Choosing a model

Set the model

Models are addressed as owner/name, exactly as in the Replicate URL:

imgen "A cat in space" --model stability-ai/sdxl --out cat.png

Without --model, imgen uses black-forest-labs/flux-1.1-pro. In a prompt file, set model in defaults or on a single job; see Batch mode.

imgen always runs the model’s latest published version. It looks the version up once per model and run, so a batch with ten jobs on the same model makes a single lookup.

Which models work

Any Replicate model that accepts a prompt input and returns an image URL (or a list whose first entry is an image URL). The size and format inputs are detected per model, see Automatic schema detection.

The README lists these as tested starting points:

Model	Notes
black-forest-labs/flux-1.1-pro	Default. Accepts exact pixel sizes, delivers WebP server-side.
black-forest-labs/flux-1.1-pro-ultra	Highest resolution of the FLUX family.
black-forest-labs/flux-schnell	Fast and cheap, good for drafts.
black-forest-labs/flux-dev	Open-source FLUX variant.
ideogram-ai/ideogram-v3-quality	The best choice when the image must contain legible text.
stability-ai/sdxl	Stable Diffusion XL.
stability-ai/stable-diffusion-3	Stable Diffusion 3, moderate text rendering.

Prices and speed are set by Replicate and change over time; check the model page before large batches.

When a model fails to resolve

A typo in the name, a private model or a model without a published version stops the run before any image is generated. The error names the model and repeats the API response, for example Failed to resolve model owner/name: 404 Not Found ….