling) with an optional skyline opener, in 16:9 and 9:16. About a dollar of stills; the cut, matte and graphics are local. For a narrated explainer with no presenter, use workflow-news-article-to-vertical-short.
- [Highlights from webcam footage](https://hypernatural.ai/skill/hn/workflow-highlights-from-webcam-footage): Turn a raw webcam or talking-head recording full of retakes into a tight rough cut of the best take of every line, approved on its own, then ask what the footage is for and carry on in the workflow for it: captions and motion graphics that react to the speaker (the 'cut my takes and go wild' request), a green-screen news spot, short clips, or just the clean cut. Free and local, about 10 minutes for the cut. Use for founder and creator recordings, launch and feature videos, and any pile of webcam retakes.
- [Launch video](https://hypernatural.ai/skill/hn/workflow-launch-video): Make a 30-90 s launch video for something new (a product, a feature, a release) presented by its founder or a teammate: either they record a quick webcam read of the approved script, or an AI avatar of them presents it, made from their own photos and voice and labelled as AI on screen. Then the product on screen, kinetic titles in the brand, karaoke captions, a name strap and the CTA, in 16:9 and 9:16. Covers what's new from the changelog or site, the launch arc (the problem, 'today we're launching', what's new, proof, the call to action), the question that picks the presenter, a storyboard before any spend, and the live spec. Use for launch videos, product announcements, feature drops, release videos and founder videos.
- [Mascot product ad](https://hypernatural.ai/skill/hn/workflow-mascot-product-ad): Make a ~30 s product ad starring the brand's own mascot: a narrated character story expressing the product's promise, with the product's real screens, integration logos and key-phrase captions composited on top. Covers breaking down the product's points (and its existing launch video, if any), a concept and shot list gate, voice auditions on the whole script, one style frame from the mascot art, stills, one motion test, H3 Max animation, one narration file per line, a music bed, the base cut, and the motion graphics shown live before the render, delivered with a side-by-side against the original. About $10 at 1080p. For SaaS, dev tools and apps with a mascot, made from the brand's own materials; for a presenter over rebuilt UI use workflow-product-ui-ad, for a faceless type-and-screenshot explainer workflow-product-explainer-video.
- [Microdrama episode](https://hypernatural.ai/skill/hn/workflow-microdrama-episode): Make one episode of a vertical soap-style microdrama (Part 1, 2, …) from a premise in a sentence or two: write a hook-first script, cast faces and voices, record the lines and the lead's inner voiceover, build location plates, render each scene as one long Seedance 2.5 reference-to-video take driven by the cast, the locations and the scene's whole voice track, then cut, fix in-world text, title, score, caption and finish. Reach for it for microdramas, vertical drama series, soap or revenge shorts, a cliffhanger episode for TikTok or Reels, or the next part of one.
- [Multi-style music video](https://hypernatural.ai/skill/hn/workflow-multi-style-music-video): Make a storytelling music video from a song and photos of the cast, switching looks scene to scene (photoreal, comic-book animation, 3D animation, paper collage) with the cast consistent across styles, lip sync on key lines, a rough cut on the song, and kinetic lyric captions styled to each scene. Use for music videos, lyric videos and mixed-media multi-style shorts.
- [Multi-style video ad for social media](https://hypernatural.ai/skill/hn/workflow-multi-style-video-ad-for-social-media): Make a 15–30 s 9:16 social ad that changes its look at the story's turn: it opens in one world (often real-looking people in a real place) and switches to the product's world in another look (a pastel dream, 3D animation, paper collage, ink comic), with the same cast recognizable across both, narrated or scored, with captions and the brand's end card, delivered at 9:16 and 16:9. Covers the brief and facts, an ad script with a hook, picking a look per part, a cast sheet per person per look, stills, shots on the model that suits each look, voice and bed, the base cut, titles and the live gate. For brands that want a scroll-stopping social ad; for a concept whose treatment is not yet chosen, use guide-make-a-video-from-a-website.
- [News article to vertical short](https://hypernatural.ai/skill/hn/workflow-news-article-to-vertical-short): Turn one news article into a ~45 s vertical (9:16) explainer short: a sourced script with the conflict in the first 3 s, 9–10 stylized visual-metaphor scenes animated on H3 Max, a quick creator-style narrator, then karaoke captions, a title bar and stickers that slam in on the key words, composited with hn4a-compositor in a chosen brand. Covers picking the story with the person (and its risks), keeping only facts traceable to the source, no likeness of real people, three looks, stills, voice auditions on the whole script with word times, shot timings on the 24 fps grid, locking what must not change in the video prompts, the cut and the live layout gate. About $7 and an hour. For same-day news shorts, newsletter-to-video and social explainers about a story; for a post's own argument retold as a paper-cut story, use workflow-blog-post-to-animated-short.
- [Pitch deck to video](https://hypernatural.ai/skill/hn/workflow-pitch-deck-to-video): Turn a finished pitch deck (PDF, PPTX, Keynote or images) into a 1-3 min presenter video for investors, prospects or partners: the real slides shown as they are, a presenter beside them (an AI avatar of whoever is pitching, made from their photos and voice, or the person on camera) walking through each slide, in a mix of cutout, picture-in-picture box and bubble layouts that never repeats twice in a row. Covers the slides as images, a script per slide (the pitch arc when there are no speaker notes), the presenter and voice, the whole track lip-synced in chunks, the layout first look, and the live spec. For a personalized sales video built from a deck template or a client-details doc, use workflow-sales-video-from-pitch-deck; for slides rebuilt as motion graphics, workflow-slide-deck-presenter.
- [Podcast promo](https://hypernatural.ai/skill/hn/workflow-podcast-promo): Turn a podcast episode, talk, interview or live session (video or audio) into 30-60 s promo clips in the show's brand: the moments picked from a diarized transcript with a hook a cold viewer gets, each bite cut in its natural pauses and levelled with one fixed gain, the speaker framed vertically (or an audiogram for audio only), karaoke captions, a guest name strap, the show's bug, stickers on the key phrases, designed title cards and an episode end card, with an optional bed. Nearly free: local cuts and renders, plus paid speaker-labelled transcription if there are several voices. For long screen recordings cut into split-screen shorts, use workflow-shorts-from-long-video.
- [Product ad from product photos](https://hypernatural.ai/skill/hn/workflow-product-ad-from-product-photos): Make a 15-30 s photoreal product ad from nothing but the product's photos (and its website, if it has one): a short script built on a moment of need and a reveal, a cast of invented people and a set designed once, stills staged with the product held right and its label exact, one motion test, the shots on the photoreal models, a voice or a music bed, the base cut, film grain where skin reads plastic, and the brand's end card, in the aspects it ships in. For consumer products, food and drink, pet, beauty and household goods; to copy an existing spot instead, use workflow-shot-for-shot-remake.
- [Product UI ad](https://hypernatural.ai/skill/hn/workflow-product-ui-ad): Make a ~30 s product ad from a company's website or product screenshots: a presenter avatar talks to camera, then the product's own UI, rebuilt as live HTML layers, is pushed into by a camera while the presenter shrinks to a face bubble, all cut to one voice track. Covers capturing the site (copy, brand from computed styles, every image, the logo), a 72-word script, a storyboard review before anything is paid for, the voice, presenter stills and two H3 talking heads (16:9 and a 3:4 crop for the bubble), one subagent per rebuilt screen, the cut and the review rounds. For SaaS ads, app demos, product overviews and explainers; for an avatar walking through slides, use workflow-sales-video-from-pitch-deck; for a faceless voiceover-and-type explainer over the real screenshots, use workflow-product-explainer-video.
- [Realistic product ad](https://hypernatural.ai/skill/hn/workflow-realistic-product-ad): Make a photoreal product ad, real-looking people and places with the product exact in their hands: first ask whether there is a reference spot to copy. With one, remake it shot for shot with the person's product swapped in (and new actors if wanted); with only a website and product photos, generate the ad from scratch around a hero beat and a reveal. Either way the label stays exact through every still and clip, the shots render on the photoreal models, and the brand's end card closes it. Use for consumer product ads, CPG, food and drink, pet, beauty and household products, 'make an ad like this one', or a product ad from packshots.
- [Recast a music video as satire](https://hypernatural.ai/skill/hn/workflow-recast-a-music-video-as-satire): Make a satire music video: put new people into an existing music video, or a section of one, keeping the original performance, camera, cuts and song: map who becomes whom, test one shot, recast the video in parts of 5 to 15 s with H3 Max video-to-video, then fix what the recast carries over or breaks (facial hair, a missed background player, garbled lyric text) and splice it back over the original audio. Reach for it for 'put us in this music video', a fan remake with friends, or any performance clip recast from one photo.
- [Recast dialogue video](https://hypernatural.ai/skill/hn/workflow-recast-dialogue-video): Replace one actor in a finished dialogue video with a new character and a new voice: cast and perform the voice, recast that character's stills in their own outfit, re-animate only their shots with the dialogue as reference audio, and export each pick with its own audio for the editor. One variant per new actor; every other shot is reused as delivered.
- [Reframe to vertical](https://hypernatural.ai/skill/hn/workflow-reframe-to-vertical): Turn a finished 16:9 spot into a 9:16 vertical version for TikTok, Reels or Shorts: crop every shot first, restage and re-animate only the shots a crop loses a story beat in, then splice them back over the source audio. Works on an original cut or a remake from workflow-shot-for-shot-remake.
