wanna.sh API

The app itself talks to wanna.sh through /api. You can use it too, for things the OpenAI API doesn’t cover: finding and downloading models, starting and stopping them, music, and settings. It’s simpler than the OpenAI API but less stable: it may change between versions. The full list, with request and response shapes, is at http://localhost:7860/docs.

The same API key rules apply.

Generate anything

POST /api/generate/{task} starts a job for image, video, speech, sound (music) or transcribe, and returns straight away. Poll GET /api/jobs/{id} until status is completed or failed.

curl http://localhost:7860/api/generate/sound \
  -H "Content-Type: application/json" \
  -d '{"model": "auto", "prompt": "calm lo-fi beat with soft piano", "duration": 10}'

{ "id": "...", "task": "sound", "status": "queued", "progress": 0, "result": null, ... }

When it’s done, result.files lists the output files, which you download from /files/{name}:

curl -O http://localhost:7860/files/<name>

For transcription, result holds text, timed segments and duration instead.

Task	Main fields
image	prompt, negative_prompt, width, height, n, quality (fast, balanced, best), steps, guidance, seed, sampler, image_upload and strength to start from a picture
video	prompt, negative_prompt, width, height, num_frames, quality, seed, image_upload
speech	text, voice, speed, format (wav, mp3, flac, ogg)
sound	prompt, duration in seconds
transcribe	file_upload (required), language, translate

Files go in by id, never by path: upload them first with POST /api/uploads (multipart field file), which returns {"id": ..., "filename": ...}, then pass that id as image_upload or file_upload.

Models

Endpoint	What it does
GET /api/models	Installed models, with whether each is loaded and where.
POST /api/models/{id}/load	Start a model. Optional body: {"gpus": [0, 1], "context": 16384, "offload": false}.
POST /api/models/{id}/unload	Stop it and free its memory.
DELETE /api/models/{id}	Delete it from disk.
GET /api/models/{id}/memory	For chat models: memory options, how each would fit, and the recommended one.
GET /api/models/{id}/log	The end of the model’s engine log, for troubleshooting.

Finding and downloading

Endpoint	What it does
GET /api/hub/overview?q=	The best few models per task for this computer.
GET /api/hub/browse?task=chat&q=&sort=popular&fits=true	Runnable models for one task, rated for this computer. sort is popular, trending, newest or liked.
GET /api/hub/model/{repo_id}	One model’s details: versions, sizes and ratings.
POST /api/downloads	Start a download: {"repo_id": "...", "variant_id": "..."}. Leave out variant_id for the recommended version.
GET /api/downloads	Download progress.
DELETE /api/downloads/{id}	Cancel one.

System and settings

Endpoint	What it does
GET /api/system	Graphics cards, memory, disk, engines and loaded models.
GET /api/system/check	The system check from Settings.
GET /api/settings, PUT /api/settings	Read or change settings. Secrets come back masked.
GET /api/keys, POST /api/keys, DELETE /api/keys/{id}	Manage API keys. POST takes {"name": "..."} and returns the key once.
GET /api/gallery, DELETE /api/gallery/{id}	Everything you’ve made.