# MuseSpan > 一套 HTTP API 调用多家厂商的原生 AI 模型能力，当前可用的类型见下方实时目录。One API for vendor-native AI model capabilities; the live catalog below lists what is available now. `/v1/models` 每条的 `id` 就是请求里 `model` 要填的值，请直接复制，不要自己拼。图片和视频的 `id` 以能力结尾（例如 `alibaba/wan-2.7/image-to-image`），一个模型可以有多个能力。音频模型只有一个能力，它的 `id` 通常就是模型 ID（`厂商/模型` 两段，与 `group.id` 相同）。`group.id` 与 `id` 不同时只用于分组，不能填进 `model`。提交时返回 404 `model_not_found`，说明这个 ID 不能调用：不要原样重发，改用响应里的 `suggested_models` 或 `/v1/models` 的 `id`。 Each `/v1/models` entry's `id` is exactly the `model` value of a request; copy it rather than composing one. Image and video IDs end with the capability (for example `alibaba/wan-2.7/image-to-image`), and one model can have several. An audio model has exactly one capability, and its `id` is normally the model ID itself (`owner/model`, the same as its `group.id`). When `group.id` differs from `id`, it only groups entries and cannot be sent as `model`. A POST that returns 404 `model_not_found` means the ID is not callable: do not resend it; use `suggested_models` from the response or an `id` from `/v1/models`. 先从下方选择能力，再读取该能力的 `llms.txt`;它由实时 Registry、公开目录和价格生成，包含当前参数、约束、计费和可运行示例。API base URL 是 `https://api.musespan.com`,鉴权使用 HTTP header `Authorization: Bearer `(密钥在 https://www.musespan.com/settings/keys 创建)。图片能力按各自 Schema 分为同步或异步；模型任务能力(如文字转语音)的结果文件(图片、音频)需用提交任务的同一把密钥下载。异步图片任务(`/v1/images/tasks`)和模型任务(`/v1/model-tasks`)的任务 ID 就是你的 `X-Request-Id`,提交是幂等的：超时、429/5xx 或查询返回 404 时，用同一个 ID 原样重新提交，不会重复扣费。受理后 15 分钟内仍未出结果(或提交后 15 分钟仍未确认)的异步图片任务和模型任务会判定失败(`error.code` 为 `generation_timeout`)并全额退款，异步图片任务轮询等待 20 分钟即可看到最终状态；模型任务超过可同时运行的数量时不会被拒绝，而是先排队(状态 `queued`，带估计的 `queue_position`)，提交 5 分钟后仍在排队的判定失败(`error.code` 为 `queue_timeout`)并全额退款，排队中的任务可用 `POST /v1/model-tasks/{id}/cancel` 取消并全额退款，模型任务轮询等待 25 分钟；生成服务暂时不可用、任务未能开始时会立即判定失败(`error.code` 为 `upstream_unavailable`)并全额退款，这不是输入问题，稍后重试即可；模型任务的输入被拒绝时任务判定失败(`error.code` 为 `invalid_parameters`)并全额退款,`error.issues` 列出被拒绝的字段(`field`)和规则码(`rule`),改好这些字段后再提交；重新生成需用新的 `X-Request-Id`,原 ID 会一直返回这条失败任务。视频(`/v1/videos`)不幂等，保存返回的任务 ID 轮询，不要自动重发。确认失败的请求不计费。 ## 文档 / Docs 每个链接都是原始 markdown(`.md` 后缀),去掉 `.md` 是给人看的 HTML 版本。 - [快速开始](https://www.musespan.com/docs/quickstart.md): 从实时能力目录选择图片、语音等模型，调用 API 并安全获取结果。 - [错误处理与计费规则](https://www.musespan.com/docs/errors.md): 区分请求拒绝、任务失败和网络超时，避免重复生成与重复费用。 ## 模型能力 / Model capabilities 以下是当前全部可用能力，由实时 Registry 和目录生成。每条的 `model` ID 都可直接放入请求的 `model` 字段。打开对应的 `llms.txt` 获取该能力当前的输入输出 schema、默认值、限制、价格和调用示例。The index is generated from the live Registry and catalog. 精简机器可读目录:https://www.musespan.com/v1/models。`?category=<类型>` 可单独使用，也可与分页组合（如 `?category=image-to-image&limit=100`）；类型：`text-to-image`、`image-to-image`、`text-to-video`、`image-to-video`、`reference-to-video`、`text-to-speech`（旧代码 `edit`、`image-edit` 等同 `image-to-image`），也可使用目录条目里出现的任意 `category` 值；多个值用逗号分隔，取并集；未知取值返回 400，`error.allowed_values` 列出可用值。条目较多时用 `?limit=100` 分页，把返回的 `next_cursor` 作为下一页 `after`。`?expand=input_schema` 只展开图片接口能力的参数约束和示例输入；模型任务（`/v1/model-tasks`）能力的 `input` 参数、请求示例和结果结构在单能力 llms.txt 与 OpenAPI 中。Filter with `?category=`, with or without `limit`; an unknown value returns 400 listing `error.allowed_values`. ### Bytedance: Seedream 5 Flash - [bytedance/seedream-5-flash/text-to-image](https://www.musespan.com/models/bytedance/seedream-5-flash/text-to-image/llms.txt): Text To Image — ByteDance Seedream 5.0 Flash generates images from text on MuseSpan, with multilingual lettering and flexible sizes for posters and ads. Designed for fast creative iteration, it handles structured compositions for product concepts and social media graphics. Describe the subject, composition, lighting and style, then choose 1K or 2K presets, portrait or landscape formats, or custom dimensions. Download results as JPEG or PNG, and integrate generation into creative workflows through the API. Put desired lettering in quotation marks to make text instructions clearer. - [bytedance/seedream-5-flash/image-to-image](https://www.musespan.com/models/bytedance/seedream-5-flash/image-to-image/llms.txt): Image To Image — ByteDance Seedream 5.0 Flash Edit transforms images with text instructions and up to 10 references for product visuals and creative edits. Change backgrounds, materials, colors or visual styles, and combine references to guide a new composition. Specify what should change and what should stay consistent with the source. MuseSpan offers this image editing API with 1K and 2K presets, custom dimensions, and