One payload
One request.Every model.
One body for every image model. You send a ratio, and Mynth maps it to a size that model serves.
Size
- prompt
- A lighthouse at dusk, film grain
- size
- 21:9
Models
Seedream 5.0 Lite and FLUX.2 Pro have a 21:9 preset. GPT Image 2 does not, so the request keeps the landscape and lands on 16:9.
The body
Same fields on every image model
Pass a model id and a prompt. Every other field is optional, and every model takes the same names.
- model
- A catalog id,
vendor/name. Pass one on every request.autois what you get when the field is omitted, and it is experimental. - prompt
- One string, 1 to 8192 characters, passed through as the positive prompt.
- size
- A ratio such as
16:9, a preset such aslandscape, orauto. You ask for a shape. The model serves pixels. Omit it and Mynth picks a ratio from the prompt. - inputs
- Up to 20 images, to edit or to reference. A pinned model that cannot take them rejects the task at creation with
VALIDATION_ERROR. - negative_prompt
- Up to 8192 characters. Sent when the model has a negative prompt. Dropped, with no error, when it does not.
count, magic_prompt, rating, destination, output, webhook, and metadata are on this body too. Generate images is the full list.
Video is a second body, on video.generate. It takes resolution and duration. It has no size field. You still switch models by changing the model id.
Outside the body
Provider settings stay with the provider
The frames above are how a size is mapped. A model serves its own preset. A ratio it lacks snaps to the closest one that keeps the orientation.
- 4k sizes
- A
4ksize needs a model with 4k presets and a 4k price. Otherwise the task fails withCAPABILITY_NOT_SUPPORTED. - Provider settings
- No request field for steps, guidance, a scheduler, LoRAs, or a provider style flag. Those stay on the provider's own API.
- Unknown fields
- A field the schema does not know is dropped, with no error. The task's
requestis the body that was accepted.
Pass-through mode is on the roadmap. It would send the request to your own provider key, with that provider's options, and it is not in the API yet. If you would use it, say so on Discord (opens in a new tab).