# GenMagic

> GenMagic is one studio for every generative AI model. Turn a single prompt into text, images, video, audio, music, or code, and route it to the best model automatically. New models the day they launch.

One studio and one API for every AI model, in every medium, from a single prompt.

## Pages
- [Models](https://genmagic.co/models)
- [Benchmarks](https://genmagic.co/benchmarks)
- [Pricing](https://genmagic.co/pricing)
- [API](https://genmagic.co/developers)
- [Blog](https://genmagic.co/blog)
- [About](https://genmagic.co/about)
- [LLM API pricing](https://genmagic.co/llm-api-pricing)
- [Video API pricing](https://genmagic.co/video-api-pricing)
- [Image API pricing](https://genmagic.co/image-api-pricing)
- [Lab API guides](https://genmagic.co/blog#api-guides)
- [AI labs](https://genmagic.co/labs)
- [Model comparisons](https://genmagic.co/compare)
- [Alternatives compared](https://genmagic.co/alternatives)
- [Pay-as-you-go AI video](https://genmagic.co/pay-as-you-go-ai-video-generator)
- [Model tests](https://genmagic.co/tests)
- [Model changes](https://genmagic.co/models/changes)
- [Measured model speed](https://genmagic.co/models/speed)
- [Our sources](https://genmagic.co/sources)
- [AI series generator](https://genmagic.co/ai-series-generator)
- [UGC video ads](https://genmagic.co/ugc-ads)
- [Virtual staging](https://genmagic.co/virtual-staging)
- [n8n workflows](https://genmagic.co/n8n)
- [Zapier steps](https://genmagic.co/zapier)
- [Make modules](https://genmagic.co/make)
- [MCP server](https://genmagic.co/mcp)
- [ComfyUI nodes](https://genmagic.co/comfyui)
- [Toonflow provider](https://genmagic.co/toonflow)
- [Photopea plugin](https://genmagic.co/photopea)
- [SketchUp and Rhino renders](https://genmagic.co/sketchup)
- [Anki audio and images](https://genmagic.co/anki)
- [Godot plugin](https://genmagic.co/godot)
- [Discord app](https://genmagic.co/discord)
- [AI in Lovable apps](https://genmagic.co/lovable)
- [Integrations](https://genmagic.co/integrations)
- [Referral program](https://genmagic.co/referrals)
- [Creator program](https://genmagic.co/creators)
- [For agencies](https://genmagic.co/agencies)
- [Consistent characters](https://genmagic.co/consistent-character-ai-video)
- [Sora 2 API alternative](https://genmagic.co/sora-2-api-alternative)

