seedanceapi
Cosmos 3 Super · Image to Video render example

NVIDIA · Cosmos 3 Super · image to video

Cosmos 3 Super · Image to Video

Render Cosmos 3 Super through the same job workflow as Seedance — from $0.068 per render.

Model idnvidia/cosmos-3-super/image-to-videoShot spec16:9 · 4:3 · 1:1 · 3:4 · 9:16 · 7sRatefrom $0.068

What it renders

NVIDIA Cosmos 3 Super Image to Video is a fast AI image-to-video generation model that creates high-quality videos from a first-frame image and a motion prompt. Hand over the model id and your shot settings, keep the job id, and collect the finished clip once the job lands. Pricing is per completed render from $0.068 and is locked in before the first frame is queued.

  1. Motion-first defaults

    Cosmos 3 Super · Image to Video opens on workable shot framing before you tune anything.

  2. Async by design

    Renders run as background jobs; poll at your own pace instead of holding a connection open.

  3. Clip-ready output

    Results arrive as a downloadable asset from private storage with a short-lived link.

  4. Image to Video shots

    Image to Video stays available for Cosmos 3 Super · Image to Video without leaving the video workspace.

  5. Cosmos 3 Super shots

    Cosmos 3 Super stays available for Cosmos 3 Super · Image to Video without leaving the video workspace.

  6. Video shots

    Video stays available for Cosmos 3 Super · Image to Video without leaving the video workspace.

From the shot board

Cosmos 3 Super · Image to Video example output.
Cosmos 3 Super · Image to Video output

Wire the render pipeline

  1. Grab a key

    Sign in, create a server-side key and add a little balance.

  2. Confirm the endpoint

    GET /v1/models lists nvidia/cosmos-3-super/image-to-video with its shot limits and current rate.

  3. Queue the render

    POST /v1/jobs with the model id, your shot settings and a unique Idempotency-Key.

  4. Collect the clip

    Poll GET /v1/jobs/{id} until it lands, then fetch GET /v1/jobs/{id}/content.

Queue the render

# Seedance API — queue one render
curl https://seedanceapi.pro/v1/jobs \
  -H "Authorization: Bearer $SEEDANCE_API_KEY" \
  -H "Idempotency-Key: $REQUEST_ID" \
  -H "Content-Type: application/json" \
  -d '{"model":"nvidia/cosmos-3-super/image-to-video","duration":7,"guidance_scale":6,"num_inference_steps":28,"prompt":"A slow dolly across a rain-lit neon street, shallow depth of field.","size":"16:9"}'

Shot settings

durationoptional
integer · 1 / 2 / 3 / 4 / 5 / 6 / 7
guidance_scaleoptional
number · default 6
imageneeded
string
negative_promptoptional
string
num_inference_stepsoptional
integer · default 28
promptoptional
string
sizeoptional
string · 16:9 / 4:3 / 1:1 / 3:4 / 9:16

Try a render from here

This queues a real job on this site's /v1/jobs using your signed-in balance, or an API key if you are not signed in. A pasted key lives only in this tab; the render is billed at the quoted USD rate.

Or sign in to use your account balance.

Before you wire it in

How do I start a Cosmos 3 Super · Image to Video render?

Authenticate, check the live model list for nvidia/cosmos-3-super/image-to-video, then POST the model id and shot settings to /v1/jobs with an Idempotency-Key. Poll the job id until it settles, then download the clip.

What can I control on Cosmos 3 Super · Image to Video?

Cosmos 3 Super · Image to Video accepts duration, guidance_scale, image, negative_prompt, num_inference_steps, prompt, size. Unknown keys are rejected before the render is queued, so a typo never burns a job.

What does a Cosmos 3 Super · Image to Video render cost?

Pricing is per render and starts at $0.068. The exact figure for your settings is returned by the model list and reserved before the job is accepted.

Is the same Seedance key valid for Cosmos 3 Super · Image to Video?

Yes — Cosmos 3 Super · Image to Video runs on the same account, key and wallet as Seedance. Each endpoint is enabled individually, so verify the route before you depend on it.

Why does my Cosmos 3 Super · Image to Video job show as queued?

Video rendering is asynchronous: queued and running are normal. Keep polling until the job succeeds, fails or asks for review, and never resubmit the same Idempotency-Key expecting a second render.

More endpoints like this