- [Sales video from Canva deck](https://hypernatural.ai/skill/hn/workflow-sales-video-from-canva-deck): Make a personalized sales video when the deck lives in Canva: from (a) a finished Canva deck, (b) a Canva deck template plus a client-details doc, or (c) a Canva brand kit or brand guide plus a client-details doc. Fill or build the deck in Canva, export the real slides, then make the avatar walkthrough with workflow-sales-video-from-pitch-deck.
- [Sales video from pitch deck](https://hypernatural.ai/skill/hn/workflow-sales-video-from-pitch-deck): Make a personalized ~90 s sales video for a named prospect from (a) a finished pitch deck, (b) a deck template plus a client-details doc, or (c) a brand guide plus a client-details doc: get to a finished, approved deck first, then an avatar of whoever is presenting (asked up front, made from their photo and voice if needed) greets the customers by name and walks through the real slides in a mix of cutout, picture-in-picture box and bubble layouts.
- [Screen recording video](https://hypernatural.ai/skill/hn/workflow-screen-recording-video): Cut a product walkthrough from paired screen recordings (no sound) and webcam recordings (with the mic): a text paper edit of the best takes, then a 16:9 video with the screen under a camera that zooms and highlights what's being said, the presenter in a corner bubble, karaoke captions, a brand motion graphic on every time jump, and finished output videos played full screen. For Loom-style demos, app tours, tutorials, onboarding and feature walkthroughs recorded as screen plus webcam.
- [Shorts from long video](https://hypernatural.ai/skill/hn/workflow-shorts-from-long-video): Turn a stream, tutorial, podcast or any long screen-plus-webcam recording into a series of 30–50 s 9:16 shorts built only from that recording: the speaker's face in one half, eased punch-ins on the right part of their screen in the other, captions on the seam in their own shorts style, and word-cued motion graphics with a different sound for each kind. Covers watching the recording and the speaker's existing shorts, a transcript with word times, picking the moments (hook, whole sentences, payoff), the rough cut, zoom targets, the base render in ffmpeg, the graphics in hn4a-compositor shown live on the first short, sound effects and loudness. Nearly free: local cuts and renders plus a few cents of sound effects. For one long walkthrough from paired screen and webcam takes, use workflow-screen-recording-video.
- [Shot-for-shot remake](https://hypernatural.ai/skill/hn/workflow-shot-for-shot-remake): Re-shoot an existing spot shot for shot with new actors or products: cut it into shots, recast the stills, swap the products, render each changed shot from pinned frames, fix the spoken lines, then grain, upscale and splice the new shots into the original cut over its own audio. Reach for it for a remake, one variant per new actor, a product swap in a finished ad, or 'copy this commercial frame by frame'.
- [Slide deck presenter](https://hypernatural.ai/skill/hn/workflow-slide-deck-presenter): Make a presenter video from a slide deck (PDF): write the narration, cast and voice it, lip-sync one locked presenter still to the whole voice track, then build motion-graphics slides around the talking head (full frame, window or corner bubble) that change and build on the spoken words. For explainers, training modules, course videos and title slides with an AI avatar presenter.
- [Website link to product walkthrough](https://hypernatural.ai/skill/hn/workflow-website-link-to-product-walkthrough): Turn a product's website and real screens into a 45–90 s narrated walkthrough of the user's journey, one step at a time. Choose recordings, held screens, reveals or rebuilt UI for what each step needs to demonstrate, with optional contextual scenes or a presenter when they serve the brand and brief; this workflow owns its script, storyboard, voice, screen build and delivery sequence.
- [Website to claymation ad](https://hypernatural.ai/skill/hn/workflow-website-to-claymation-ad): Turn a company's website into a ~20 s claymation ad delivered at 16:9, 1:1 and 9:16: five stylized toy-world shots animated on H3 Max, a narrator and a music bed, then kinetic titles, a logo lockup, benefit chips, a proof stat and an end card in the company's own font, colours, logo and CTA button. Every fact comes from the site. A worked conceptual-ad treatment in clay: a five-beat script, one locked style still, native 9:16 redraws instead of crops, one test clip before the batch, the cut on the voice's word times, and the titles shown live before the render. About $5 of model spend. For outreach gift videos, agency demos and brand promos made from nothing but a URL; for a faceless explainer over the product's real screenshots, use workflow-product-explainer-video.

## Skills

- [Add sound effects and ambience](https://hypernatural.ai/skill/hn/guide-add-sound-effects): How to add sound effects and background ambience to any video (a door creaking, birds singing, wind, a room's tone, pops and whooshes for graphics): which route to take, how to place, loop, pan and mix sounds against picture, and how to check them. Use for SFX, foley, ambience, soundscapes, background sound, room tone.
- [Add titles and motion graphics](https://hypernatural.ai/skill/hn/guide-add-titles-and-motion-graphics): Choose graphics by what the viewer needs: emphasize a message, understand a product or process, identify a speaker or context, read speech, or see an interaction with the scene. Then combine the appropriate compositor leaves for layout, captions, UI, camera movement, masking and perspective, with a first look and the shared live-spec review before rendering.
- [Animate a shot](https://hypernatural.ai/skill/hn/guide-animate-a-shot): Turn a still into a moving shot, or re-animate a shot from an existing spot: image-to-video (img2vid, i2v) from a pinned first frame, first and last frames pinned, reference-to-video (ref2vid, r2v) anchored on the shot's still or composed freely from references, or a video-to-video recast that keeps a take's performance. Weighs the routes, then cuts, briefs, renders and checks the shot.
- [Caption a video](https://hypernatural.ai/skill/hn/guide-caption-a-video): Choose whether a video needs spoken captions, selected message titles, or both, then a caption treatment suited to the listening context, brand tone and layout. Covers karaoke, calm words and seam chunks, readable sizes and placement, word timing, and coordinating captions with titles or chyrons in the first look.
- [Capture a brand](https://hypernatural.ai/skill/hn/guide-capture-a-brand): Capture a brand's creative direction and usable assets in refs/brand.md: audience, personality, tone, existing imagery, desired impression, fonts, colours, logos and usage rules. Use a website, brand kit, repo or footage to distinguish what the brand already establishes from a proposed interpretation, so the video's concept and visual treatment fit the brand.
- [Cast a voice](https://hypernatural.ai/skill/hn/guide-cast-a-voice): How to choose a voice for a character: a short brief, a search of the ElevenLabs shared library for voices verified on eleven_v4, 3 picked on their metadata and auditioned at once on part of the script's copy (a narrator for a short reads the whole script instead, and the pick is the take), and the person's pick, with a performance guide for every script. Only when the person wants a voice matched to a face or a detailed character does a listening model rank a longer shortlist first. The person's own voice can be cloned with ElevenLabs, with their consent, from about 30 s of varied speech on a good microphone. Use for voice casting, choosing a TTS voice, voice selection, auditions.
- [Choose a video model](https://hypernatural.ai/skill/hn/guide-choose-a-video-model): Choose which video model renders a shot: MiniMax H3 Max, Seedance 2.5, Kling v3 Pro, and the ones to avoid (plain H3, Veo 3.1, Omni Flash). Compares time per render, top resolution, look, the motion each handles, inputs (first frame, last frame, references, audio) and cost tier, with the default strategy: draft every shot on H3 Max, then upgrade only the shots that need 1080p, photographic realism or complex motion.
- [Cut best takes](https://hypernatural.ai/skill/hn/guide-cut-best-takes): How to turn raw recordings full of retakes, false starts and long pauses into a clean rough cut of the best takes before anything else is built: one webcam or talking-head file cut on its speech islands, or many takes and paired screen and webcam recordings cut as a text paper edit from their transcripts. Covers picking the last complete take, checking every cut inside a sentence by ear and by transcript, squeezing pauses, the frame grid, and the review before graphics. Free and local. Use for best takes, retakes, rough cut, tighten pauses, remove ums, paper edit, talking head edit.
- [Depict real people and places](https://hypernatural.ai/skill/hn/guide-depict-real-people-and-places): How to make stills and illustrations of real people and real places that a viewer recognises: list the real subjects, get photos first, write each place's facts, name everything exactly in prompts, lock location plates restyled from photos for every camera direction, stage people into them, and check every still against the photos and the facts. Use for tributes, documentaries, news explainers, sports stories, history, celebrities, public figures, famous landmarks, stadiums, arenas, city skylines and any picture of someone or somewhere real.
- [Design a set](https://hypernatural.ai/skill/hn/guide-design-a-set): How to build a location (a set) before any shot stills: write its geography down, render the main image of the empty set, derive every camera angle from it, then stage the cast into cropped plates, with brand colour that holds and a review per angle. Use for set design, environments, backgrounds, studio or game-show sets, location plates and set angles.
- [Edit a frame before animating](https://hypernatural.ai/skill/hn/guide-edit-a-frame-before-animating): Change a still before it is animated, and choose the image model: nano banana 2.1 edit (the default for recasts: natural placement and composition) or nano banana Pro edit (precise detail edits that leave every other pixel the same), a whole-frame pass or a crop-and-patch of one region. Covers actor swaps, one targeted fix at a time (hair, a prop, a pose), small print and logo fixes, removing a badge or pin, restaging a frame for 9:16 or 3:4, and conforming 2K edits to 1080p with the grain kept.
- [Keep small details](https://hypernatural.ai/skill/hn/guide-keep-small-details): Keep a logo, label, small print or a small branded product exact through image edits and video animation, product swaps included: when pure or anchored reference-to-video (ref2vid) is enough, and when to edit the frame in an enlarged crop, pin first and last frames and gate every step. Use for product placement, packaging, logos that smear or slide, and badges or pins to remove.