## Guides and model tests
- [Wan 3.0 API: Every Way to Get Access, and What Each Costs (2026)](https://genmagic.co/blog/wan-3-0-api): How to call Alibaba's Wan 3.0, Wan 3.0 Prime, 2.7 and 2.6 from code or a workflow: through Alibaba Cloud Model Studio (its regions, the limited-time 30% discount, the 30-second free quota and the 24-hour download window), with open weights for Wan 2.2, or with no Alibaba account through Replicate, kie.ai, fal, WaveSpeed, OpenRouter or GenMagic. Prices per second, working requests and which route fits which job.
- [Suno API in 2026: The Invite-Only Beta, the New MCP Connector, the Unofficial APIs, and Music APIs You Can Call Today](https://genmagic.co/blog/suno-api): Suno has no public API: its developer API is an invite-only beta, and on October 28 it launched an MCP connector for Claude, ChatGPT and Gemini. What Suno's terms say about automation and commercial use, which unofficial Suno APIs closed and which still sell access, and the official music APIs (Lyria, ElevenLabs, Stable Audio) with prices and working requests.
- [Seedream API: ByteDance's Seedream 5.0 Pro, Flash and Lite, Every Way to Get Access, and What Each Costs (2026)](https://genmagic.co/blog/seedream-api): How to call ByteDance's Seedream 5.0 Pro, 5.0 Flash, 5.0 Lite and 4.5 from code: price per image (from $0.018), what each model does, the default that doubles Pro's price, the watermark that is on unless you turn it off, the US restriction on BytePlus, the China retirements on November 24, and every platform that resells Seedream, with working requests.
- [Nano Banana API: Nano Banana 2.1, Pro and Lite, Every Way to Get Access, and What Each Costs (2026)](https://genmagic.co/blog/nano-banana-api): How to call Google's Nano Banana 2.1, Nano Banana Pro and Nano Banana 2 Lite from code: the Gemini API and its Interactions call, the Cloud Agent Platform (formerly Vertex AI), prices per image at 1K, 2K and 4K, why there is no free tier for images, which old model ids have shut down, and every platform that resells Nano Banana, from $0.02 an image.
- [Grok Imagine API: xAI's Video and Image Models, Every Way to Get Access, and What Each Costs (2026)](https://genmagic.co/blog/grok-imagine-api): How to call xAI's Grok Imagine Video 1.5, 1.5 Lite and Grok Imagine Image 2.0 from code: prices per second by resolution (from $0.02), what each model does, the image model retired November 2, rate limits, moderation charges and data rules, and every platform that resells Grok Imagine, with working requests.
- [FLUX API: Black Forest Labs' FLUX.2 and FLUX 3 Models, Every Way to Get Access, and What Each Costs (2026)](https://genmagic.co/blog/flux-api): How to call FLUX 3 Image, FLUX 3 Klein, FLUX.2 pro, max, flex and klein, and FLUX 3 Video from code: prices per image, per megapixel and per second (from $0.008), what each model does, megapixel rounding, Black Forest Labs' rules on data and reselling, open-weight licences, and every platform that resells FLUX, with working requests.
- [Seedance 2.0 API: Every Way to Get Access, and What Each Costs (2026)](https://genmagic.co/blog/seedance-2-0-api): How to call ByteDance's Seedance 2.0, 2.0 Fast, 2.0 Mini and 2.5 from code or a workflow: through BytePlus ModelArk (with its $30 activation gate, its region and country limits and its rule on real faces), or with no BytePlus account through OpenRouter, fal, Replicate, WaveSpeed, kie.ai or GenMagic. How token pricing turns into dollars per second, working requests, and which route fits which job.
- [Kling API: Kling 3.0 and 4.0, Every Way to Get Access, and What Each Costs (2026)](https://genmagic.co/blog/kling-api): How to call Kuaishou's Kling 4.0, 3.0, 3.0 Turbo, 3.0 Omni and O1 from code or a workflow: through Kling's own developer platform (units, resource packs, an API key and the Singapore endpoint), or with no Kling account through OpenRouter, fal, WaveSpeed, Replicate, kie.ai or GenMagic. Prices per second with and without sound, Kling 4.0's launch on October 28, working requests and which route fits which job.
- [AI Video and Image APIs Compared: 2026 Prices and Picks](https://genmagic.co/blog/ai-video-image-api-comparison): fal, Replicate, kie.ai, WaveSpeed, OpenRouter, Google direct and GenMagic, side by side: what Veo 3.1, Kling 3.0, Seedance 2.0, Wan and Nano Banana cost on each, how each one bills, which work in n8n, Zapier and MCP, and which to pick for what.