JPEG or PNG output for design workflows and content production. ### OpenAI: GPT Image 2.5 Flare - [openai/gpt-image-2.5-flare/text-to-image](https://www.musespan.com/models/openai/gpt-image-2.5-flare/text-to-image/llms.txt): Text To Image — OpenAI GPT Image 2.5 Flare is OpenAI's everyday image model, fast and high quality with natural lighting, rich textures and complex layouts. Render crisp text, transparent backgrounds and up to 10 images per call, with six quality levels from low to max and seven size presets. - [openai/gpt-image-2.5-flare/image-to-image](https://www.musespan.com/models/openai/gpt-image-2.5-flare/image-to-image/llms.txt): Image To Image — OpenAI GPT Image 2.5 Flare edits images from plain instructions, changing only what you ask while subjects, layout and background stay intact. Combine up to 16 reference images, add a mask for local edits, choose from six quality levels and get up to 10 results per call. ### OpenAI: GPT Image 2.5 Sunburst - [openai/gpt-image-2.5-sunburst/text-to-image](https://www.musespan.com/models/openai/gpt-image-2.5-sunburst/text-to-image/llms.txt): Text To Image — OpenAI GPT Image 2.5 Sunburst is the precision tier of GPT Image 2.5, trading longer generation time for extra fidelity in intricate detail. Built for premium visuals, maps, packaging and dense layouts, with quality up to max, transparent backgrounds and up to 10 images per call. - [openai/gpt-image-2.5-sunburst/image-to-image](https://www.musespan.com/models/openai/gpt-image-2.5-sunburst/image-to-image/llms.txt): Image To Image — OpenAI GPT Image 2.5 Sunburst makes the most tightly scoped GPT Image edits, keeping subject and composition steady across many revisions. Use up to 16 reference images and an optional mask, with quality up to max and up to 10 results per call. ### OpenAI: GPT Image 2 - [openai/gpt-image-2/text-to-image](https://www.musespan.com/models/openai/gpt-image-2/text-to-image/llms.txt): Text To Image — OpenAI GPT Image 2 creates highly detailed images with fine, legible typography, from invitations and posters to product shots. Choose low to high quality and seven size presets, add a transparent background, export PNG, JPEG or WebP and generate up to 4 images per call. - [openai/gpt-image-2/image-to-image](https://www.musespan.com/models/openai/gpt-image-2/image-to-image/llms.txt): Image To Image — OpenAI GPT Image 2 makes fine-grained edits to your images, from rewriting in-image text to restyling props and scenes. Combine up to 16 reference images, add a mask for targeted changes, pick low to high quality and get up to 4 results per call. ### OpenAI: GPT Image 1.5 - [openai/gpt-image-1.5/text-to-image](https://www.musespan.com/models/openai/gpt-image-1.5/text-to-image/llms.txt): Text To Image — OpenAI GPT Image 1.5 generates high-fidelity images with strong prompt adherence, getting counts, colors and placements right. Choose square, landscape or portrait sizes, low to high quality and an optional transparent background, with up to 4 images per call. - [openai/gpt-image-1.5/image-to-image](https://www.musespan.com/models/openai/gpt-image-1.5/image-to-image/llms.txt): Image To Image — OpenAI GPT Image 1.5 edits images while preserving composition, lighting and fine detail, with a high-fidelity mode for faces and products. Use up to 10 reference images and an optional mask, with low to high quality and up to 4 results per call. ### OpenAI: GPT Image 1 Mini - [openai/gpt-image-1-mini/text-to-image](https://www.musespan.com/models/openai/gpt-image-1-mini/text-to-image/llms.txt): Text To Image — OpenAI GPT Image 1 Mini is the low-cost GPT Image model, built for quick, efficient generation of icons, social graphics and drafts. Pick square, landscape or portrait sizes, low to high quality and a transparent background, with up to 4 images per call. - [openai/gpt-image-1-mini/image-to-image](https://www.musespan.com/models/openai/gpt-image-1-mini/image-to-image/llms.txt): Image To Image — OpenAI GPT Image 1 Mini makes fast, low-cost edits from text instructions, ideal for quick variations of icons, graphics and product images. Use up to 10 reference images, choose low to high quality and a transparent background, and get up to 4 results per call. ### Black Forest Labs: FLUX.1 Kontext Max - [black-forest-labs/flux-kontext-max/text-to-image](https://www.musespan.com/models/black-forest-labs/flux-kontext-max/text-to-image/llms.txt): Text To Image — Black Forest Labs FLUX.1 