# Fashion Diffusion

> AI fashion design and commercial photography platform for brands and creators.
> The Fashion Diffusion Public API is a bearer-key REST API for asynchronous
> generative fashion imagery tasks: virtual try-on and AI inpainting today, with a
> shared task-query surface. Base URL: https://www.fashiondiffusion.ai/api/public/v1/
>
> Generated by API Evangelist from the provider's published surfaces on 2026-08-31.
> Fashion Diffusion does not publish an llms.txt of its own (/llms.txt returns 404).

## Getting started

- [API landing page](https://www.fashiondiffusion.ai/api-access): capability list, latency expectations, FAQ, and the contact-sales entry point for API access.
- [Public API reference](https://www.fashiondiffusion.ai/playground/api/docs): the complete public contract — authentication, both submit endpoints, task query and list, parameters, defaults and the error envelope. Readable anonymously; marked noindex.
- [Developer API area](https://www.fashiondiffusion.ai/playground/api): where API keys are created. Requires sign-in.
- [Pricing](https://www.fashiondiffusion.ai/pricing): four subscription tiers, all of which include Developer API access.

## Authentication

Send the full API key secret as a bearer credential on every call:

    Authorization: Bearer fd_live_xxx

The secret is displayed only once at creation. There is no OAuth, no OpenID Connect
and no scope surface. An API key can read only tasks owned by its user and current
team. An invalid or missing key returns HTTP 401 with
`{"error":{"code":"INVALID_API_KEY","message":"Invalid API key","requestId":"<uuid>"}}`.

## Operations

All operations are relative to `https://www.fashiondiffusion.ai/api/public/v1`.

- `POST /virtual-try-on/tasks` — submit a garment image and a model image. Returns 202 with the task id.
  Parameters: `garmentImageUrl` or `garmentImageBase64`, `modelImageUrl` or `modelImageBase64`,
  `imageCount` (1, 2, 4; default 1), `resolution` (2K, 4K; default 2K),
  `ratio` (auto, 1:1, 3:4, 4:3, 9:16, 16:9; default auto), `mode` (FAST, REFINE; default FAST),
  `model` (standard, pro; default standard), `prompt`.
- `POST /ai-inpainting/tasks` — replace or repair a selected area of an image from a text prompt.
  Returns 202 with the same task object. Parameters: `originalImageUrl` or `imageBase64`, `prompt`,
  `model` (standard, pro), `resolution` (2K, 4K), `ratio`, optional `referenceImages`, optional mask image.
- `GET /tasks/{taskId}` — read one task. Statuses: QUEUED, RUNNING, SUCCEEDED, FAILED. Output image URLs are in `outputs`.
- `GET /tasks` — list tasks submitted through the Public API by the current user.
  Query: `limit` (1-100, default 20), `cursor` (from the previous response's `nextCursor`), `createdAfter` (ISO 8601).

Each image may be a public HTTPS URL or a Base64 data URL. JPEG, PNG and WebP only,
20 MB maximum, exactly one input form per image field.

## Conventions

- Execution model: submit-and-poll. Submission returns 202; completion is discovered only by polling `GET /tasks/{taskId}`.
- Tracing: every response carries an `X-Request-Id` header; the same id appears as `error.requestId`.
- Errors: proprietary JSON envelope `{"error":{"code","message","requestId"}}`, not RFC 9457 problem+json.
- Pagination: cursor-based on `GET /tasks` only.
- Versioning: URI path, `v1`. No version policy or deprecation policy is published.

## What this API does NOT have

An agent planning against this API should know these absences, all of which are
stated or verified rather than assumed:

- **No idempotency.** The reference states verbatim: "This version has no idempotency
  protection, so do not automatically retry a timed-out submission unless duplicate
  charges are acceptable." Each accepted request creates and charges a new task.
- **No reversal.** There is no cancel, abort, void, delete or refund operation for a
  submitted task, and no window is stated. A submission cannot be taken back.
- **No dry-run.** There is no validate-only or estimate mode; every accepted call bills.
- **No rate-limit signal.** No RateLimit-*, X-RateLimit-* or Retry-After header is
  published and no 429 behaviour is documented. Throughput is bounded by the plan's
  monthly credit allowance, not by a published request rate.
- **No webhooks or events.** No callback URL, no event stream, no AsyncAPI.
- **No machine-readable contract.** No OpenAPI, Swagger, GraphQL schema or Postman
  collection is served; `/openapi.json`, `/swagger.json` and `/api-docs` all return 404.
- **No MCP server** and **no A2A agent card**. `/.well-known/agent-card.json`,
  `/.well-known/agent.json` and every other named well-known path return 404.
- **No SDKs.** npm and PyPI carry no first-party package, and the company's GitHub
  account has zero public repositories.
- **No status page and no changelog.**
- **Marketed surface exceeds documented surface.** The API landing page advertises
  "more than 18 purpose-built fashion AI endpoints"; four operations are publicly
  documented. The marketing snippet also calls `https://api.fashiondiffusion.ai/v1/...`,
  a host that does not resolve — the working base is `www.fashiondiffusion.ai/api/public/v1/`.

## Plans

Four self-serve monthly tiers, each including Developer API access, metered in image
credits: Starter $12 (100 credits), Creator $29 (320), Pro $59 (760), Studio $199 (2800).
Yearly billing gives two months free. There is no free tier. Enterprise API volume is a
contact-sales conversation with no published price or SLA.

## Compliance and data handling

- GDPR rights and Standard Contractual Clauses for US transfers; CCPA rights for California residents.
- Payments processed by Stripe and Apple under PCI-DSS; Fashion Diffusion holds no PCI attestation of its own.
- Hosted on AWS in the United States; TLS in transit, AES-256 at rest.
- Retention: account data for the life of the account plus 30 days; uploaded images until deleted; billing records 7 years.
- Uploaded images are not used to train Fashion Diffusion models without explicit consent.
- No SOC 2, ISO 27001 or other independently audited certification is published, and there is no trust center or security.txt.

## Company

- [Home](https://www.fashiondiffusion.ai/)
- [About](https://www.fashiondiffusion.ai/about-us)
- [Blog](https://www.fashiondiffusion.ai/blog)
- [Terms of Service](https://www.fashiondiffusion.ai/terms-of-service)
- [Privacy Policy](https://www.fashiondiffusion.ai/privacy-policy)
- [GitHub](https://github.com/FashionDiffusion) (no public repositories)
- Support: support@fashiondiffusion.ai