- [GPT Image API: GPT Image 2.6 and 2.5 Sunburst and Flare, Every Way to Get Access, and What Each Costs (2026)](https://genmagic.co/blog/gpt-image-api): OpenAI shut GPT Image 1 down on October 23, 2026 and removes GPT Image 1.5, 1 Mini and chatgpt-image-latest on December 1. How to call GPT Image 2.6 and 2.5 Sunburst and Flare and GPT Image 2 now: token prices turned into dollars per image at every quality, usage tiers, organization verification, the free credit ChatGPT Pro and Business subscribers can claim, and routes that need no OpenAI account.
- [Sora 2 API: Shut Down on September 24, 2026. What to Use Instead, Including a Drop-in Replacement for the Videos API](https://genmagic.co/blog/sora-2-api): OpenAI removed the Sora 2 API, sora-2 and sora-2-pro on September 24, 2026, with no replacement of its own. GenMagic answers the same Videos API calls (client.videos.create, retrieve, download_content; n8n's OpenAI node), so existing code moves with a base URL and a model name. Plus what Sora 2 cost and the models that replace it (Veo 3.1, Kling, Seedance, Wan).
- [Runway API (Runway Dev): Gen-4.5, Aleph 2.0 and 60 Other Models, What Each Costs (2026)](https://genmagic.co/blog/runway-api): How to call Runway's API, now called Runway Dev: Gen-4.5 at $0.12 a second, Gen-4 Turbo, Aleph 2.0 video editing and Act-Two, plus the Veo, Kling 4.0, Seedance, MiniMax, Wan and Grok models it resells under the same key. Credits, usage tiers, moderation charges, training terms, and how its prices compare with each model maker's and with GenMagic's.
- [Recraft API: SVG and Raster Image Generation, Every Way to Get Access, and What Each Costs (2026)](https://genmagic.co/blog/recraft-api): How to call Recraft V4.1, V4.1 Flash, the vector models and V4 Styles from code: price per image for raster and SVG (from $0.007), what pro, utility, flash and styles mean, the editing tools only Recraft's own API has, its rights and rules, and every platform that resells Recraft, with working requests.
- [Qwen Image API: Alibaba's Qwen-Image 3.0 and 2.1, Every Way to Get Access, and What Each Costs (2026)](https://genmagic.co/blog/qwen-image-api): How to call Alibaba's Qwen-Image 3.0, 3.0 Pro, 2.1 Pro and 2.1 Turbo from code: price per image (from $0.016), the 10 free images per model, regions that are cheaper than Singapore, the 2K rule that nearly doubles Pro's price, which Qwen-Image weights you may run commercially, and every platform that resells Qwen-Image, with working requests.
- [Midjourney API in 2026: Is There One, What the Unofficial Ones Risk, and What Developers Use Instead](https://genmagic.co/blog/midjourney-api): Midjourney still has no public API: what its own rules say about automation, the Enterprise API survey from 2025, which unofficial Midjourney APIs shut down and which still sell access, the risks they carry, and the official image APIs that do each Midjourney job, with prices and working requests.
- [Krea API: Krea 2 and Krea 3 (K3 Edit), Every Way to Get Access, and What Each Costs (2026)](https://genmagic.co/blog/krea-api): How to call Krea 2 Large, Medium and Turbo and Krea 3 from code: price per image (from $0.015), style references, moodboards and LoRAs, what Krea 3's edits cost by resolution and reference, why the same Krea 3 edit costs $0.08 on one platform and $0.24 on another, the open Krea 2 weights and their licence, and every platform that resells Krea, with working requests.
- [HeyGen API in 2026: Avatar IV and HeyGen Video Prices, the v3 Switch on October 31, and Every Way to Get Access](https://genmagic.co/blog/heygen-api): How to call HeyGen's avatar engines and its new HeyGen Video model from code: pay-as-you-go prices per minute and per second, the v1 and v2 endpoints retired on October 31, 2026, the launch discount that ended the same day, limits, consent and data rules, and every platform that resells HeyGen, with working requests.
- [Hailuo API (MiniMax H3): Every Way to Get Access, and What It Costs (2026)](https://genmagic.co/blog/hailuo-api): How to call MiniMax's Hailuo video models from code: MiniMax H3 (what most people call Hailuo 3), H3 Highspeed, Lite and Max, and the older Hailuo 2.3, through MiniMax's own API at $0.025 to $0.13 a second, or with no MiniMax account through Replicate, fal, kie.ai, WaveSpeed, OpenRouter or GenMagic. Prices, working requests, open weights and which route fits which job.
