Skip to content

Cosmos 3 Super Image to Video

Integrate

Model id nvidia/cosmos-3-super/image-to-video
# Run the model; wait up to 20 s for the finished result in the same call
curl -X POST "https://api.doitong.ru/v1/run/nvidia/cosmos-3-super/image-to-video?wait=20" \
  -H "Authorization: Bearer $DOITONG_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "A red fox trots through fresh snow at sunrise, soft golden light, shallow depth of field",
    "image": "https://content.doitong.com/images/showcase/v3/gen_fox_cartoon.jpg"
  }'

# Still "queued" or "running"? Poll with the id from the response:
curl "https://api.doitong.ru/v1/requests/REQUEST_ID?wait=20" \
  -H "Authorization: Bearer $DOITONG_API_KEY"

Files like the example: $0.40 400 API credits · ≈ ₽42.93 your files are measured before the run Run it on this page

Get API key
NVIDIA Image to Video Lipsync #stylized#transform

Animates a first-frame image into a video, with optional prompt expansion and an agentic refinement loop

NVIDIA · Video

Your first result goes here

Build a request below and run it with your API key. Outputs are copied to our storage, kept forever or 7 days for free.

Build a request

Controls are generated from the model schema. The code and the price update as you change them.

•required 88 / 4096

Video-generation instructions in text: specify what the scene contains and the motion it should show.

•required image_url
Drop a file, or

Image URL used to condition the video's first frame.

image_size default {"height":480,"width"...

Video dimensions requested before automatic adjustment. To obtain a supported size, the system clamps the request and chooses the nearest NVIDIA tier—256p, 480p, or 720p—and supported aspect ratio.

default 2

Upper limit on prompt stages with agentic generation enabled.

default 2

Per-iteration video candidate count for the agentic loop, determining how many videos compete in evaluation before the best proceeds to prompt rewriting.

default 189

Frame count for generation; increasing it produces a longer video.

default 24

Output video frame rate, measured in frames per second.

Advanced negative_prompt · num_inference_steps · guidance_scale · seed · enable_prompt_expansion · agentic_early_stop · enable_safety_checker
0 / 2048

Negative prompt for discouraging content such as unwanted motion or artifacts. Omitting it uses NVIDIA's recommended negative prompt for i2v, while an empty string disables negative guidance.

default 28

Denoising step count, trading longer processing for better quality as the count increases.

default 6

Strength of classifier-free guidance.

Seed controlling repeatability: an identical prompt and seed produce an identical video when the model version is unchanged.

POST /run/nvidia/cosmos-3-super/image-to-video?wait=20 quote → $0.40
# Run the model; wait up to 20 s for the finished result in the same call
curl -X POST "https://api.doitong.ru/v1/run/nvidia/cosmos-3-super/image-to-video?wait=20" \
  -H "Authorization: Bearer $DOITONG_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "A red fox trots through fresh snow at sunrise, soft golden light, shallow depth of field",
    "image": "https://content.doitong.com/images/showcase/v3/gen_fox_cartoon.jpg"
  }'

# Still "queued" or "running"? Poll with the id from the response:
curl "https://api.doitong.ru/v1/requests/REQUEST_ID?wait=20" \
  -H "Authorization: Bearer $DOITONG_API_KEY"

Run this request

Your key stays in this browser: it is sent only to the Doitong API and saved (localStorage) only if you tick Remember. Get an API key

About Cosmos 3 Super Image to Video

Uses an `image_url` as the opening frame and a `prompt` up to 4096 characters. `num_frames` goes up to 189 and `frames_per_second` from 4 to 60. Prompt expansion is on by default, and `enable_agentic_generation` runs iterative candidates with critique. Also offers `negative_prompt`, `guidance_scale` and `num_inference_steps` up to 50.