- [Lip-sync video](https://hypernatural.ai/skill/hn/guide-lipsync-video): Make someone speak on screen in sync with a voice track: lip-sync a still to audio (a presenter or avatar, a long script in chunks), fix only the mouth of an existing take, or render a dialogue scene with several people from stills and reference audio (ref2vid with audio). Weighs cost per second, quality, length limits, cutting the line by ear, and checking sync over time.
- [Make a script from source](https://hypernatural.ai/skill/hn/guide-make-a-script-from-source): How to get a job's script from what the person brings: a finished video, a written script or screenplay, a slide deck, or a source to retell (a website, a post, an article, a repo, a help page, a board) where you write the narration and the person approves it word for word. Ends in script.json (scenes, shots, a gendered cast, every line verbatim with a delivery note and emotion, action, beats, sound cues, all timed). Holds the length table (words per second by kind of read) and the beat-table format. Use for transcribing a video into a screenplay, script extraction, converting a script, writing narration from a source, script length, words per second, shot lists, Fountain.
- [Make a video from a website](https://hypernatural.ai/skill/hn/guide-make-a-video-from-a-website): Choose a creative approach for a video made from a company's website: a conceptual product or brand film, a product demonstration, or a product explainer. Start from the audience, purpose and brand direction, then choose the presentation and visual treatment within that approach and hand off to the appropriate workflow.
- [Pace a short for social](https://hypernatural.ai/skill/hn/guide-pace-a-short-for-social): How to shape a short video for where it will be posted: the aspect and size per platform (9:16 for TikTok, Reels and Shorts, 1:1 for a feed, 16:9 for a site), the length (20 s for an ad, 30–50 s for an explainer short), shot lengths of 2.5–6 s with the hook in the first 3, the 9:16 safe areas that keep captions and titles clear of the platform's buttons, loudness (−14 LUFS for shorts platforms, −16 elsewhere), and delivering several aspects from generated shots (action in the middle third, native 9:16 redraws rather than 608 px crops, a 1:1 crop shifted per shot). Use for TikTok, Reels, YouTube Shorts, vertical video, social ads, multi-aspect delivery, safe zones, short-form pacing.
- [Pick a stylized look](https://hypernatural.ai/skill/hn/guide-pick-a-stylized-look): Choose and hold a generated visual treatment for an approved concept, using the brand's personality, imagery and intended audience to select relevant style frames. Covers consistent prompts and references, recurring characters when needed, stills and a motion test; claymation, 3D animation and other looks are treatments, not separate video purposes.
- [Prepare renders for delivery](https://hypernatural.ai/skill/hn/guide-prepare-renders-for-delivery): What to do with picked renders before they go to the client: native 1080p or 720p plus a Topaz upscale, frame-exact trims, grain for plastic skin, splicing clips into the source cut over its audio, 9:16 vertical crops, loudness, and small and no-audio copies. Use for finishing, post, conform, upscale, trim, film grain, splice, vertical or TikTok versions, delivery files.
- [Present with an avatar](https://hypernatural.ai/skill/hn/guide-present-with-an-avatar): How to put a presenter in a video beside slides, product screens or titles: who presents (asked, never assumed), an AI avatar of a real person from their own photos and voice, a designed presenter, or the person on camera recording a quick webcam read; the presenter still (3:4, three-quarter body, room for open arms, badges removed before any render); the voice, their own cloned or a library one; the whole track lip-synced as one take in chunks, checked, retimed and joined; and the layouts the take goes into (full frame, window, cutout over a real matte, a round bubble on the side they look toward; one take, two framings; never the same layout twice in a row). Use for slide-deck presenters, sales and pitch videos, launch videos, avatar ads and walkthroughs with a talking head.
- [Pull product highlights](https://hypernatural.ai/skill/hn/guide-pull-product-highlights): How to pull what a product video says from what the person has, before any script is written: the headline, the value props in the source's own order, every number with the line it came from and the date it was read, the named parts verbatim, the CTA's exact words and the one thing the viewer must do, written to source/facts.md and checked against the person's brief. Starts by asking what they have: a website (its copy read in page order), a GitHub repo (README, releases, commits and where the marketing lives), the product running on this computer (screenshots of its key screens), docs or a help page, a details doc, or an existing launch video. Use for product highlights, value props, key features, product facts, site copy, reading a repo, product screenshots, fact checking a brief.
- [Recast a person](https://hypernatural.ai/skill/hn/guide-recast-a-person): Replace a person in existing footage with a new actor or character, and know what each route keeps: swap the still and re-animate (a new performance), a video-to-video recast that keeps the original performance exactly, or a recast plus a new voice. Covers character sheets as single views, back views for shots from behind, keeping or changing wardrobe, and scrubbing the footage for the old actor.
- [Select highlights from long footage](https://hypernatural.ai/skill/hn/guide-select-highlights-from-long-footage): How to pick the moments worth a clip from a long recording (a stream, tutorial, demo, talk, interview or podcast episode): watch it as well as transcribe it, build a sentence menu with times, choose pieces that open on a hook a cold viewer gets, run 5 to 9 whole sentences on one idea and end on a payoff, then land every edge in a natural pause and check each piece by its own transcript. Routes for one speaker, several speakers (diarized) and recordings over an hour. Free except optional paid transcription. Use for clips, shorts, soundbites, highlights, best moments, podcast clips, promo bites.
- [Show a product on screen](https://hypernatural.ai/skill/hn/guide-show-a-product-on-screen): Choose how to show real product evidence from what the viewer needs: inspect a held screen, follow an actual recording, discover parts through a staged reveal, or see editable UI states change. Then choose the framing, camera pace and highlights for readability; supports screenshots, device frames, cards, recordings and rebuilt screens without making prominence alone a reason to animate.
- [Time a video to a song](https://hypernatural.ai/skill/hn/guide-time-a-video-to-a-song): How to time a video to a song: split the song into stems, time the sung words from the vocal, map the bar grid, write the shot list on the bars or on the sung lines, render each shot long enough, cut every shot frame-exact onto the song, and time the lyric sheet word by word for captions. Routes: a cut on the bar grid (the picture changes on the downbeats) or a cut on the sung lines (one shot per line, 3.5–6.5 s); lip-synced lines from the vocal stem's window. Free and local apart from the stems (under $1 a song). Use for music videos, lyric videos, montages cut to music, and anything that must land on the beat.
- [Voice a script](https://hypernatural.ai/skill/hn/guide-voice-a-script): How to make the spoken audio for a script: one voice at a time with ElevenLabs TTS v4, a whole scene in one take with text-to-dialogue, or a narrator for a short auditioned on the whole script so the picked audition is the take and its alignment gives the word times; one whole take versus one file per line; audio tags as direction; lossless output and loudness; a top-end check for voices that sound dull on phones; respelling what TTS misreads; every line timed on one track; and the order to fix notes in. Use for voiceover, VO, dialogue audio, TTS, narration, a narrator for a short or ad, performing a script, a dialogue master track.
- [BiRefNet matte](https://hypernatural.ai/skill/hn/birefnet-matte): Cut the people (and the podium, chair or desk in front of the back wall) out of a span of video as a foreground cutout with a real alpha channel, a VP9 WebM that lines up frame for frame with the source: a frame-exact segment, BiRefNet v2 video on fal for the mask, and an ffmpeg alpha merge. Use it to put titles behind people, or a presenter over graphics; never colour-key instead.
- [Split a song into stems](https://hypernatural.ai/skill/hn/demucs-split-song-stems): Split a song into four stems (vocals, drums, bass, other) with demucs htdemucs_ft on fal: give it the song, get four WAVs on the song's own clock. The vocal stem is what speech-to-text should hear to time the lyrics, and the drum stem sharpens downbeat finding; use it first when cutting a music video or lyric captions to a song.
- [Clone a voice](https://hypernatural.ai/skill/hn/elevenlabs-clone-a-voice): Clone a person's own voice with ElevenLabs (instant voice cloning, elevenlabs/voices/add) from about 30 s of them alone on the best microphone they have, varying pitch and intonation, with their consent, and get a voice id that speaks any later script on eleven_v4 like a library voice. Covers asking for consent and the right recordings, preparing and checking the samples (one speaker, no music, enough speech, no clipping), the run call and its fields, what it costs and holds on Hypernatural's account or the person's own key, and auditioning the clone on the script's copy. Use for voice cloning, clone my voice, a presenter's own voice, the founder's voice, custom voice, IVC; to pick a stock voice instead use guide-cast-a-voice.
- [ElevenLabs music bed](https://hypernatural.ai/skill/hn/elevenlabs-generate-music): Generate a music bed with ElevenLabs Music: from a prompt and a length (3 to 300 s), or from a composition plan of timed sections (3 to 120 s each, 300 s in all, local and global styles) whose lengths are enforced so a lift or a final hit lands on a cue mark. Instrumental by default, MP3 out. Covers laying the plan's sections on the cut (sparse under the pain, a swell at the turn, a hit on the end card), two contrasting beds, picking the bed under the picked voice, and its level under a voice. Use for a background bed under a voiceover, an ad, a presenter video or motion graphics; the person picks the music.
- [ElevenLabs sound effects](https://hypernatural.ai/skill/hn/elevenlabs-sound-effects): Generate realistic sound effects and loopable ambience from a text description with ElevenLabs' sound effects model through hn4a (door creaks, birdsong, rain, crowds, impacts, whooshes), then place them on picture. Falls back to the sound-effects guide's other routes when hn4a doesn't offer the model.
- [ElevenLabs speech-to-text](https://hypernatural.ai/skill/hn/elevenlabs-speech-to-text): Transcribe a voice track or a video's soundtrack with ElevenLabs speech-to-text (on fal): every word with its start and end time, a diarized speaker id per word, and optional audio events like (laughs). The basis for speaker turns, pauses to cut at, cue times for titles, and checking a voiced take's words and pace. Use whenever you need to know when something is said.