- [Free AI Video and Image API Credits in 2026: Who Gives What, and What It Buys](https://genmagic.co/blog/free-ai-video-image-api-credits): Which AI video and image APIs give free credit in late 2026, and how many images or seconds of video each one actually buys: Google, OpenAI, Alibaba, ByteDance, Kling, kie.ai, WaveSpeed, fal, Replicate, OpenRouter and GenMagic, read on their own pages, with the conditions attached.
- [Veo 3.1 API: Every Way to Get Access, and What Each Costs (2026)](https://genmagic.co/blog/veo-3-1-api): Google shut Veo 3.1 down on the Gemini API on October 22, 2026. How to call Veo 3.1, Fast and Lite now: from Google's Agent Platform (formerly Vertex AI), or with no Google Cloud account through OpenRouter, fal, Replicate, WaveSpeed, kie.ai or GenMagic, with prices per second, what each tier can do, working requests and which route fits which job.
- [Best AI Virtual Staging Models in 2026: Nine Tested on Two Empty Rooms](https://genmagic.co/blog/best-ai-virtual-staging-tested): Nine image models furnished two empty rooms for a property listing. Eight kept the windows, doors, radiator, beam and skylight where they were, from 1.8 cents a photo in the test; we measured how much of each room they changed, and one rebuilt both rooms.
- [Best AI Thumbnail Generator Models in 2026: Nine Tested on Two YouTube Thumbnails](https://genmagic.co/blog/best-ai-thumbnail-generator-tested): Nine image models made a YouTube thumbnail with a creator's face and one without. All nine set the faceless title exactly; seven kept the face and the title on the harder one, from 1.8 cents a thumbnail in the test. One invented a channel name and three claims.
- [Best AI Photo Restoration Models in 2026: Nine Tested on One Damaged Photo](https://genmagic.co/blog/best-ai-photo-restoration-tested): Nine image models restored a creased, stained, torn 1950s photo, once in black and white and once in color. Six removed all the damage, five of them keeping the couple's faces, from 1.2 cents a photo in the test; two left the torn corner or stains, and one kept the damage and replaced the people.
- [Best AI Headshot Generator Models in 2026: Nine Tested on Two Selfies](https://genmagic.co/blog/best-ai-headshot-generator-tested): Nine image models turned two casual selfies into professional headshots. Six kept both faces, freckles, glasses and grey beard included, from 1.2 cents a headshot in the test; two added glasses nobody wears, and one replaced the person.
- [Best AI Coloring Page Generator Models in 2026: Nine Tested on Two Pages](https://genmagic.co/blog/best-ai-coloring-page-generator-tested): Nine image models drew a children's coloring page and an intricate adult one. All nine kept to black lines on white with no color or gray fill; they differed on line weight, how much of the page they used and what they added. The cheapest cost 1.8 cents a page.
- [Best AI Ad Generator Models in 2026: Nine Tested on One Product Photo](https://genmagic.co/blog/best-ai-ad-generator-tested): Nine image models turned one product photo into a square feed ad and a vertical story ad with exact wording. Seven got every word and the label right, the cheapest for about a cent an ad in the test; one misspelled, one added a claim nobody wrote, one swapped the product.
- [Veo 3.1 vs Veo 3.1 Fast vs Lite: One Ad, Three Tiers](https://genmagic.co/blog/veo-3-1-lite-vs-fast-vs-veo-3-1): Google's Veo 3.1 comes in three tiers, and the top one costs eight times the cheapest. We made the same six-second spoken UGC ad on each: all three said the line word for word, and the cheapest kept the product label sharpest.
- [How to Price AI UGC Ads for Clients: 2026 Rates and Margins](https://genmagic.co/blog/how-to-price-ai-ugc-ads): What clients pay for AI-made UGC and video ads in 2026, what one really costs to produce (generation, retakes, people's time), and how to set per-ad, batch and retainer prices with the margin written out, from real model prices.
- [Consistent Characters in AI Video: Four Models Tested](https://genmagic.co/blog/consistent-character-ai-video-test): One character sheet, one scene, four video models that follow reference images: Wan 3.0, Seedance 2.0 Mini, HappyHorse 1.1 and Grok Imagine Video. All four kept the character; here is how they differed, what each cost and how long it took.
