Cosmos 3 Super Image to Video

Cosmos 3 Super Image to Video · NVIDIA — Image to video · 1–7 s. Call it through one queue API and pay for the options you pick.

modellane/cosmos-3-super-image-to-video · Image to video · NVIDIA

Image to video model: 1–7 s.

Prompt prompt — required, at most 2500 characters Describe what you want produced as concretely as you can.
Source image image_url — required The frame the generation starts from.
Negative prompt negative_prompt — optional Things you don't want to see in the output.
Image size image_size — optional The pixel size of the output image.
Duration duration — optional, default: 5, options: 1, 2, 3, 4, 5, 6, 7 A longer duration raises the price directly.
Steps num_inference_steps — optional, default: 28, range: 1–50 More steps usually give a cleaner result and take longer.
Prompt adherence guidance_scale — optional, default: 6, range: 1–20 A high value stays close to the prompt, a low value leaves the model more room.
Seed seed — optional The same seed and the same settings produce a similar result; leave it empty and one is picked at random.
Safety checker enable_safety_checker — optional, default: No When on, the output is passed through a safety filter.

Output and retention

The generated file and your input are kept for seven days and then deleted.

Submit to the queue: POST /v1/queue/modellane/cosmos-3-super-image-to-video

Status, result, cancellation and error codes work the same for every model; see the API docs.

Everything on this page is public; you need an account to run the model.