- [ElevenLabs text-to-dialogue](https://hypernatural.ai/skill/hn/elevenlabs-text-to-dialogue): Perform a multi-voice scene in one ElevenLabs text-to-dialogue generation on eleven_v4, so the turns play off each other instead of being stitched from separate reads, and get each line's start and end time. Lines are directed with audio tags written from the script's parentheticals, emotions, beats and each character's performance guide. Use to voice a script with two or more characters; for one voice alone use elevenlabs-tts-v4.
- [ElevenLabs TTS v4](https://hypernatural.ai/skill/hn/elevenlabs-tts-v4): Speak a line or a script in one ElevenLabs voice with text-to-speech on eleven_v4, directed by bracketed audio tags like [warmly] or free-form ones like [lower, knowing]. Lossless pcm_48000 output, then -16 LUFS on the finished track. Use for voice auditions, narration, a presenter's track or re-voicing one line; for a multi-voice scene in one take use elevenlabs-text-to-dialogue.
- [ElevenLabs voice isolation](https://hypernatural.ai/skill/hn/elevenlabs-voice-isolation): Pull a clean voice out of a mix with ElevenLabs audio isolation (on fal): music, sound effects and room noise dropped, speech kept. Use it to drive a lip-sync or reference-audio model with an original actor's line, then cut the line from the isolated voice by ear from a ladder of trims.
- [ElevenLabs voice search](https://hypernatural.ai/skill/hn/elevenlabs-voice-search): Find candidate voices for a character in the ElevenLabs shared voice library from a casting brief (gender, age buckets, accent, search queries), keeping only voices verified for eleven_v4, then pick the 3 to audition on their metadata (12 when a listening model will rank them first). Use when casting a library voice, from a plain request ("a British woman") or from images and guidance.
- [Crop to vertical](https://hypernatural.ai/skill/hn/ffmpeg-crop-to-vertical): Reframe a finished 16:9 cut to 9:16 for TikTok, Reels or Shorts with ffmpeg: a vertical window per shot that holds still, pans across the shot to follow the subject and action, or fits the whole frame with padding for end cards; then the full vertical cut over the source audio, per-shot files and a contact sheet to check framing. Restaged renders splice in at exact shot length.
- [Cut an audio window for a lip-sync model](https://hypernatural.ai/skill/hn/ffmpeg-cut-audio-window): Cut a window of an audio or video track into a mono 48 kHz WAV for a lip-sync or reference-to-video model: one speaker's line, a talking-head chunk, or a shot's dialogue window silenced from where the next speaker starts. Pads short windows with silence (H3 Max lip sync wants 5 to 15 s; fal rejects reference audio under about 2 s). Use before h3-max-lipsync-a-still, a mouth-only re-sync, or a dialogue render with reference audio.
- [Cut clips to a voiceover](https://hypernatural.ai/skill/hn/ffmpeg-cut-clips-to-a-voiceover): Cut a narrated piece's generated clips to its voiceover in one ffmpeg pass: each shot's window from the voice's word times (it starts at the midpoint of the gap before its line, the last ends 0.6 s after the final word), every boundary on the frame grid, each clip cut from frame 0 for exactly its frames and held on its last frame when short, scaled and cropped to the aspect with lanczos, the clips' own sound under the voice with short fades, one whole take or one file per line, and an optional music bed ducked under the voice. Gives the untitled base cut that titles and captions go over, per aspect from the same shot list. Use for the rough cut, base cut, picture cut or assembly of an ad, explainer, news short or story short made of generated shots.
- [Eased punch-in on a screen recording](https://hypernatural.ai/skill/hn/ffmpeg-eased-punch-in): Push or pan smoothly onto one element of a screen recording (a header, a button, a stat) with ffmpeg: crop a fixed region, upscale it 2x with lanczos, then animate zoompan one output frame per input frame, eased between keyframed crop rectangles with a smoothstep. Smooth, legible punch-ins for split-screen shorts, walkthroughs and tutorials, rendered into the base video before any graphics go on top. For a camera over layers you built yourself, use the compositor's camera instead.
- [Frame-exact trim](https://hypernatural.ai/skill/hn/ffmpeg-frame-exact-trim): Cut a render or clip to an exact number of frames at its native frame rate with ffmpeg, re-encoded, so it drops into a cut frame for frame: trim to shot length, skip an idle opening, drop pulldown duplicates. Reach for it after every render, before review and delivery, instead of speeding a clip up to fit.
- [Grain for plastic skin](https://hypernatural.ai/skill/hn/ffmpeg-grain-for-plastic-skin): Add moving film grain to a finished take whose skin reads plastic, waxy or over-smooth, with ffmpeg, instead of re-rendering it. Builds a ladder of grain strengths at size 2 and a 100% crop sheet against the source so the strength is picked by eye. Use for a client note like 'skin looks plastic' on an otherwise good take.
- [Normalize loudness](https://hypernatural.ai/skill/hn/ffmpeg-normalize-loudness): Bring a voice or dialogue track to a target loudness (EBU R128, -16 LUFS integrated, -1.5 dBTP true peak) with ffmpeg loudnorm in two passes and linear gain, so dynamics and timing are untouched and line timings stay valid. Use on a finished, assembled track before delivery or lip-sync, never per line.
- [Find scene cuts](https://hypernatural.ai/skill/hn/ffmpeg-scene-cuts): Find a video's shot boundaries with ffmpeg's scene score: hard cuts, and soft cuts or dissolves that a normal threshold misses, then a shot table with each shot's first and last frame and a contact sheet to confirm them. Use to split a source spot or a finished cut into shots before recreating, reframing, scripting or titling it.
- [Splice into the source cut](https://hypernatural.ai/skill/hn/ffmpeg-splice-into-source-cut): Rebuild a source video frame for frame on its own timeline and soundtrack with picked clips replacing chosen frame ranges, using one ffmpeg filter graph; every frame without a replacement stays the original footage. Use to assemble a shot-for-shot remake, a re-cut with a few shots redone, or any delivery cut over the original audio, plus its small and no-audio copies.
- [Inspect media](https://hypernatural.ai/skill/hn/ffprobe-inspect-media): Read a media file's facts with ffprobe: its streams, codecs, duration, frame rate, exact frame count, size, and whether it has an audio stream. Use before any trim, cut, mux, measurement or model input, and on every render a provider returns, because later steps rely on these numbers and providers change frame rates.
- [Flux 3: remove a detail from a clip](https://hypernatural.ai/skill/hn/flux-3-edit-video-remove-a-detail): Remove one small detail from a person across a whole clip with Flux 3 edit-video (video to video): a mustache, a mark, an accessory, with motion, timing and framing kept. Cheap ($0.033 a second) but returns about 1248x704, so the result is upscaled back; never use it on a clip with on-screen text, which it rewrites. To change who the person is, use h3-max-recast instead.
- [Gemini ranks voices](https://hypernatural.ai/skill/hn/gemini-rank-voices): Have Gemini listen to candidate voices' ElevenLabs library previews next to the character's images and rank them by fit, 1 to 10, with what it actually heard. A noisy listener: use it to narrow a shortlist to the 3 worth auditioning and to annotate the review, then let the person pick. Use only when the person wants a voice matched to a face or a detailed character; a plain request ("a British woman") skips it and auditions 3 picked on metadata.
- [Script from video](https://hypernatural.ai/skill/hn/gemini-script-from-video): Write a dialogue script from any video by watching and listening to it: a cast with gendered labels, every line verbatim and attributed by mouth and voice, a delivery note and emotion per line, stage directions, beats, sound and music cues, and scenes and shots taken from the picture's cuts. Writes script.json, script.md and script.fountain. Use for transcribing a finished ad, drama, interview or vlog into a screenplay to re-voice or remake.
- [Shot recreation brief](https://hypernatural.ai/skill/hn/gemini-shot-recreation-brief): Have Gemini watch one reference shot with its sound at 24 fps and break it down: camera, timed motion beats for every person, other motion, sound effects, dialogue, music and what a recreation must get right. Then turn that into a recreation brief for a video model, with your direction overriding it where it reads the shot wrong. Use when recreating a shot beat for beat, shot-for-shot, or with a new actor.
- [H3 Max: dialogue shot from a still and a voice track](https://hypernatural.ai/skill/hn/h3-max-dialogue): Render a dialogue shot with MiniMax H3 Max reference-to-video plus a voice track: the shot's still as first frame, the cast's reference pictures, the shot's window of the dialogue track as Audio 1, and optionally a pinned end frame. Better lip sync on large faces, stronger blocking and excellent identity, but it pushes in, may cut to new angles on long lines, and can duplicate a person. The only model that takes an end frame together with audio, for exits and transitions. Up to 15 s of audio and 9 references; for locked framing and voiceover, use seedance-2-5-dialogue.
- [H3 Max: image to video](https://hypernatural.ai/skill/hn/h3-max-image-to-video): Animate a still with MiniMax H3 Max image-to-video (or its reference-to-video mode when characters or products need reference pictures): the fastest and cheapest video model, with good motion and performance but smoothed, painted-looking hair and skin up close. Use to draft every shot quickly, and for wide shots or subjects without fine texture; upgrade to kling-v3-image-to-video or seedance-2-5-image-to-video where close-up realism, 1080p grain or complex physics matter.