- [Cheapest AI Video Models in 2026: Six Tested on One Clip](https://genmagic.co/blog/cheapest-ai-video-models-tested): We made the same five-second clip on the six cheapest AI video models, for 6.5 cents to 28 cents in the test. All six were usable; here is what each cost, how long it took and where the cheaper ones fall short.
- [Best AI Model for Building a Website: Eight Tested on One Brief](https://genmagic.co/blog/best-ai-website-generator-tested): Claude Sonnet 5, Claude Opus 5.5, GPT-6.1 Terra, Gemini 4 Argon, Kimi K3, Muse Wonder, Muse Spark 1.3 and GLM 5.1 built the same landing page from one brief. Seven delivered a complete, phone-ready page for 5 to 38 cents in the test; three invented reviews; one rarely finished.
- [Best AI Video Model for UGC Ads: Five Models, One Brief](https://genmagic.co/blog/best-ai-video-model-for-ugc-ads): We gave Veo 3.1 Lite, Veo 3.1 Fast, Kling 3.0, Wan 3.0 and Seedance 2.0 the same creator photo and the same spoken line. Four said it word for word, one refused, and the cheapest was also the fastest. Real clips, costs, render times and transcripts.
- [Best AI Video Generator in 2026: Eight Models Tested on Two Scenes](https://genmagic.co/blog/best-ai-video-generator-tested): Veo 3.1 Fast, Kling 3.0 Pro, Seedance 2.5, Wan 3.0, Hailuo 3, Runway Gen-4.5, Grok Imagine Video 1.5 and HappyHorse 1.1 on the same action shot and the same spoken line. Seven of eight said the line word for word; in the test the clips cost $0.55 to $1.51 each and took from under a minute to four.
- [Best AI Logo Generator Models in 2026: Eight Tested on Two Briefs](https://genmagic.co/blog/best-ai-logo-generator-tested): Nano Banana 2.1, GPT Image 2.5, Recraft V4.1 and its vector model, FLUX 3, Seedream 5.0 Pro, Qwen Image 3 Pro and FLUX 3 Klein on the same two logo briefs. All eight spelled the name right; one returned a real SVG; the best ideas cost about 5 cents in the test.
- [Best AI Image Model for Product Photos: Six Models Tested](https://genmagic.co/blog/best-ai-image-model-for-product-photos): One packshot, one instruction, six image models: Nano Banana 2.1, Nano Banana Pro, GPT Image 2, Seedream 5.0 Pro, FLUX.2 Pro and Qwen Image 3 Pro. All six kept the label down to the small print; price and speed decided it.
- [Best AI Image Generator in 2026: Nine Models Tested on Four Jobs](https://genmagic.co/blog/best-ai-image-generator-tested): Nano Banana 2.1, GPT Image 2.5, FLUX 3, Seedream 5.0 Pro, Qwen Image 3 Pro, Recraft V4.1, Grok Imagine, MAI Image 2.6 and FLUX 3 Klein on the same four prompts: exact text, a candid photo, a flat illustration and a scene with things to count. All nine passed; in the test an image cost 1.2 to 7.8 cents and took 6 to 78 seconds.
- [Consistent Characters in AI Video: The 2026 Guide](https://genmagic.co/blog/consistent-characters-ai-video): Why AI video forgets who your characters are between shots, which models in late 2026 can hold a face and a product steady, how to build references and prompts that work, and what a consistent minute of video costs.
- [AI UGC Ads in 2026: Make Them, Test Them, Stay Legal](https://genmagic.co/blog/ai-ugc-ads): How to turn one product photo into a spoken, vertical UGC-style video ad for under a dollar, what it costs against creators and AI UGC tools, how to test hooks, and the FTC, platform and EU rules that decide whether you can run it.
- [LLM Benchmarks in 2026: Which Ones Still Matter](https://genmagic.co/blog/llm-benchmarks-2026): Most LLM benchmarks are saturated, flawed or easy to game in 2026. Which ones still separate models, how to read a score, and how to choose a model with them.
- [Media Generation for AI Agents: The 2026 Guide](https://genmagic.co/blog/media-generation-for-ai-agents): How to give AI agents the power to generate images, video, speech, music, and on-brand assets through one API and one MCP server, with verified 2026 models.
- [Introducing GenMagic: one place for every AI model](https://genmagic.co/blog/introducing-genmagic): The best AI models are scattered across a dozen companies. GenMagic puts every one of them behind a single prompt box, across text, image, speech, video, and music.
- [You shouldn't have to pick an AI model](https://genmagic.co/blog/you-shouldnt-have-to-pick-a-model): Choosing the right model for every task is a job nobody signed up for. Here is why routing beats picking, and how a single prompt box gets you a better result with none of the homework.