Parameters

Send either the canonical name or the native name, not both. Canonical names

NameType DefaultAllowed Description
promptrequired
Typestring
≤ 4,096 characters
Video-generation instructions in text: specify what the scene contains and the motion it should show.
image_urlrequired image
Typeurl · image
Image URL used to condition the video's first frame.
negative_promptoptional
Typestring
DefaultThe video captures a series of frames showing macroblocking artifacts, chromatic aberration, high-frequency noise, and rolling shutter distortion. It includes static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, bit-depth compression artifacts, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, hard cut, visual noise, and flickering. It features moiré patterns, edge halos, and temporal aliasing. Furthermore, the content defies common sense, generating illogical scenarios, nonsensical entities, absurd character behaviors, and conceptual paradoxes that violate basic human reasoning and everyday reality. The video looks like a surreal or glitchy hallucination. Overall, the video is of poor quality.
≤ 2,048 characters
Negative prompt for discouraging content such as unwanted motion or artifacts. Omitting it uses NVIDIA's recommended negative prompt for i2v, while an empty string disables negative guidance.
enable_prompt_expansionoptional
Typeboolean
Defaulttrue
Prompt expansion that, when true, uses Cosmos3-Nano Reasoner to produce the dense-caption format used to train Cosmos3; this VLM examines the first frame. COSMOS_PROMPT_UPSAMPLER_BASE_URL selects a configured Reasoner service; otherwise, the app launches one locally. Expansion failure leaves the unmodified prompt in use.
enable_agentic_generationoptional
Typeboolean
Defaultfalse
Iterative Cosmos agentic generation, substantially slower and costlier than generating once because every candidate requires a full render. The loop upsamples the prompt, generates video candidates, uses a VLM to critique sampled frames, and rewrites the prompt.
agentic_max_iterationsoptional
Typeinteger
Default2
1 – 3
Upper limit on prompt stages with agentic generation enabled.
agentic_samples_per_iterationoptional
Typeinteger
Default2
1 – 3
Per-iteration video candidate count for the agentic loop, determining how many videos compete in evaluation before the best proceeds to prompt rewriting.
agentic_early_stopoptional
Typeboolean
Defaulttrue
Early-stop control that treats a critic score above the strict quality threshold as sufficient to end the agentic loop.
image_sizeoptional
Typeobject
Default{"height":480,"width":832}
square_hdsquareportrait_4_3portrait_16_9landscape_4_3landscape_16_9
Video dimensions requested before automatic adjustment. To obtain a supported size, the system clamps the request and chooses the nearest NVIDIA tier—256p, 480p, or 720p—and supported aspect ratio.
num_framesoptional
Typeinteger
Default189
5 – 189
Frame count for generation; increasing it produces a longer video.
frames_per_secondoptional
Typeinteger
Default24
4 – 60
Output video frame rate, measured in frames per second.
num_inference_stepsoptional
Typeinteger
Default28
1 – 50
Denoising step count, trading longer processing for better quality as the count increases.
guidance_scaleoptional
Typenumber
Default6
0 – 20
Strength of classifier-free guidance.
seedoptional
Typeinteger | null
Seed controlling repeatability: an identical prompt and seed produce an identical video when the model version is unchanged.
enable_safety_checkeroptional
Typeboolean
Defaulttrue
Moderation control for the supplied image and prompt, with disabling permitted only for authorized accounts. Requests lacking authorization undergo checks regardless of the setting.
required Canonical name, also accepted

File parameters take a public https:// URL, a data: URI up to 10 MB, or the file_url of an upload. Files

Output

GET /requests/req_8k2m4q7x1z9c3v5b6n0p2r4t
{
  "id": "req_8k2m4q7x1z9c3v5b6n0p2r4t",
  "object": "request",
  "model": "nvidia/cosmos-3-super/image-to-video",
  "status": "succeeded",
  "created_at": "2026-10-20T12:00:00Z",
  "started_at": "2026-10-20T12:00:02Z",
  "completed_at": "2026-10-20T12:00:09Z",
  "queue_position": null,
  "cost": {
    "credits": 400,
    "usd": "0.40",
    "charged": 400,
    "refunded": 0
  },
  "input": {
    "prompt": "A red fox trots through fresh snow at sunrise, soft golden light, shallow depth of field",
    "image_url": "https://content.doitong.com/images/showcase/v3/gen_fox_cartoon.jpg"
  },
  "output": {
    "video": {
      "url": "https://content.doitong.com/dev-api/2026/10/20/req_8k2m4q7x1z9c3v5b6n0p2r4t/0-output.mp4",
      "content_type": "video/mp4",
      "file_name": "0-output.mp4",
      "file_size": 1843200,
      "width": 1024,
      "height": 768,
      "fps": 1.0,
      "duration": 5,
      "num_frames": 1
    },
    "seed": 42
  },
  "outputs": [
    {
      "type": "video",
      "url": "https://content.doitong.com/dev-api/2026/10/20/req_8k2m4q7x1z9c3v5b6n0p2r4t/0-output.mp4",
      "content_type": "video/mp4",
      "file_size": 1843200,
      "width": 1280,
      "height": 720,
      "duration": 5,
      "stored": true
    }
  ],
  "error": null,
  "metrics": {
    "inference_seconds": 6.8
  },
  "urls": {
    "self": "https://api.doitong.ru/v1/requests/req_8k2m4q7x1z9c3v5b6n0p2r4t",
    "cancel": "https://api.doitong.ru/v1/requests/req_8k2m4q7x1z9c3v5b6n0p2r4t/cancel"
  }
}
Your files, your choiceEvery outputs[].url is a copy on our storage: kept forever (paid per GB-month) or 7 days for free, by your account setting or ?storage=.
Same shape on every modeltype, url, content_type, width, height and duration, for images, video and audio alike. output is the model’s own result object.
Fields of output
video
object {url, content_type, file_name, file_size, width, height, fps, duration, num_frames}Video produced by the model.
seed
integerSeed value applied during generation.

More models: Image to Video

For AI agents: this page as Markdown is at https://doitong.com/en-US/developers/models/nvidia/cosmos-3-super/image-to-video.md. All models and guides: https://doitong.com/developers/llms.txt