- [H3 Max: lip-sync a still](https://hypernatural.ai/skill/hn/h3-max-lipsync-a-still): Lip-sync a still to an audio clip with H3 Max image-to-video and target audio: one locked presenter or speaker still plus 5 to 15 s of speech becomes a talking head that opens and closes on the still. Use for avatar and presenter videos, a long voice track cut into chunks, or a line on a shot with no usable take; to fix only the mouth of an existing take, use heygen-v3-resync-existing-take or veed-v2-resync-existing-take instead.
- [H3 Max: recast a person in a take](https://hypernatural.ai/skill/hn/h3-max-recast): Swap a person in an existing video take for someone else with MiniMax H3 Max recast (video to video): the original performance, timing, camera and sound kept exactly, the face and body taken from reference pictures. Use when the whole original take is usable and only who is in it changes; to change the performance too, swap the still and re-animate instead (guide-recast-a-person).
- [H3 prompt format](https://hypernatural.ai/skill/hn/h3-prompt-format): The prompt grammar for MiniMax H3 Max video: <Subject N> and <Picture N> labels, (Sn) speaker IDs, <d> dialogue blocks, the Audio 1 binding, the soundscape rules and the fixed header lines, for image-to-video, reference-to-video with dialogue, lip sync and recast. Load it before writing any H3 Max prompt; the h3-max- leaves cover each mode's request.
- [HeyGen v3 Precision: re-sync the mouth of an existing take](https://hypernatural.ai/skill/hn/heygen-v3-resync-existing-take): Re-sync only the mouth of an existing video take to new or corrected audio with HeyGen v3 Lipsync Precision (video-to-video lip sync), keeping the take's skin, light and performance. The best-looking mouth-only fix: picked over VEED v2 and H3 Max lip sync on a 3/4 close-up. Use when a well-acted take speaks the wrong line, drifts out of sync, or was re-voiced; for the cheapest fix use veed-v2-resync-existing-take.
- [hn4a-compositor](https://hypernatural.ai/skill/hn/hn4a-compositor): The base for everything drawn with hn4a-compositor, hn4a's own compositor: a JSON layer spec (fills, images, video, text, SVG paths, groups that work as cameras or lie on a surface) played live in a browser page and rendered frame-exact to an MP4 with the audio mixed in. Holds the spec format, the keyframes, the commands, the first look in the chosen direction or a few relevant candidate treatments, the one gate every compositor job keeps (the person approves the live spec, then you render), re-rendering only a changed span, and the checks. The base every hn4a-compositor-* leaf loads; which of them fits the job (overlay motion graphics, ad titles, in-scene titles, hand-reactive titles, a product screen cut up or rebuilt, ad and explainer cameras, a news chyron, a storyboard, kinetic lyrics, an audio waveform, graphics on a surface in perspective) is picked in guide-add-titles-and-motion-graphics. Not Remotion, no build step.
- [Audio waveform](https://hypernatural.ai/skill/hn/hn4a-compositor-audio-waveform): Draw a sound's real waveform with hn4a-compositor, from the audio file itself: bars from its peak envelope as a path layer, drawn on, revealed left to right, or scrolling past a fixed playhead as an audiogram for an audio-only podcast clip. Use for audiograms, waveform graphics, a sound effect or ambience shown on screen, and a running waveform band under a sound-themed piece.
- [Brand ad titles](https://hypernatural.ai/skill/hn/hn4a-compositor-brand-ad-titles): The title kit for a short ad in the brand's own font, colours and logo, built with hn4a-compositor and cued on the voice's words: the hook as a big stack whose number slams in the accent, lines that build word by word with one accent word and hand-placed breaks, the logo lockup on the brand name, glass benefit chips that pop on their words, the proof stat with a kicker, key-phrase captions, and the end card with the site's CTA button recreated and the URL, in 16:9, 1:1 and 9:16 from one set of helpers. Covers what to show on each line, what to leave to the narration, and the end card's timing. Use for ad titles, kinetic type, logo lockups, end cards, CTA cards and benefit chips over a generated or real ad cut.
- [Calm explainer camera](https://hypernatural.ai/skill/hn/hn4a-compositor-calm-explainer-camera): A calm camera treatment for hn4a-compositor when the brief needs inspection or measured reveals: one slow drift or push per screen, short eased moves that start just before their cue, and a slow creep in on every hold when subtle movement helps the hold, kept inside the screen's edges. Numbers that worked: 0.65 s moves starting 0.3 s before the cue with a 3% creep (a narrated summary); framings at 1.15-1.6x, 0.7 s eases landing 0.15 s before the phrase, a 4% creep (a screen-recording walkthrough). Writes the moves for a camera group. For a brief that needs faster emphasis, use hn4a-compositor-fast-ad-camera.
- [News chyron](https://hypernatural.ai/skill/hn/hn4a-compositor-chyron): News graphics composited with hn4a-compositor: a lower third with a kicker tab (SPECIAL REPORT, BREAKING), the station bug, a ribbon whose headline and name straps swap on cue, a ticker crawl that is already full and moving on its first frame, a LIVE bug with the place, and a vertical short's pill title bar; laid out for 16:9 and 9:16 from the same calls, clear of captions and platform buttons. Use for a news spot, a mock broadcast, a presenter's name strap, a guest's name in a podcast clip, or any story title bar.
- [Fast ad camera](https://hypernatural.ai/skill/hn/hn4a-compositor-fast-ad-camera): The default camera for the presenter-over-UI ad recipe in hn4a-compositor: a push on every spoken beat that lands just after the word and holds, hard cuts and whip pans to the next screen instead of drifting there, kept inside the screen's edges, with a pace check against what read as frantic and what read as sleepy (9 moves in 15 s at 1.45-1.85x was too busy; 2 at 1.12-1.42x too dull; about 3 pushes per screen at 1.35-1.6x was right). Writes the moves for the camera group over a rebuilt or cut-up product screen. For an explainer, a summary or slides, use hn4a-compositor-calm-explainer-camera.
- [Hand-interactive titles](https://hypernatural.ai/skill/hn/hn4a-compositor-hand-interactive-titles): Titles that react to a person's face and hands in real footage: a hero word squashed flat by both hands pushing down, one shoved off screen by a sideways push, a title glued to the side of the head, a title stretched between the palms, words dropped onto the head, with the hands kept in front of the type. Tracks the face and hands locally (free), uses the hands-only matte, and writes per-frame keys into an hn4a-compositor spec.
- [Compose in-scene titles](https://hypernatural.ai/skill/hn/hn4a-compositor-in-scene-titles): Titles that live in the footage, rendered with hn4a-compositor: type behind people (a foreground cutout over the words), words that pop or pulse on spoken cues, light spilling on the set, glitches timed to the picture, typing with a caret, a phone ring, a drop, a focus pull, a screen powering on, a floor reflection, karaoke captions on the spoken word, a camera pushing into a product screen rebuilt as layers. Uses a small Python helper that writes the JSON layer spec. Use for kinetic and in-scene titles over a finished cut; type that lies on a floor, sits on an angled wall or recedes into the set goes in a plane group (hn4a-compositor-perspective-planes).
- [Kinetic lyrics](https://hypernatural.ai/skill/hn/hn4a-compositor-kinetic-lyrics): Lyric captions for a music video, composited with hn4a-compositor over the cut: each sung word slams in on its syllable and pulses on the beat, one layer per word on a line that never reflows, type that follows the look on screen, punch words slammed in big, white flashes and camera shake on the hits, lines that act out their words, lines skipped where the shot already has them drawn in, and lines lying on a floor or a wall in perspective. Widths measured with the same font files the spec embeds. Reads the timed lyrics, the song's beats and the shot list. Use for kinetic lyrics, lyric videos and music-video captions.
- [Make product UI exciting on screen](https://hypernatural.ai/skill/hn/hn4a-compositor-make-product-ui-exciting): Animate selected parts of a real product screenshot when a staged reveal helps explain the screen or direct attention: cut cards, rows, charts and buttons into exact image layers and cue their entrances. Use a held screenshot for inspection, existing footage for an actual operation, or hn4a-compositor-rebuild-product-screen when the story needs editable data or behaviour.
- [Compose motion graphics](https://hypernatural.ai/skill/hn/hn4a-compositor-motion-graphics): Build overlay motion graphics and title slides with hn4a-compositor: headlines, kickers, lists that build on the words, logos, hand-drawn arrows and circles, lower thirds, a presenter as a full frame, a window, a cutout or a round bubble, all cued to word timings. Covers laying a slide out beside a talking head, cueing each entrance a touch before its word, moving a presenter between layouts (full frame, window, bubble), where the camera's pace lives, and a complete worked slide. Use for slides, explainers and kinetic type over or beside footage; for type that lives inside the footage, use hn4a-compositor-in-scene-titles.
- [Perspective planes](https://hypernatural.ai/skill/hn/hn4a-compositor-perspective-planes): Lay graphics flat on a surface in the picture with hn4a-compositor plane groups, so they read as receding into the scene: a route drawn off down a street, type lying on a floor or painted on a wall, an inset on a screen or a card, captions and titles that tilt into the set. Covers the 3x3 matrix from four corner points, a street from a level camera, keying the plane per frame from corners tracked through the footage, how the group's px relate to sharpness, the limits, and putting a person in front with a matte. Fakes depth with flat planes; it is not 3D models or a 3D-animation look.
- [Rebuild a product screen as compositor layers](https://hypernatural.ai/skill/hn/hn4a-compositor-rebuild-product-screen): Rebuild one product screen (a SaaS table, a dashboard, an app view) from a marketing screenshot as hn4a-compositor layers, so a camera can push into it at 2x and stay crisp and every part that moves is its own layer: rows that cascade and insert, cells that type, pills that flip, counts that climb, donuts and bars that draw. Covers reading the screenshot (measuring in source pixels, sampling colours per pixel, the app's own font), the builder contract (POI plus build(group_id, cues) returning layers, and a test spec), believable data, the checks, and a brief for one subagent per screen. For a product ad, app demo, UI animation or explainer; titles and cameras over footage are hn4a-compositor-in-scene-titles.