## API

GenMagic exposes an OpenAI-compatible REST API and a remote Model Context Protocol (MCP) server, so an agent can generate text, images, speech, music, and video itself.

- Base URL: https://genmagic.co/api/v1
- Auth: `Authorization: Bearer gm_live_<key>` (create a free account and a key at https://genmagic.co/get-api-key). The model catalog needs no key.
- Billing: pay-as-you-go against a prepaid USD balance; cost and remaining balance ride response headers.

### Endpoints
- `POST /api/v1/chat/completions` - OpenAI-compatible chat/text completion. Point any OpenAI SDK at this base URL. Routes the messages to the selected model through GenMagic, billed against the key's real USD balance at the model's price in /api/v1/models (cached input and long-context prompts bill at the cache and tier prices listed there). The request body is forwarded to the model unchanged, so every parameter in the model's supported_parameters works as sent (temperature, tools, response_format, seed, ...). Streams with server-sent events when stream=true. When the key owner has Brand personalization on, a brand system message is prepended automatically.
- `POST /api/v1/text` - Typed, brand-steered text artifact. GenMagic-native. Returns a finished artifact rather than raw model output: pass type to unlock the artifact's system prompt and, for svg/website, the brand's exact palette and typeface. Omit type for plain, voice-branded text.
- `POST /api/v1/images/generations` - OpenAI-compatible image generation and edit. Generate an image from a prompt, or transform a reference image (image-to-image) by passing image. Set type=logo to steer as a brand mark. On-brand automatically when personalization is on. Returns { created, model, data: [{ url }] } (or b64_json), where model is the model that made the image, also when you left model out.
- `POST /api/v1/images/edits` - OpenAI-compatible image edit (upload a picture). OpenAI's image edit call, so client.images.edit in the OpenAI SDKs and apps built on them work unchanged: upload the picture as the image file and describe the change in prompt. The same engine, price and response as /images/generations with a reference image. A JSON body with image as an https or data:image URL also works. Masks are refused (describe the change in the prompt instead).
- `POST /api/v1/audio/speech` - OpenAI-compatible text-to-speech. Turn text into spoken audio. Returns raw audio bytes; a durable hosted URL rides the X-Media-Url header. Voiced verbatim (never brand-altered).
- `POST /api/v1/audio/music` - Original music track from a description. GenMagic-native (OpenAI has no music endpoint). Returns raw audio bytes; a hosted URL rides X-Media-Url. Brand-steered in mood when personalization is on.
- `POST /api/v1/videos` - Start an asynchronous video job. Returns 202 immediately with { id, object: "video.generation", model, status: "queued", created (Unix seconds), status_url }; nothing is charged until the clip completes. Poll GET /api/v1/videos/{id} until status is completed or failed. Also OpenAI's Videos API: a multipart request (what the OpenAI SDKs' client.videos.create sends) or one with seconds, size or input_reference is answered with OpenAI's Video object, an id starting video_, at HTTP 200, so code written for Sora 2's Videos API runs unchanged with a GenMagic model id; a request that still names sora-2 runs on google/veo-3.1-fast and sora-2-pro on google/veo-3.1 (the answer's model says which ran).
- `GET /api/v1/videos/{id}` (free) - Poll a video job. Returns { id, object, model, status } with status queued or processing while the clip renders; when completed it adds the durable hosted url and cost_cents. A render failure is a terminal failed status at HTTP 200 with error, a plain-text message. Charged once, on completion. An id starting video_ (from an OpenAI-style create) is answered with OpenAI's Video object instead: status queued, in_progress, completed or failed, progress, created_at, and error as { code, message }.
- `GET /api/v1/videos/{id}/content` (free) - Download a finished clip. OpenAI's Videos API download (client.videos.download_content): redirects (302) to the finished clip's durable file. 409 video_not_ready while it renders, 409 video_failed for a failed job. Only the clip exists: variant=thumbnail or spritesheet answers 404. Works with either id form.
- `GET /api/v1/models` (free, no key needed) - List models with full detail. OpenAI-compatible model list enriched with capabilities, pricing, and benchmarks from the same live catalog the studio uses, newest first (by created, a Unix timestamp in seconds). Text models also carry supported_parameters, capabilities.reasoning_efforts and reasoning_mandatory, and cache, reasoning and long-context (tiers) prices where the model publishes them; audio models carry audio_kind (speech or music), the endpoint that runs them. A pure lookup: spends no credits, and works without a key.
- `GET /api/v1/models/{id}` (free, no key needed) - Retrieve one model's full detail. OpenAI-compatible client.models.retrieve. A pure lookup: spends no credits, and works without a key. Model ids contain a slash, e.g. /api/v1/models/openai/gpt-image-2.
- `GET /api/v1/models/changes` (free, no key needed) - What changed in the model catalog, day by day. New models (by the catalog's listing date), price changes with the price before and after, retirements, returns, changed details, and benchmark scores that appeared or changed, per UTC day, with how many model pages each day touched: the data behind genmagic.co/models/changes. A pure lookup: spends no credits, and works without a key.
- `GET /api/v1/benchmarks` (free) - List benchmarks with definitions. Every benchmark in the matrix, described identically (what it tests, metric, status, maintainer, source). A pure lookup: spends no credits. Each links to /api/v1/benchmarks/{id} for its model leaderboard.
- `GET /api/v1/benchmarks/{id}` (free) - Benchmark detail and model leaderboard. One benchmark's full definition plus its model leaderboard ranked best-first, each score carrying its source and a link to the original. Answers 'which model is best at benchmark X'. A pure lookup: spends no credits.
- `GET /api/v1/usage` (free) - Read the key's credit balance and spend. Returns the authenticated key owner's live balance and lifetime spend so a program can check budget before a run. Spends no credits.
- `POST /api/v1/agent/turn` (free) - Agent reply plus a cross-modal asset plan. The API face of the studio Agent. Send a conversation; it replies in character and returns a plan of on-brand assets to make. Planning is orchestration and is not billed (you then generate each asset via the matching endpoint).