Kontext Max generates images from text with top-tier prompt adherence and clean, legible typography. Ideal for signs, packaging and posters, with 9 aspect ratios, guidance from 1 to 20, prompt enhancement and up to 4 images per call. - [black-forest-labs/flux-kontext-max/image-to-image](https://www.musespan.com/models/black-forest-labs/flux-kontext-max/image-to-image/llms.txt): Image To Image — Black Forest Labs FLUX.1 Kontext Max is the premium Kontext editor, with stronger prompt adherence and cleaner typography for precise, consistent edits. Add or replace signs, labels and headlines, restyle scenes or swap objects, with guidance from 1 to 20, 9 aspect ratios and up to 4 results per call. ### Black Forest Labs: FLUX.1 Kontext Pro - [black-forest-labs/flux-kontext-pro/text-to-image](https://www.musespan.com/models/black-forest-labs/flux-kontext-pro/text-to-image/llms.txt): Text To Image — Black Forest Labs FLUX.1 Kontext Pro also generates images from text, with strong prompt adherence and consistent, polished results. Pick from 9 aspect ratios, adjust guidance from 1 to 20, turn on prompt enhancement and create up to 4 images per call. - [black-forest-labs/flux-kontext-pro/image-to-image](https://www.musespan.com/models/black-forest-labs/flux-kontext-pro/image-to-image/llms.txt): Image To Image — Black Forest Labs FLUX.1 Kontext Pro edits a single image from plain instructions, changing only what you ask while keeping people, objects and composition consistent. Use it for scene and lighting changes, object swaps and style edits, with guidance from 1 to 20, 9 aspect ratios and up to 4 results per call. ### Black Forest Labs: FLUX 3 - [black-forest-labs/flux-3/text-to-image](https://www.musespan.com/models/black-forest-labs/flux-3/text-to-image/llms.txt): Text To Image — Black Forest Labs FLUX 3 is the newest FLUX model, built for detailed, well-composed images with precise layout and cleaner in-image text. Render in native 2K or 4K, or as fast 512 and 768 square drafts, across 14 aspect ratios plus auto, with optional prompt expansion for packaging, ads and editorial art. - [black-forest-labs/flux-3/image-to-image](https://www.musespan.com/models/black-forest-labs/flux-3/image-to-image/llms.txt): Image To Image — Black Forest Labs FLUX 3 makes precise local edits from written instructions, changing only what you ask while the rest of the image stays intact. Combine up to 10 reference images into one scene, output at 1K, 2K or 4K or as quick square drafts, and keep the input framing with auto aspect ratio or pick from 14 ratios. ### Black Forest Labs: FLUX.2 Dev - [black-forest-labs/flux-2-dev/text-to-image](https://www.musespan.com/models/black-forest-labs/flux-2-dev/text-to-image/llms.txt): Text To Image — Black Forest Labs FLUX.2 Dev is the open-weight FLUX.2 model, pairing strong realism and crisp text with full control over quality and speed. Tune guidance from 0 to 20 and steps from 4 to 50, add prompt expansion or acceleration, and generate up to 4 images per call. - [black-forest-labs/flux-2-dev/image-to-image](https://www.musespan.com/models/black-forest-labs/flux-2-dev/image-to-image/llms.txt): Image To Image — Black Forest Labs FLUX.2 Dev edits images from written instructions with the open-weight FLUX.2 model, keeping text and details sharp. Combine up to 10 reference images and tune guidance, steps and acceleration, with six size presets and up to 4 results per call. ### Black Forest Labs: FLUX.2 Klein 9B - [black-forest-labs/flux-2-klein-9b/text-to-image](https://www.musespan.com/models/black-forest-labs/flux-2-klein-9b/text-to-image/llms.txt): Text To Image — Black Forest Labs FLUX.2 Klein 9B is the compact FLUX.2 model, built for fast, low-cost images with sharp realism and clean text. Use 4 to 8 steps for quick drafts and high-volume work, with six size presets, PNG, JPEG or WebP output and up to 4 images per call. - [black-forest-labs/flux-2-klein-9b/image-to-image](https://www.musespan.com/models/black-forest-labs/flux-2-klein-9b/image-to-image/llms.txt): Image To Image — Black Forest Labs FLUX.2 Klein 9B makes fast, low-cost edits from written instructions, ideal for quick touch-ups at scale. Combine up to 10 reference images, choose 4 to 8 steps, pick from six size presets and get up to 4 results per call. ### Black Forest Labs: FLUX.2 Max - [black-forest-labs/flux-2-max/text-to-image](https://www.musespan.com/models/black-forest-labs/flux-2-max/text-to-image/llms.txt): Text