- [Storyboard in the compositor](https://hypernatural.ai/skill/hn/hn4a-compositor-storyboard): A storyboard of the whole piece, built in hn4a-compositor from cheap parts before anything is paid for: the real screenshots under the planned camera, a grey presenter silhouette at its planned size that shrinks into the bubble where the cut will, the headline in the brand font, props as labelled boxes at their real size, and every camera target as a dashed box with its push written on it, cued from estimated word times. One still per beat goes to the person with the board itself playing live, then optionally an animatic on the approved voice. Catches slides-not-an-ad layouts, a bubble too small, too many camera moves and a set that doesn't hold up while they cost nothing to change. Use for presenter ads, product walkthroughs, launch videos, pitch videos and any piece with paid renders.
- [hn4a-lipsync-check: is the mouth in sync, over time?](https://hypernatural.ai/skill/hn/hn4a-lipsync-check): Check whether a talking head's mouth is in sync with its audio over time, and where it isn't: hn4a-lipsync-check measures the mouth's lag every 0.2 s (SyncNet with MediaPipe face tracking, run locally), flags sustained spans out of tolerance, and names the words spoken there. Catches lip-sync drift that a duration check or a quick watch misses, such as H3 Max lips running 3 to 5% slow. Use on every lip-synced take before picking it; internal QA only.
- [Kling v3: image to video](https://hypernatural.ai/skill/hn/kling-v3-image-to-video): Animate a still into a shot with Kling v3 Pro image-to-video: the first frame pinned (frame 0 holds at about 0.996 against the still), an optional pinned last frame, and characters or products held from @Element references. Smooth, photographic 1080p for close-ups where texture must hold; use for frame-accurate shot recreation. For physics-heavy motion use seedance-2-5-image-to-video, for fast cheap drafts h3-max-image-to-video.
- [Kling v3 prompt format](https://hypernatural.ai/skill/hn/kling-v3-prompt-format): The prompt grammar for Kling v3 (Kling 3.0) video: short plain-prose prompts anchored on the first frame, @ElementN labels for the characters and products sent as elements, Kling's [speaker, voice]: "line" dialogue tags, and the Background audio line. Load it before writing any Kling v3 prompt, so characters and products hold to their references; kling-v3-image-to-video covers the request.
- [Convert a written script](https://hypernatural.ai/skill/hn/llm-convert-written-script): Turn a written script (a screenplay, a shot list, a draft with quotes and stage directions) into script.json, the same format gemini-script-from-video writes, so it can be voiced and shot: scenes, the script's own shots, and ordered elements with every dialogue line copied word for word and checked, a delivery note and emotion per line, and estimated timings. Use when the person brings a script as text, .md, .docx or a doc rather than a video.
- [Draft a gift video note](https://hypernatural.ai/skill/hn/llm-draft-a-gift-video-note): Draft the short message that goes with an outreach or gift video made for someone who didn't ask for it (a founder, an open-source maintainer, a creator, a company in the news): why this person, what it was made from, honest time and cost, and an honest ask, in the person's own voice, leading with the video. The person sends it themselves, privately; the video is posted only once the recipient says yes, and nothing in the note claims what the source contradicts. Use for gift videos, outreach videos, cold outreach, spec ads, a note to a founder or maintainer, sharing a side-by-side.
- [Pick a visual metaphor](https://hypernatural.ai/skill/hn/llm-pick-a-visual-metaphor): Develop a visual concept for a product or brand promise, a codebase or a news story: a transformation, contrast, pattern, sense of scale, material behaviour, system or character story that makes the idea legible. Returns a concept and shot-by-shot table grounded in the facts and brand direction; character worlds and name-based puns are optional approaches, with delivered examples.
- [Source a news story](https://hypernatural.ai/skill/hn/llm-source-a-news-story): Pick a news story with the person and source it before any script: two or three of the day's stories with a line each on fit and risk (deaths, partisan fights, a stale peg), then source/article.md with the link, date and author and only the facts the script will use, each traceable to the source and marked 'reportedly' where the source is, with fallbacks when the article won't load. Text work the agent does itself. Use before scripting a news short, a news spot or any explainer about a story in the news.
- [Write a hook](https://hypernatural.ai/skill/hn/llm-write-a-hook): Write the first 3–6 seconds of a short video or TikTok ad so a cold viewer stays: the hook patterns that worked (conflict then counter, orientation then the clever line, the customer's own mess with one turn word, the pain with a number, the result shown before it is explained, the product on screen in seconds), the picture that goes with each, three candidates tested against a short list, and the dead ends (slow scene-setters, a long webcam opening). Use for hooks, openers, cold opens, scroll-stoppers, the first line of a short, reel, TikTok, social ad or explainer.
- [Write an ad script](https://hypernatural.ai/skill/hn/llm-write-an-ad-script): Write the script for a short product ad or marketing explainer (20–45 s) from the product's facts: a beat order that worked for its shape (a 30 s presenter ad over the product's UI, a 20 s five-beat generated-world ad, a 35 s faceless explainer, a 28 s mascot story, a 45 s repo explainer), one shot per point, narration that names the specifics, on-screen type that says the gist and never prints what is spoken, and no invented reviews, numbers or claims. Returns a beat table (time, picture, narration, on screen) and the narration word for word with audio tags. Use for ad scripts, ad copy, promo scripts, explainer scripts, beat sheets, product video scripts.
- [Character sheet](https://hypernatural.ai/skill/hn/nano-banana-character-sheet): Make a 2x2 character sheet (turnaround) with nano banana Pro edit so a character stays the same from shot to shot: front close-up, profile close-up, full body front, full body back, exactly one figure per panel, a fixed outfit spelled out, no labels; drawn in the job's look from the picked style frame, or from a real person's photo for a cast member, or around a product that becomes a character. Covers the prompt that worked, two takes and rejecting sheets whose views disagree, 4x2 grids, skin tone under strong colour light, stray lettering, one sheet per person per look, then cutting it into single views. Use for a recurring hero, a mascot in a new look, cast consistency, model sheets, turnarounds and character references before any shot still.
- [Derive the last frame from the locked first](https://hypernatural.ai/skill/hn/nano-banana-derive-last-frame): Make the pinned last frame of a shot by editing the locked first frame with nano banana Pro, as text edits only, so face, hair, props and grain stay continuous by construction; then crop it to the source's push-in with python-match-push-in. Use before a first-and-last-frame render (Seedance, Kling, H3 Max) whenever the shot must end on an exact state.
- [Design the main image of an empty set](https://hypernatural.ai/skill/hn/nano-banana-design-empty-set): Design the main image of an empty set (a location: a TV studio, a quiz-show stage, an office) with nano banana Pro from a written geography, a palette and style references, brand colour included. Use first when a job needs a location built before any shot stills; the angles and the staged shots are made from this image.
- [Nano banana prompt format](https://hypernatural.ai/skill/hn/nano-banana-prompt-format): How to write a nano banana image prompt (2.1 and Pro, generate and edit) that draws what you meant: every image labelled with what it is for, real people and places named exactly, the camera and which way every body, face and hand points, every word on a sign or jersey spelled out, and what an edit keeps. Anything a prompt asks for but doesn't describe gets invented. Load it before writing any nano banana prompt, for stills, illustrations, shot stills, set plates and edits; the nano-banana- leaves cover each task's request.
- [Recast a person in their own outfit](https://hypernatural.ai/skill/hn/nano-banana-recast-in-own-outfit): Replace one person in a still with a new character of any gender or build, dressed in their own outfit from their character sheet (badge included), keeping the pose, framing, set, light and everyone else, on nano banana 2.1 edit by default (Pro for a precise detail edit). Single views cut from the sheet as references; a back view for shots from behind. Use to recast a role in a finished dialogue video or any shot where the costume changes with the actor.
- [Remove a small object](https://hypernatural.ai/skill/hn/nano-banana-remove-small-object): Remove a small object (a badge, a pin, a stain, a stray logo) from a still without touching anything else: crop about 256 px around it at 2K, enlarge 4x, have nano banana edit it out, and paste back only a feathered circle, so the face and grain stay bit-identical. Also edit the character's reference sheets, and never name the object in the video prompt. Use before animating a still that must lose a detail.
- [Re-render small print in an enlarged crop](https://hypernatural.ai/skill/hn/nano-banana-rerender-small-print): Get small print or a logo on a prop exactly right in a 1080p still: crop the prop, enlarge it about 6x so it fills the image, re-render only its print with nano banana Pro edit against the product photo, and paste back only the pixels that changed, so grain and faces stay. With a match reference cut at the same scale from the shot's other pinned frame, the print keeps its size and place between first and last frame. Use for garbled lettering, wrong logos, labels and packaging text.
- [Restage a frame for 9:16 or 3:4](https://hypernatural.ai/skill/hn/nano-banana-restage-for-new-aspect): Restage a 16:9 frame for a vertical deliverable when a plain crop would lose a story beat: either a 16:9 edit that moves people into the middle third, judged by its centred 9:16 crop, or a native 9:16 (or 3:4) redraw as if shot in portrait, on nano banana 2.1 and Pro edit. Use for TikTok, Reels and Shorts versions, and for portrait talking-head stills.