### MCP server

Point VS Code, Cursor, Claude Code, Claude Desktop, or any MCP client at `https://genmagic.co/api/mcp` with a GenMagic API key as the Bearer token (one-click VS Code and Cursor installs at https://genmagic.co/developers#mcp). Streamable HTTP, stateless, protocol 2025-06-18. Tools:
- `list_models` - List the models you can pick, with ids and prices (filter by category or search); no key needed.
- `generate_text` - Generate text, code, an SVG graphic, or a full web page with any text model (optional type and model).
- `generate_image` - Generate or edit an image with any image model; type=logo for a brand mark.
- `generate_speech` - Text to speech with any speech model and voice, voiced verbatim.
- `generate_music` - Original music track from a description.
- `create_video` - Start an async video job with any text-to-video model, length, resolution and sound, or animate an image (returns an id, charges nothing yet).
- `get_video` - Poll a video job; charges once on completion.

Apps that connect by signing in instead of sending a key (Claude and ChatGPT custom connectors, any client that follows the MCP authorization spec) use `https://genmagic.co/api/mcp/connect`: the person signs in to GenMagic, approves, and the app receives a token (OAuth 2.1 with PKCE, dynamic client registration; metadata at https://genmagic.co/.well-known/oauth-authorization-server). Setup for each app: https://genmagic.co/mcp.

