mattt
commented
3 weeks ago Hi @Wheest. That information is available in the schema of the model version you're running.
>>> import replicate
>>> v = replicate.models.get("stability-ai/stable-diffusion-3").latest_version
>>> list(v.openapi_schema["components"]["schemas"]["Input"]["properties"].keys())
['seed', 'image', 'prompt', 'num_outputs', 'aspect_ratio', 'output_format', 'guidance_scale', 'output_quality', 'negative_prompt', 'prompt_strength', 'disable_safety_checker']
Wheest
Different models have different prediction parameters (e.g., some look for an image, others look for text input).
It would be useful if the API had an endpoint we could probe to get the input a particular model expects, without running inference.
We could then use this to create the PredictionParams for a given model.