- [Restyle a photo as a location plate](https://hypernatural.ai/skill/hn/nano-banana-restyle-photo-as-plate): Turn a photo of a real place into a location plate in the job's art style with nano banana 2.1 edit, keeping the photo's camera, perspective and architecture exactly, its ads and logos cleared and its real signs spelled out. Use when stills must show a real stadium, building, street or landmark from several angles: one plate per camera direction, then stage people into the plates instead of generating the place.
- [Render the set from a new angle](https://hypernatural.ai/skill/hn/nano-banana-set-from-new-angle): Render the same set from another camera position (a reverse angle, a podium single, an over-the-shoulder, a wide of the audience) with nano banana Pro edit, from the picked main image of the set. Use after the main image is picked, once per camera the script needs, to get empty plates that keep the set's design, colour and signage.
- [Stage the cast in a set](https://hypernatural.ai/skill/hn/nano-banana-stage-cast-in-set): Stage the cast into an empty set plate (the set's main image or one of its angles) with nano banana Pro edit to make a shot still: who stands where, doing what, lit by the set, with a picture on any screen. Use for each shot's first still once the set's angles are picked; crop the plate first when the shot is tighter than the plate.
- [Swap the actor, keep the wardrobe](https://hypernatural.ai/skill/hn/nano-banana-swap-actor-keep-wardrobe): Swap a new actor into an existing frame from a character sheet with nano banana edit, keeping the original's wardrobe, pose, eyeline, framing, light and grain: three seeds per frame at 2K, then Lanczos to 1080p. Use to recast the stills of a spot shot for shot; on Pro it can restyle the held product in the same pass.
- [Swap the product on a held prop](https://hypernatural.ai/skill/hn/nano-banana-swap-held-product): Put a client's product onto a prop someone is holding in a still, with nano banana Pro edit: a pre-hazed product photo for the design, a crop of the source frame's prop for angle, tilt and grip, so the product sits in the shot's light instead of looking pasted on. Use for product swaps and for restaged or redrawn stills that lost the product's look; fix the logo afterwards with nano-banana-rerender-small-print.
- [One targeted fix on a picked still](https://hypernatural.ai/skill/hn/nano-banana-targeted-fix): Make one small correction to an already-picked still with nano banana edit (a hair strand, a prop's position, a pose detail, a letter), using a crop of the original or matching frame as the reference so the fix sits in the shot's own light, focus and grain. Use after a swap or recast when one thing is wrong; for small print use nano-banana-rerender-small-print, for removals nano-banana-remove-small-object.
- [Assemble a dialogue track](https://hypernatural.ai/skill/hn/python-assemble-dialogue-track): Join performed dialogue passages (ElevenLabs text-to-dialogue takes) into one voice track, WAV and MP3, with every line's start and end known in track.json. Also speeds one character's lines up with pitch kept, lengthens or shortens a pause for a director's note or the picture's action time, and gives each line's exact speech times from a transcript. Use after voicing a script, before loudness normalization and lip-sync.
- [Assemble lip-sync chunks](https://hypernatural.ai/skill/hn/python-assemble-lipsync-chunks): Turn a presenter's lip-sync chunk renders into one talking head: check every chunk (opens on the pinned still, back in the pose at the cut, the mouth moving, in sync over time with the drift measured as a slope), retime a chunk whose mouth drifts steadily behind (the free fix for H3 Max's 3 to 5% slow lips, no re-render and no paid re-sync), then join the picks on one frame grid with a 5-frame dissolve at each seam and the track under it, and check the whole against the track. Free and local. Use after python-plan-lipsync-chunks and h3-max-lipsync-a-still, for a slide-deck presenter, a sales or pitch video, or any long talking head.
- [Check pinned frames](https://hypernatural.ai/skill/hn/python-check-pinned-frames): Check whether video renders landed on their pinned first and last frames and how far each runs past the shot's length: per-frame SSIM against both stills, the frame where it lands on the last still, and the overage in frames and seconds. Use after image-to-video renders with a first frame, or first and last frames, pinned, to decide whether to trim the tail or re-brief a take that lands late.
- [Check title occlusion](https://hypernatural.ai/skill/hn/python-check-title-occlusion): Measure how much of each title behind a person the person hides: for every text, image or path layer between the footage and the person cutout in an hn4a-compositor spec, every 2nd frame it is on, the share of its ink the cutout covers, through every camera, shake and group that moves either. Reports the worst share and when, and fails any layer over 0.2. Free and local. Run after every build of type-behind-people, before the live spec goes to the person.
- [Conform a still to 1080p](https://hypernatural.ai/skill/hn/python-conform-to-1080p): Conform a 2K nano banana render (or any still) to 1920x1080, or another delivery size, by centre-cropping to the aspect and resizing with Lanczos; optionally crop a box first to reframe a plate tighter or cut an element out of a set image. Use after every image edit or generation before the still goes to a video model, a review or a paste-back.
- [Contact sheet](https://hypernatural.ai/skill/hn/python-contact-sheet): Make labelled contact sheets to look at many pictures at once: frames of a video at even steps, every N seconds, at given times or at fractions of its length (each labelled with its time and frame number), one row per take or shot comparing first, middle and last frames, or a folder of stills, renders or a site's scraped images labelled with name and size, plus a px grid for reading boxes and zoom targets off a frame. Use to watch a long recording, find where the action is in a render, compare takes side by side, check stills for baked words, or measure a region.
- [Crop a region, edit it big, paste it back](https://hypernatural.ai/skill/hn/python-crop-and-paste-back): Cut a small region (a logo, a label, a badge, a pin) out of a frame, enlarge it for an image-model edit, then paste only that region back so the rest of the frame stays bit-identical: grain, faces and background untouched. Also keeps a full-frame edit only inside a box after aligning it to the original. Use for any small fix on a still that a whole-frame pass would smooth.
- [Cut best takes by speech islands](https://hypernatural.ai/skill/hn/python-cut-best-takes-by-speech-islands): Cut one recording full of retakes (a webcam or talking-head file) into a tight rough cut of its best takes: split it into utterances from the sound, transcribe each on its own, pick the last complete take of every line, then build the cut in one ffmpeg graph with each piece trimmed to its speech, pauses over 0.14 s squeezed to about 0.11 s, 6 and 10 ms fades and every edge on the frame grid, plus a timeline of what was kept. Free and local. For takes split across files or paired screen and webcam recordings, cut from the transcript with python-cut-phrases-from-takes instead.
- [Cut phrases from takes](https://hypernatural.ai/skill/hn/python-cut-phrases-from-takes): Cut a paper edit into frame-exact tracks: give it each take's word timings (whisper-cli or ElevenLabs JSON) and a text cut list of (take, approximate second, the exact words kept) plus markers for time jumps, sped-up stretches, clips played inside the screen and full-screen inserts; get one or two cropped video tracks (a screen recording and a webcam, say) and the mic, cut on one frame grid with each phrase padded clear of its neighbours and long pauses squeezed, plus timeline.json with every kept word in comp time for captions and cues. Also finds where a file played back during recording sits in the mic by envelope cross-correlation. Use for transcript edits of screen recordings, product demos, interviews and multi-camera talks.
- [Cut clips to the bar grid](https://hypernatural.ai/skill/hn/python-cut-to-bar-grid): Cut a music video's clips onto the song's bar grid: each shot gets exactly the frames between its start and end bars so cuts never drift, optional stepping on twos (12 fps look) per shot, a start offset to skip an idle opening, a short clip holding its last frame, shots given in seconds instead of bars for a cut on the sung lines, all joined with the song underneath, and a --check that finds every cut with ffmpeg's scene score and reports any more than a frame off. Use for every rough and final cut on a song.
- [Find a recorded line in a render](https://hypernatural.ai/skill/hn/python-find-line-in-render): Find where in a video render a recorded dialogue line is spoken, by matching loudness envelopes in 10 ms steps, and get the offset and start frame to cut the render on so its line lands where the original's does. Use on dialogue takes rendered with an untimed audio reference (they speak the line late or early), or to align any clip to a reference line; check the result by ear.
- [Find speech pauses](https://hypernatural.ai/skill/hn/python-find-speech-pauses): Find where the speech and the pauses are in a recording from the sound itself (a 10 ms RMS envelope), not from word times: every pause with its length and depth, the energy dips inside a run of speech, the utterances split out to their own WAVs to transcribe one by one, and a picked piece's edges moved into the pauses around it. Free and local. Use to cut takes, soundbites and moments cleanly, to transcribe a long recording utterance by utterance, and to split a track at pauses; for pauses read from a transcript, python-read-word-timings.
- [Map a song's bars](https://hypernatural.ai/skill/hn/python-map-song-bars): Map a song's bar grid with librosa: tempo, beats, downbeats (4/4, phase picked by the strongest drum onsets), drum onsets, an energy curve, and a printout of every sung phrase with its bar number, the grid a music video's shot list is written on. Give it the song WAV, optionally the drum stem and the vocal's word timings (ElevenLabs STT or whisper-cli JSON); also turns a bar number into a time and a frame for the cut.
- [Match the source's push-in on a pinned last frame](https://hypernatural.ai/skill/hn/python-match-push-in): Crop a derived last frame so it is pushed in from your locked first frame exactly as far as the source shot's last frame is from its first, measured, so a first-and-last-frame render keeps the original camera move. Run after nano-banana-derive-last-frame; image models ignore 'push in 10%'.
- [Matte a talking head](https://hypernatural.ai/skill/hn/python-matte-a-talking-head): Cut a presenter or talking head out of their video for free on this computer with Robust Video Matting (RVM): a soft, steady alpha that follows hair and holds from frame to frame, written as a VP9 WebM with alpha that lines up frame for frame with the source, plus the grey mask. Made for one person facing the camera, such as an AI avatar's lip-synced take or a webcam read, to put them over slides, a set or graphics. For people with furniture that must stay in the cutout, or a busy shot, use birefnet-matte; for a matte with the face and hands tracked too, python-track-face-and-hands.