### Integrations

GenMagic also runs inside workflow and agent builders and creative tools, with the same account and balance:
- Dify: [Dify Marketplace plugin](https://marketplace.dify.ai/plugin/genmagic/genmagic). Install the GenMagic plugin from the Dify Marketplace and paste your API key into its settings.
- n8n: [n8n verified community node (npm package n8n-nodes-genmagic)](https://www.npmjs.com/package/n8n-nodes-genmagic). In n8n Cloud or self-hosted n8n, open the nodes panel, search for GenMagic, select it under More from the community and install it (an owner or admin installs it once per instance), then add a GenMagic API credential. Self-hosted n8n can also install the package n8n-nodes-genmagic from Settings > Community Nodes.
- ComfyUI: [ComfyUI custom node pack on the Comfy Registry (node id genmagic)](https://registry.comfy.org/nodes/genmagic). In ComfyUI Manager, search for GenMagic and install it (or run comfy node install genmagic), restart ComfyUI and paste your API key in Settings > GenMagic > API key. Nodes: GenMagic Image (text-to-image, image-to-image), GenMagic Video (text-to-video, image-to-video), GenMagic Speech, GenMagic Music and GenMagic Text.
- Vercel AI SDK: [OpenAI-compatible provider guide (@ai-sdk/openai-compatible)](https://genmagic.co/developers#ai-sdk). npm install ai @ai-sdk/openai-compatible, then createOpenAICompatible with your GenMagic API key and the API base URL; generateText, streamText, tool calling and generateImage work through it.
- Postman: [Public Postman workspace (the GenMagic API collection)](https://www.postman.com/genmagic-1301793/genmagic-api/overview). Fork the GenMagic API collection into your own workspace, set its Bearer token to your API key, and send any request; GET /models answers without a key.
- Zapier: [Zapier integration (beta, joined by invite link until Zapier lists it in its app directory)](https://zapier.com/developer/public-invite/247302/515325/8c30f92e3875254f2f4aa5dedc7d67b8/). Open the invite link while signed in to Zapier to add GenMagic to your account, then connect it with your GenMagic API key; the steps are Generate Image, Generate Text, Generate Speech, Start Video and Find Video.
- Make: [Make app (published, in Make's app review; added by invite link until Make approves it)](https://www.make.com/en/hq/app-invitation/6ff111f8c59c3853cd1b922ef92fadc8). Open the invite link while signed in to Make to install GenMagic in your organization, then create a GenMagic connection with your API key; modules: Generate an image, Generate text, Generate speech, Create a video, Get a video, Search models and Make an API call.
- SketchUp: [SketchUp extension GenMagic Render (.rbz; in the SketchUp Extension Warehouse review, installable from this site today)](https://genmagic.co/sketchup). Download genmagic_render.rbz, then in SketchUp 2017 or later open Extensions > Extension Manager > Install Extension and pick it; Extensions > GenMagic Render renders the current view (photoreal, interior, dusk, watercolor or concept) with the image model you pick, after Sign in with GenMagic or an API key.
- Rhino 8: [Rhino 8 Python script GenMagic Render (in Food4Rhino's review, installable from this site today)](https://genmagic.co/sketchup). Download and unpack the zip, run RunPythonScript in Rhino 8 and pick GenMagicRender.py; it renders the active viewport with the image model you pick and can place the render in the model as a picture frame.
- Anki: [Anki add-on GenMagic: AI audio and images for your notes (.ankiaddon, installable from this site; AnkiWeb listing to follow)](https://genmagic.co/anki). Download genmagic.ankiaddon, then in Anki open Tools > Add-ons > Install from file; in the Browse window select notes and choose Notes > GenMagic: add audio or images to selected notes to add AI speech or an image to every selected note.
- Godot: [Godot 4 editor plugin (in the Godot Asset Store and Asset Library review, installable from this site today)](https://genmagic.co/godot). Download the zip, copy addons/genmagic into your project, enable GenMagic in Project > Project Settings > Plugins and paste your API key into the GenMagic dock; it makes images (sprites, textures, concept art), speech clips and music into res://genmagic.

## Pricing

Prepaid, pay-as-you-go on a real USD balance (1 credit = 1 US cent), not a per-provider subscription. Free to start: create an account and verify your email for a one-time $0.50 starter balance, one per person (temporary and forwarding email addresses, and inboxes or company domains that already received it, do not qualify). Each generation draws its model's price, listed in GET /api/v1/models and shown before you run it; the API and the studio charge the same. Top-ups are one-time credit purchases of $5, $10, $25, $50, $100 or $500. Optional auto top-up refills the balance from a saved card when it falls below a line the account sets, and before a request the balance cannot cover, within a monthly limit the account chooses. Optional monthly plans add more credit than they cost every month: Pro at $18/mo adds $20, Studio at $49/mo adds $55, Agency at $149/mo adds $167. Plan credit is for its month: what is unused when the plan renews does not carry over, and it is spent before any top-up credit. Top-up credit never expires.

## Machine-readable

- OpenAPI spec: [https://genmagic.co/openapi.json](https://genmagic.co/openapi.json)
- Agent card (A2A): [https://genmagic.co/.well-known/agent-card.json](https://genmagic.co/.well-known/agent-card.json)
- MCP manifest: [https://genmagic.co/.well-known/mcp.json](https://genmagic.co/.well-known/mcp.json)
- Agent Skills: [https://genmagic.co/.well-known/skills/index.json](https://genmagic.co/.well-known/skills/index.json), install with `npx skills add https://genmagic.co/`
- Official packages: `npm install genmagic` (JavaScript and the `genmagic` command), `pip install genmagic` (Python, with `genmagic.tools` for agent frameworks), `composer require genmagic/genmagic` (PHP and Laravel), `npm install genmagic-ai-provider` (Vercel AI SDK), `npm install genmagic-tanstack-ai` (TanStack AI)
- Coding-agent plugin (MCP server with sign-in plus a skill): `codex plugin marketplace add https://codeberg.org/genmagic/genmagic-plugin.git` or `claude plugin marketplace add https://codeberg.org/genmagic/genmagic-plugin.git`