To Image — Black Forest Labs FLUX.2 Max is the top FLUX.2 tier, built for the most realistic and precise images in the family, from product shots to campaign art. Choose from six size presets with JPEG or PNG output and a fixed seed for repeatable results. - [black-forest-labs/flux-2-max/image-to-image](https://www.musespan.com/models/black-forest-labs/flux-2-max/image-to-image/llms.txt): Image To Image — Black Forest Labs FLUX.2 Max delivers the family's most precise edits, preserving materials, lighting and identity while applying your changes. Combine up to 10 reference images, keep the input framing with auto sizing or pick a preset, and export JPEG or PNG. ### Black Forest Labs: FLUX.2 Pro - [black-forest-labs/flux-2-pro/text-to-image](https://www.musespan.com/models/black-forest-labs/flux-2-pro/text-to-image/llms.txt): Text To Image — Black Forest Labs FLUX.2 Pro is the studio-grade FLUX.2 model, built for photorealistic images with accurate lighting, materials and fine detail. Use it for product shots, editorial photos and campaign visuals, with six size presets from square to 16:9, JPEG or PNG output and repeatable seeds. - [black-forest-labs/flux-2-pro/image-to-image](https://www.musespan.com/models/black-forest-labs/flux-2-pro/image-to-image/llms.txt): Image To Image — Black Forest Labs FLUX.2 Pro edits photos from written instructions with studio-grade realism, keeping people, products and lighting consistent. Combine up to 10 reference images for retouching, product placement and scene changes, with auto sizing or six presets, JPEG or PNG output and repeatable seeds. ### Black Forest Labs: FLUX 1.1 Pro - [black-forest-labs/flux-1.1-pro/text-to-image](https://www.musespan.com/models/black-forest-labs/flux-1.1-pro/text-to-image/llms.txt): Text To Image — Black Forest Labs FLUX 1.1 Pro delivers polished, well-composed images with strong prompt adherence and refined artistic detail. Suited to illustrations, concept art and marketing visuals, it offers six size presets, optional prompt enhancement, JPEG or PNG output and up to 4 images per call. ### Black Forest Labs: FLUX 1.1 Pro Ultra - [black-forest-labs/flux-1.1-pro-ultra/text-to-image](https://www.musespan.com/models/black-forest-labs/flux-1.1-pro-ultra/text-to-image/llms.txt): Text To Image — Black Forest Labs FLUX 1.1 Pro Ultra generates images at up to 2K resolution with lifelike photographic detail. Turn on raw mode for a natural, less processed photo look, guide results with a reference image, and pick from 9 aspect ratios from 21:9 to 9:21, with up to 4 images per call. ### Black Forest Labs: FLUX.1 Dev - [black-forest-labs/flux-dev/text-to-image](https://www.musespan.com/models/black-forest-labs/flux-dev/text-to-image/llms.txt): Text To Image — Black Forest Labs FLUX.1 Dev is the open-weight FLUX model tuned for quality, with fine control over how closely images follow your prompt. Set guidance from 1 to 20 and steps up to 50 for detailed art, illustration and concept work, with optional acceleration, six size presets and up to 4 images per call. - [black-forest-labs/flux-dev/image-to-image](https://www.musespan.com/models/black-forest-labs/flux-dev/image-to-image/llms.txt): Image To Image — Black Forest Labs FLUX.1 Dev reimagines an input image from a text prompt, keeping its composition while changing mood, lighting or details. Set strength from 0.01 to 1 to control how far it drifts, tune guidance and steps, and get up to 4 results per call. ### Black Forest Labs: FLUX.1 Schnell - [black-forest-labs/flux-schnell/text-to-image](https://www.musespan.com/models/black-forest-labs/flux-schnell/text-to-image/llms.txt): Text To Image — Black Forest Labs FLUX.1 Schnell is the fastest FLUX model, producing usable images in 1 to 4 steps for rapid drafts and high-volume pipelines. Use it for thumbnails, previews and quick prompt iteration, with steps up to 12, optional acceleration, six size presets and up to 4 images per call. ### Google: Nano Banana 2.1 - [google/nano-banana-2.1/text-to-image](https://www.musespan.com/models/google/nano-banana-2.1/text-to-image/llms.txt): Text To Image — Google Nano Banana 2.1 is the newest Nano Banana Flash model, with better visual design and more natural-looking images than earlier versions. Strong for posters, ads and packaging with readable text, offering 1K to 4K output, three thinking levels, optional web search and 14 aspect ratios plus