- [Measure fine detail](https://hypernatural.ai/skill/hn/python-measure-detail): Measure how much fine detail (film grain, pores, hair, small print) is left in a box of a 1080p frame: the spread of a high-pass filter, compared across the source and its edits or renders. Use to catch an edit or render that smoothed a face into plastic skin or lost the grain, and to pick a grain strength from a ladder against the source.
- [Measure judder](https://hypernatural.ai/skill/hn/python-measure-judder): Score how smooth a video's motion is: a judder number from its per-frame motion steps (mean change in step over the median step), plus a count of repeated frames that shows a padded, dup-then-skip cadence. Use to compare takes of one shot, or a render against its source footage, when motion looks jerky, stuttery or choppy.
- [Measure layout drift](https://hypernatural.ai/skill/hn/python-measure-layout-drift): Measure whether a render or an edited still kept the layout of its pinned still: the best zoom and shift between them, and how well they match after that alignment. Use after every image-to-video or ref2vid render to check it opened on its still (under 0.95 = it didn't), and to decide whether to accept a slight zoom or counter-crop it. Raw SSIM can't tell; this can.
- [Plan lip-sync chunks](https://hypernatural.ai/skill/hn/python-plan-lipsync-chunks): Plan a long voice track's lip-sync chunks before any render: 5 to 14 s each, cut in a pause read off the sound itself (never inside a word, however a transcript smears), a pause after a sentence first, every cut snapped to the 24 fps frame grid, and each chunk's audio written sample-exact from the track, with the words each chunk says for its prompt. The chunks join back into the track with no gap or overlap. Free and local. Use before lip-syncing a presenter still to a whole voice track with H3 Max; then python-assemble-lipsync-chunks checks the renders and joins them.
- [Read word timings](https://hypernatural.ai/skill/hn/python-read-word-timings): Read word times from any source (a whisper-cli transcript, ElevenLabs speech-to-text with diarized speakers, or the character alignment a TTS take returns) into what an edit needs: the words, when a phrase is spoken, named cue times for a build script, a take checked word for word against its script (tags read aloud included), speaker turns, the pauses safe to cut at, and where a clip's own speech ends. Use to cue titles and motion graphics on the words, check a voice take, or find who speaks when; to cut a voice track into lip-sync chunks, python-plan-lipsync-chunks.
- [Split a character sheet into single views](https://hypernatural.ai/skill/hn/python-split-character-sheet): Cut a grid character sheet (the 2x2 turnaround: front close-up, profile, full body front, full body back) into single-view reference images, keeping only the views a shot needs. Use before giving a character to any image or video model (nano banana staging and recasts, ref2vid, dialogue renders): models get single views, never the whole sheet.
- [Synthesized SFX for motion graphics](https://hypernatural.ai/skill/hn/python-synth-sfx-for-motion-graphics): Add sound effects to a motion-graphics or interactive-titles cut: cue each sound from the hn4a-compositor build, synthesize it from a 24-sound kinetic-type palette (pops, clicks, stamps, impacts, bass drops, thwacks, squashes, splats, boings, panned whooshes, risers, record scratches, glitches, dings, confetti cannons) locally with numpy/scipy (free), panned by where its graphic sits, and mix the track under the voice. Use for free, slightly cartoony effects that land exactly on graphics events, or SFX in place of the original audio; for realistic or described sounds use elevenlabs-sound-effects (guide-add-sound-effects weighs the two).
- [Time a lyric sheet](https://hypernatural.ai/skill/hn/python-time-lyrics): Time a song's lyric sheet word by word against speech-to-text of the vocal stem (ElevenLabs STT or whisper-cli JSON): each sung line matched in order, a misheard word keeping the STT's timing under the lyric's text, a dropped word interpolated and marked ?, a clean section's timing copied onto a repeat STT garbled, and every line printed with its bar. Use to cue kinetic lyric captions and karaoke-style words on the syllable.
- [Track face and hands](https://hypernatural.ai/skill/hn/python-track-face-and-hands): Track the person in a cut locally and free with MediaPipe: a per-frame person matte, both hands (palm centre, lowest point, box) and the face (box, nose, chin, forehead), written as track.json, plus a person cutout and a hands-only cutout as VP9 video with alpha and an end freeze frame with its cutout. Use for titles behind a person, graphics that react to the face and hands, a presenter cut out over a backdrop, and the occlusion check; pay for birefnet-matte only where its edges are needed.
- [Seedance 2.5: dialogue shot from a still and a voice track](https://hypernatural.ai/skill/hn/seedance-2-5-dialogue): Render a dialogue shot with ByteDance Seedance 2.5 reference-to-video plus a voice track: the shot's still as the opening, the cast's reference pictures, and the shot's window of the dialogue track as @Audio1, with each line pinned to its time. Holds a locked framing, keeps mouths closed on voiceover, and takes up to 30 s of audio and 29 references. Use for locked talking heads and restrained two-handers; for bigger blocking, better close-up lip sync or an end frame with audio, use h3-max-dialogue.
- [Seedance 2.5: image to video](https://hypernatural.ai/skill/hn/seedance-2-5-image-to-video): Animate a still with ByteDance Seedance 2.5 image-to-video, first and last frames pinned: the exact opening (about 0.997 against the still), photographic grain at 720p or 1080p, and complex physical motion such as paper, cloth or debris falling through the air. Use for physics-heavy shots and shots where a small detail must hold between two pinned frames; slow (6 to 30 min a render). For fast drafts use h3-max-image-to-video; for close-up character shots, kling-v3-image-to-video.
- [Seedance 2.5 prompt format](https://hypernatural.ai/skill/hn/seedance-2-5-prompt-format): The prompt grammar for ByteDance Seedance 2.5 video: @Image1 as the first frame, @ImageN binding lines for reference pictures, @Audio1 for a dialogue track, timed spoken lines (From 0.09 to 1.48 seconds, @Image2 speaks {…}), integer-second marks, <sound> effects and the fixed closing lines, for image-to-video and reference-to-video. Load it before writing any Seedance prompt; seedance-2-5-image-to-video and seedance-2-5-dialogue cover the requests.
- [3D animation look](https://hypernatural.ai/skill/hn/style-3d-animation): The 3D-animation look for generated stills and clips: polished, stylized 3D feature-film characters with appealing proportions, large expressive eyes, soft subsurface skin and warm cinematic light, in a fully stylized world or dropped into a real-looking one, including a product turned into a character (a can that walks and shops) with its label kept exact. Holds the style line used verbatim in every nano banana prompt, the H3 Max phrasing, the resolution, the caption type that pairs with it, and when 3D beats claymation. Use for 3D animation, CG cartoon, animated product mascots, product-as-character ads and kids' or learning videos in a 3D look. Not perspective tricks on flat graphics (hn4a-compositor-perspective-planes); how to pick among looks is guide-pick-a-stylized-look.
- [Claymation look](https://hypernatural.ai/skill/hn/style-claymation): The claymation look for generated stills and clips: a warm, stylized 3D miniature toy world, as if photographed with a tilt-shift macro lens, soft matte clay and painted wood, rounded shapes, toy-like figures with simple friendly faces. Holds the style line used verbatim in every nano banana prompt, the H3 Max style phrase (3D CG claymation), why 768P is enough for clay (half the price of 1080P), sizes given by body parts, devices that grow real logos, and why clay animates cleanly. Use for claymation, clay, stop-motion, toy-world and miniature-diorama ads, explainers and news shorts; how to pick among looks is guide-pick-a-stylized-look.
- [Topaz upscale](https://hypernatural.ai/skill/hn/topaz-upscale): Upscale a rendered clip faithfully (720p to 1080p, any factor) with Topaz video upscale on fal, enhancing what is there without inventing new detail, so faces stay the faces in the render. Use on a 720p render before trimming, when rendering natively at 1080p is too slow or not offered.
- [VEED v2: re-sync the mouth of an existing take, cheapest](https://hypernatural.ai/skill/hn/veed-v2-resync-existing-take): Re-sync only the mouth of an existing video take to its audio with VEED Lipsync v2 (video-to-video lip sync), the cheapest mouth-only fix at $0.07/s. The fix for an H3 Max lip-sync chunk whose lips are out of sync in a way a free retime (python-assemble-lipsync-chunks) can't remove, flagged by hn4a-lipsync-check. Returns 25 fps, so convert back to the take's rate. For a close-up where the mouth is watched, use heygen-v3-resync-existing-take.
- [Transcribe with word times (whisper-cli)](https://hypernatural.ai/skill/hn/whisper-cli-transcribe-with-word-times): Transcribe speech locally and free with whisper-cli, one word a segment with start and end times: pull the audio to a 16 kHz mono WAV, run several files at once, read the JSON with python-read-word-timings, and check a voice take or a finished mix against its script, word for word with no audio tag spoken. Covers which model to offer, the audio-start offset of a movie's sound, and the traps: word times that drift on long files and short windows, names misspelled, numbers written as digits, sentence ends lost. Use before any paid speech-to-text; use elevenlabs-speech-to-text instead when speakers must be told apart.

## Hypernatural web app

- [AI models](https://hypernatural.ai/ai-models): the AI video and image models Hypernatural runs
- [Web Creator](https://hypernatural.ai/web-creator): the browser-based video editor
- [Web Creator pricing](https://hypernatural.ai/web-creator/pricing): plans for the web app
- [Guides](https://hypernatural.ai/guides): guides to making AI video
- [Help](https://hypernatural.ai/help): help center