Models

Every model is called the same way: POST /v1/tasks with the model’s id and an input object that matches its schema. Each model’s page lists its inputs, limits and defaults, with an example request.

Model	Tasks	Runs on	Per image	Per reference image	1,000 images
FLUX 3 Image Klein flux-3-klein	Text to image, Image editing	Our GPUs	$0.01	+$0.004	$10.00
Z-Image Turbo z-image-turbo	Text to image	Our GPUs	$0.006	—	$6.00
Qwen Image 3.0 qwen-image-3.0	Text to image, Image editing	Publisher's API	$0.035	+$0.004	$35.00
Qwen Image 3.0 Pro qwen-image-3.0-pro	Text to image, Image editing	Publisher's API	$0.045	+$0.004	$45.00
Seedream 5.0 Pro seedream-5.0-pro	Text to image, Image editing	Publisher's API	$0.05	+$0.004	$50.00

Prices are per output image. Edits also pay the per-reference-image price for each reference image, for every image they produce.

Models that run on our GPUs are uncensored, and every image they make passes our output safety check. Models on a publisher’s API are resold: they keep the publisher’s moderation, which refuses adult content. Both kinds take the same requests and are billed the same way.

Read models from the API

GET /v1/models lists every model with its pricing and its input_schema (a JSON Schema), and GET /v1/models/{id} returns one. Neither needs an API key.

curl https://api.velvetapi.dev/v1/models

Price a request

POST /v1/quotes returns the exact price of a request without running it or needing a key:

curl https://api.velvetapi.dev/v1/quotes \
-H "Content-Type: application/json" \
-d '{"model": "flux-3-klein", "input": {"prompt": "...", "num_images": 2}}'

{
  "model": "flux-3-klein",
  "price_usd": 0.02
}

Invalid input gets the same 400 invalid_input error that creating the task would.