auto. - [google/nano-banana-2.1/image-to-image](https://www.musespan.com/models/google/nano-banana-2.1/image-to-image/llms.txt): Image To Image — Google Nano Banana 2.1 edits images from written instructions, keeping subjects consistent and results more natural than earlier Nano Banana models. Combine up to 10 reference images for retouching or product compositing, with 1K to 4K output, three thinking levels and optional web search. ### Google: Nano Banana 2 - [google/nano-banana-2/text-to-image](https://www.musespan.com/models/google/nano-banana-2/text-to-image/llms.txt): Text To Image — Google Nano Banana 2 (Gemini 3.1 Flash Image) is the everyday workhorse of the series, balancing image quality, text rendering and turnaround time. Draft at 0.5K and finish at 4K, frame banners or stories with 14 aspect ratios up to 8:1, and turn on web search or high thinking for harder prompts. - [google/nano-banana-2/image-to-image](https://www.musespan.com/models/google/nano-banana-2/image-to-image/llms.txt): Image To Image — Google Nano Banana 2 (Gemini 3.1 Flash Image) makes quick, instruction-based image edits for retouching, background swaps and product variants. Work from up to 10 reference images, preview at 0.5K before exporting at 4K, and add web search or high thinking for harder requests. ### Google: Nano Banana 2 Lite - [google/nano-banana-2-lite/text-to-image](https://www.musespan.com/models/google/nano-banana-2-lite/text-to-image/llms.txt): Text To Image — Google Nano Banana 2 Lite (Gemini 3.1 Flash-Lite Image) is the speed-first Nano Banana 2 model, made for generating many images at scale. Good for drafts, social assets and batch pipelines, with 1K output, 14 aspect ratios plus auto, a system prompt for a house style and up to 4 images per call. - [google/nano-banana-2-lite/image-to-image](https://www.musespan.com/models/google/nano-banana-2-lite/image-to-image/llms.txt): Image To Image — Google Nano Banana 2 Lite (Gemini 3.1 Flash-Lite Image) makes fast, lightweight image edits from text, suited to many small touch-ups in a row. Use up to 10 reference images for quick fixes, object removal or background swaps, with 1K output, optional thinking and up to 4 results per call. ### Google: Nano Banana - [google/nano-banana/text-to-image](https://www.musespan.com/models/google/nano-banana/text-to-image/llms.txt): Text To Image — Google Nano Banana (Gemini 2.5 Flash Image) is the original model of the series, turning plain-language prompts into clean, natural-looking images. A dependable pick for illustrations, social graphics and everyday visuals, with 10 aspect ratios, PNG, JPEG or WebP output and up to 4 images per call. - [google/nano-banana/image-to-image](https://www.musespan.com/models/google/nano-banana/image-to-image/llms.txt): Image To Image — Google Nano Banana (Gemini 2.5 Flash Image) edits photos from a written instruction while keeping faces, products and characters recognizable. Combine up to 10 reference images to change outfits, backgrounds or styles, with auto or 10 fixed aspect ratios and up to 4 results per call. ### Google: Nano Banana Pro - [google/nano-banana-pro/text-to-image](https://www.musespan.com/models/google/nano-banana-pro/text-to-image/llms.txt): Text To Image — Google Nano Banana Pro (Gemini 3 Pro Image) is the top tier of the series, built for images that need sharp, legible text and careful layout. Use it for infographics, posters, menus and product shots in 1K, 2K or 4K, with optional web search for current facts and up to 4 images per call. - [google/nano-banana-pro/image-to-image](https://www.musespan.com/models/google/nano-banana-pro/image-to-image/llms.txt): Image To Image — Google Nano Banana Pro (Gemini 3 Pro Image) handles demanding image edits, from relighting and restyling to rewriting the text inside a picture. Use up to 10 reference images to place products into scenes or keep a character consistent across shots, with 1K to 4K output and optional web search. ### ElevenLabs: Eleven v4 - [elevenlabs/eleven-v4](https://www.musespan.com/models/elevenlabs/eleven-v4/llms.txt): Text To Speech — ElevenLabs' most expressive text-to-speech model. Natural, emotionally rich speech in 90+ languages, with inline audio tags such as [excited], [whispers] or [laughs] to direct delivery, plus stability and similarity controls to shape the voice. Up to 5,000 characters per request. ### ElevenLabs: Eleven v4 Turbo - [elevenlabs/eleven-v4-turbo](https://www.musespan.com/models/elevenlabs/eleven-v4-turbo/llms.txt): Text To Speech — A faster, lower-cost version of Eleven v4 for interactive apps and high-volume narration. Keeps v4's expressive delivery, inline audio tags and 90+ languages, with stability and similarity controls. Up to 5,000 characters per request. ### ElevenLabs: Eleven v3 - [elevenlabs/eleven-v3](https://www.musespan.com/models/elevenlabs/eleven-v3/llms.txt): Text To Speech — ElevenLabs' expressive v3 text-to-speech model. Inline audio tags such as [excited] or [whispers] give fine control over emotion and delivery across 70+ languages. Up to 5,000 characters per request. ### Alibaba: Z-Image Base - [alibaba/z-image-base/text-to-image](https://www.musespan.com/models/alibaba/z-image-base/text-to-image/llms.txt): Text To Image — Alibaba's Z-Image Base is the standard, non-Turbo Z-Image model, using more steps and guidance for careful renders. Choose it when detail outweighs turnaround in portraits, illustration and magazine-style scenes, with negative prompts, guidance 1-20, up to 50 steps, custom sizes and seeds. ### Alibaba: Z-Image Turbo - [alibaba/z-image-turbo/text-to-image](https://www.musespan.com/models/alibaba/z-image-turbo/text-to-image/llms.txt): Text To Image — Alibaba's Z-Image Turbo is a compact 6B-parameter model that turns prompts into lifelike images in 8 steps or fewer. Suits drafts, social posts and high-volume pipelines, with preset or custom sizes, prompt expansion, acceleration levels, seeds, PNG/JPEG/WebP files and batches of 1-4. - [alibaba/z-image-turbo/image-to-image](https://www.musespan.com/models/alibaba/z-image-turbo/image-to-image/llms.txt): Image To Image — Alibaba's Z-Image Turbo reworks a reference image to follow a new prompt, using the original as a starting point. Good for restyling, variations and quick concept rounds, with a strength slider for how far results drift from the source, matched or custom sizes and prompt expansion. ### Alibaba: Z-Image Turbo ControlNet - [alibaba/z-image-turbo-controlnet/image-to-image](https://www.musespan.com/models/alibaba/z-image-turbo-controlnet/image-to-image/llms.txt): Image To Image — Alibaba's Z-Image Turbo ControlNet generates pictures that match the structure of an edge, depth or pose guide. Keeps layouts, silhouettes and poses intact for character art, product scenes and architecture, with canny, depth or pose preprocessing, control strength and start/end timing. ### Alibaba: Z-Image Turbo Inpaint - [alibaba/z-image-turbo-inpaint/image-to-image](https://www.musespan.com/models/alibaba/z-image-turbo-inpaint/image-to-image/llms.txt): Image To Image — Alibaba's Z-Image Turbo Inpaint repaints only the masked region of a photo from a prompt and leaves the rest intact. Handy for swapping objects, fixing flaws or changing outfits and backdrops, with adjustable inpaint strength, control scale and timing, output size and repeatable seeds. ### Alibaba: Z-Image Turbo Tiling - [alibaba/z-image-turbo-tiling/text-to-image](https://www.musespan.com/models/alibaba/z-image-turbo-tiling/text-to-image/llms.txt): Text To Image — Alibaba's Z-Image Turbo Tiling generates seamless, repeatable textures and patterns from a text description. Built for game materials, fabric prints, wallpapers and 3D surfaces, tiling horizontally, vertically or both, with adjustable tile size and stride and an optional source image or mask. ### Alibaba: Wan 2.7 - [alibaba/wan-2.7/text-to-image](https://www.musespan.com/models/alibaba/wan-2.7/text-to-image/llms.txt): Text To Image — Alibaba's Wan 2.7 turns text prompts into detailed, high-quality images with strong prompt understanding. A cost-efficient choice for concept art, social content and everyday visuals, with negative prompts, custom sizes, seeds, JPEG/PNG/WebP output and up to 5 images per request. - [alibaba/wan-2.7/image-to-image](https://www.musespan.com/models/alibaba/wan-2.7/image-to-image/llms.txt): Image To Image — Alibaba's Wan 2.7 edits and transforms images with plain-text instructions, using up to 4 reference images. Refer to them as image 1 to image 4 to restyle, combine or retouch, with optional prompt expansion, negative prompts, custom sizes, seeds and up to 4 results per request. ### Alibaba: Wan 2.7 Pro - [alibaba/wan-2.7-pro/text-to-image](https://www.musespan.com/models/alibaba/wan-2.7-pro/text-to-image/llms.txt): Text To Image — Alibaba's Wan 2.7 Pro turns text prompts into premium, richly detailed images. Crisp textures and accurate lighting suit product shots, marketing visuals and editorial art, with negative prompts, custom sizes, seeds and up to 5 images per request. - [alibaba/wan-2.7-pro/image-to-image](https://www.musespan.com/models/alibaba/wan-2.7-pro/image-to-image/llms.txt): Image To Image — Alibaba's Wan 2.7 Pro makes precise, professional-grade image edits from text instructions. Combine up to 4 reference images (image 1 to image 4 in your prompt) for product retouching, material and style changes or scene swaps, with negative prompts, custom sizes, seeds and up to 4 results. ### Alibaba: Wan 2.5 Preview - [alibaba/wan-2.5-preview/text-to-image](https://www.musespan.com/models/alibaba/wan-2.5-preview/text-to-image/llms.txt): Text To Image — Alibaba's Wan 2.5 Preview turns text prompts into polished images and can expand short prompts automatically. Choose a preset or custom size, add a negative prompt, fix a seed for repeatable results and get up to 4 images per request. - [alibaba/wan-2.5-preview/image-to-image](https://www.musespan.com/models/alibaba/wan-2.5-preview/image-to-image/llms.txt): Image To Image — Alibaba's Wan 2.5 Preview edits an image from a text instruction, or blends two reference images into a new one. Restyle scenes, change materials or swap backgrounds with instructions up to 2,000 characters, a negative prompt, sizes from 384 to 1440 px, seeds and up to 4 results per request. ### Wan 2.2 5B - [alibaba/wan-2.2-5b/text-to-image](https://www.musespan.com/models/alibaba/wan-2.2-5b/text-to-image/llms.txt): Text To Image — Alibaba's Wan 2.2 5B, the lighter Wan 2.2 model, produces detailed photorealistic images that follow the prompt. Fits realistic scenes, product mockups and nature shots, with prompt expansion, negative prompts, preset or freeform sizes, guidance, steps, seeds and PNG or JPEG files. ### Wan 2.2 A14B - [alibaba/wan-2.2-a14b/text-to-image](https://www.musespan.com/models/alibaba/wan-2.2-a14b/text-to-image/llms.txt): Text To Image — Alibaba's Wan 2.2 A14B renders prompts as photorealistic images in two expert passes, for layout and then detail. Built for cinematic stills, product visuals and rich environments, with per-stage guidance, preset or pixel sizes, negative prompts, seeds, steps, shift and prompt expansion. ### Alibaba: Qwen Image - [alibaba/qwen-image/text-to-image](https://www.musespan.com/models/alibaba/qwen-image/text-to-image/llms.txt): Text To Image — Alibaba's Qwen-Image creates images from text prompts and stands out at rendering complex, multi-line text. Aimed at posters, slides and signage, with negative prompts, size presets or exact dimensions, guidance and steps, seeds, up to 3 LoRAs, turbo mode and 1-4 images per call. ## 厂商 / Vendors - [Alibaba](https://www.musespan.com/models/alibaba): Alibaba Wan, Qwen-Image and Z-Image via API: text-to-image, image editing, inpainting and ControlNet-guided image generation models. - [Black Forest Labs](https://www.musespan.com/models/black-forest-labs): Black Forest Labs FLUX.1, FLUX 1.1 Pro, Kontext, FLUX.2 Pro and FLUX 3 via API: text-to-image and instruction-based image editing models. - [ByteDance](https://www.musespan.com/models/bytedance): ByteDance Seedream 5.0 Flash on MuseSpan: text-to-image generation and reference-guided image editing through one API. - [ElevenLabs](https://www.musespan.com/models/elevenlabs): ElevenLabs Eleven v3, Eleven v4 and Eleven v4 Turbo via API: text-to-speech with inline audio tags, 90+ languages in v4 and a low-latency Turbo variant. - [Google](https://www.musespan.com/models/google): Google Nano Banana, Nano Banana 2, 2 Lite, 2.1 and Pro via API: Gemini-based text-to-image and image editing models from Google DeepMind. - [OpenAI](https://www.musespan.com/models/openai) ## 接口 / API 数据面路径相对于 `https://api.musespan.com`,需要 `Authorization: Bearer `;目录接口使用下方完整网址：`/v1/models` 无需密钥即可读取公开目录，携带 Bearer API key 时返回相同的目录，只校验密钥是否有效；`/v1/videos/models` 必须携带密钥。 - `GET https://www.musespan.com/v1/videos/models` — 旧版视频兼容目录，使用组级 ID。新接入统一使用网站 /v1/models 条目的 `id` 作为 `model`，并读取对应 Schema。必须携带 Bearer API key，列出全部已上架的视频模型。Legacy video compatibility catalog with group-level IDs; new integrations must send an `id` from the website /v1/models catalog as `model` and read its Schema. Requires a Bearer API key and lists every listed video model. (https://www.musespan.com/docs/quickstart.md) - `POST /v1/videos` — 提交视频生成任务(统一接口)。Body 使用 `{ model, prompt, resolution, aspect_ratio, seconds }` 或 `{ model, prompt, size, seconds }`;两种尺寸表示不能同时出现。立即返回任务 `id`,不等生成完成。Submit an async text-to-video / image-to-video job. (https://www.musespan.com/docs/quickstart.md) - `POST /v1/images/generations` — 同步生成或编辑图片。请求的 `model` 使用 `/v1/models` 条目的 `id`，并且只传该能力 schema 声明的字段。Generate or edit images synchronously; `model` is an `id` from /v1/models, with only the fields its schema declares. (https://www.musespan.com/docs/quickstart.md) - `POST /v1/images/tasks` — 提交异步图片任务，带唯一 X-Request-Id，任务 ID 就是它；提交幂等，超时或 429/5xx 时用同一 ID 原样重发，不会重复扣费。Submit an asynchronous image task only when the selected Schema declares this endpoint; it is idempotent by X-Request-Id. (https://www.musespan.com/docs/quickstart.md) - `GET /v1/images/tasks/{id}` — 使用提交时的同一把密钥每 5 秒查询异步图片任务；completed 时读取全部 data[].url；返回 404 task_not_found 表示提交没有送达，用同一 ID 重发。Poll an image task; a 404 means the POST never arrived, so resend it. (https://www.musespan.com/docs/quickstart.md) - `POST /v1/model-tasks` — 提交模型任务：请求体 {"model": `/v1/models` 条目的 `id`, "input": {...}}，必须带 X-Request-Id，任务 ID 就是它；提交幂等，重发时请求体字节必须完全一致。返回 404 `model_not_found` 表示这个 ID 不能调用：不要原样重发，改用响应里的 `suggested_models`。Submit a model task with an `id` from /v1/models as `model`; X-Request-Id is required and idempotent (resend byte-identical bodies). A 404 model_not_found means the ID is not callable: do not resend it, use suggested_models. (https://www.musespan.com/docs/quickstart.md) - `GET /v1/model-tasks/{id}` — 使用同一把密钥每 5 秒查询模型任务；只有 completed 含输出；排队中的任务为 queued 并带估计的 queue_position；返回 404 task_not_found 表示提交没有送达，用同一 ID 重发。Poll a model task; only completed includes output; a queued task may carry its estimated queue_position. (https://www.musespan.com/docs/quickstart.md) - `POST /v1/model-tasks/{id}/cancel` — 取消仍在排队（`queued` 且带 `queue_position`）的模型任务，任务以 `cancelled` 失败并全额退款；已不在排队中（正在发出、已发出或已结束）的任务不能取消（409 `task_not_cancellable`）。Cancel a model task still waiting in the queue; it fails as cancelled and is fully refunded. A task no longer waiting in the queue (being sent, sent, or ended) cannot be cancelled (409). (https://www.musespan.com/docs/errors.md) - `GET /v1/model-tasks/{id}/artifacts/{index}` — 用提交任务的同一把密钥下载模型任务声明的结果文件（图片、文字转语音的音频等），以 application/octet-stream 返回；音频按文件头识别格式（ID3 或 MPEG 帧头为 MP3，RIFF…WAVE 为 WAV，OggS 为 Ogg/Opus，fLaC 为 FLAC）。Download a declared model-task result file (images, text-to-speech audio) with the submitting key; bytes arrive as application/octet-stream, so identify audio formats from the leading bytes. (https://www.musespan.com/docs/quickstart.md) - `GET /v1/videos/{id}` — 轮询任务状态。`status` ∈ `queued` | `in_progress` | `completed` | `failed` | `cancelled`;必须用提交时的同一把密钥。Poll every 5s. (https://www.musespan.com/docs/errors.md) - `GET /v1/videos/{id}/content` — 任务完成后下载 MP4 字节流。必须使用提交任务时的同一把密钥，并及时转存。Download the completed video bytes. (https://www.musespan.com/docs/quickstart.md) - `GET https://www.musespan.com/v1/models` — 返回已上架模型、独立能力和公开价格。添加 `?expand=input_schema` 返回图片接口能力的参数约束和示例输入；模型任务（`/v1/model-tasks`）能力的 `input` 参数请读取该能力的 llms.txt 或 OpenAPI。无需密钥时返回公开目录；携带 Bearer API key 时返回相同的目录，只校验密钥是否有效。每条的 `id` 就是请求的 `model`：图片和视频的 `id` 以能力结尾（如 `alibaba/wan-2.7/image-to-image`），音频模型的 `id` 通常就是模型 ID（`厂商/模型` 两段，这时与 `group.id` 相同）；`group.id` 与 `id` 不同时只用于分组，不能填进 `model`。查询参数：`category=<类型>[,<类型>]`（可单独使用，也可配合分页）、`limit=1..200` 加 `after=`、`expand=input_schema`；`category` 或 `expand` 取值未知时返回 400 `invalid_parameter`，`error.allowed_values` 列出可用值，`limit`/`after` 无效时返回 400 `invalid_pagination`。Returns the live public catalog; the `id` of each entry is the request `model` (image and video IDs end with the capability, the `id` of an audio model is normally the model ID; a `group.id` that differs from `id` is not callable). `?expand=input_schema` adds parameter contracts and example inputs for image endpoints, while model tasks document their `input` in their llms.txt and OpenAPI. Query: `category=[,]` (with or without pagination), `limit=1..200` + `after=`, `expand=input_schema`; an unknown `category` or `expand` value returns 400 `invalid_parameter` with `error.allowed_values`, a bad `limit`/`after` returns 400 `invalid_pagination`. (https://www.musespan.com/docs/quickstart.md) ## 更多 / More - [llms-full.txt](https://www.musespan.com/llms-full.txt): 全部文档正文和已上架能力说明。内容较长，优先按需读取单能力文档；价格和可用状态以调用前的实时目录为准。 - [agent skill](https://www.musespan.com/.well-known/agent-skills/musespan/skill.md): 可直接装载的 SKILL.md(索引在 https://www.musespan.com/.well-known/agent-skills/index.json)。