Saltar al cuntegna

Cosmos 3 Super Image to Video

Integrar

ID dal model 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"

Datotecas sco l'exempel: $0,40 400 credits API · ≈ ₽43,03 tias datotecas vegnan mesiradas avant l'execuziun Lantschar sin questa pagina

Obtegna ina clav API
NVIDIA Maletg a video Lipsync #stylized#transform

Animescha in maletg dal prim frame en in video, cun engrondiment opziunal dal prompt e in ciclus da perfecziunament agentic

NVIDIA · Video

Tes emprim resultat vegn qua

Construis ina dumonda sutvart e l'exequescha cun tia clav API. Outputs vegnan copiads en nossa memorisaziun, conservads per adina u 7 dis gratuitamain.

Construir ina dumonda

Ils controls vegnan generads dal schema dal model. Il code e ​​il pretsch s'actualiseschan durant che ti als midas.

•obligatori 88 / 4096

Instrucziuns da generaziun video en text: inditgescha tge che la scena cuntegna e tge moviment ch’ella duai mussar.

•obligatori image_url
Trametta in datoteca u

URL dal maletg utilisà per definir il first frame dal video.

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

Dimensiuns dal video dumandadas avant l’adattaziun automatica. Per obtegnair ina grondezza sustegnida, limita il sistem la dumonda e tscherna la stgalimada NVIDIA la pli datiers — 256p, 480p u 720p — e la proporziun sustegnida.

standard 2

Limita superiura da las fasas dal prompt cun generaziun agentica activa.

standard 2

Dumber da candidats video per iteraziun dal ciclus agentic; determinescha quant videos che concurran en l’evaluaziun avant che il meglier cuntinuescha a la reelavuraziun dal prompt.

standard 189

Dumber da frames per la generaziun; augmentar quel producescha in video pli lung.

standard 24

Ritmus da frames dal video d'output, mesirà en frames per secunda.

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

Prompt negativ per evitar cuntegns sco moviments u artefacts nungiavischads. Senza quest prompt vegn utilisà il prompt negativ recumandà da NVIDIA per i2v; ina chadaina vida disactivescha la guida negativa.

standard 28

Dumber da pass da denoisaziun, che baratta in'elavuraziun pli lunga per ina meglra qualitad cun l'augment dal dumber.

standard 6

Furma da la guida senza classificatur.

Seed che controlla la reproducibilitad: in prompt e seed identics produceschan in video identic, sch’il model resta la medema versiun.

POST /run/nvidia/cosmos-3-super/image-to-video?wait=20 calculaziun → $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"

Executar questa dumonda

Tia clav resta en quest browser: ella vegn tramessa mo al Doitong API e memorisada (localStorage) mo sch'ella scheccas Tiar a memoria. Obtegnair ina clav API

Davart Cosmos 3 Super Image to Video

Utilisescha in `image_url` sco frame d'avertura ed in `prompt` da maximalmain 4096 caracters. `num_frames` va fin a 189 e `frames_per_second` da 4 fin a 60. L'engron­diment dal prompt è activà per standard, e `enable_agentic_generation` exequescha candidats iterativs cun critica. Offerescha er `negative_prompt`, `guidance_scale` e `num_inference_steps` fin a 50.

Parameters

Invia u il num canonic u il num nativ, betg omaduas. Nums canonics

NumTip StandardPermess Descripziun
promptobligatori
Tipstring
≤ 4 096 caracters
Instrucziuns da generaziun video en text: inditgescha tge che la scena cuntegna e tge moviment ch’ella duai mussar.
image_urlobligatori image
Tipurl · image
URL dal maletg utilisà per definir il first frame dal video.
negative_promptopziunal
Tipstring
StandardThe 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 caracters
Prompt negativ per evitar cuntegns sco moviments u artefacts nungiavischads. Senza quest prompt vegn utilisà il prompt negativ recumandà da NVIDIA per i2v; ina chadaina vida disactivescha la guida negativa.
enable_prompt_expansionopziunal
Tipboolean
Standardtrue
Expansiun dal prompt che, sche true, utilisescha Cosmos3-Nano Reasoner per producir il format dense-caption utilisà per trenar Cosmos3; quest VLM examinescha il first frame. COSMOS_PROMPT_UPSAMPLER_BASE_URL tscherna in servetsch Reasoner configurà; autramain l’app lantscha in localmain. Sche l’expansiun falla, vegn utilisà il prompt betg midà.
enable_agentic_generationopziunal
Tipboolean
Standardfalse
Generaziun agentica iterativa da Cosmos, bler pli plauna e pli chara che la generaziun unica, perquai che mintga candidat pretenda ina renderisaziun cumpletta. Il ciclus amplifitgescha il prompt, generescha candidats video, utilisescha in VLM per crititgar frames campiunads e reela­vura il prompt.
agentic_max_iterationsopziunal
Tipinteger
Standard2
1 – 3
Limita superiura da las fasas dal prompt cun generaziun agentica activa.
agentic_samples_per_iterationopziunal
Tipinteger
Standard2
1 – 3
Dumber da candidats video per iteraziun dal ciclus agentic; determinescha quant videos che concurran en l’evaluaziun avant che il meglier cuntinuescha a la reelavuraziun dal prompt.
agentic_early_stopopziunal
Tipboolean
Standardtrue
Controlla d’arrest anticipà che considerescha in resultat dal critic sur la limita da qualitad strict sco suffizient per terminar il ciclus agentic.
image_sizeopziunal
Tipobject
Standard{"height":480,"width":832}
square_hdsquareportrait_4_3portrait_16_9landscape_4_3landscape_16_9
Dimensiuns dal video dumandadas avant l’adattaziun automatica. Per obtegnair ina grondezza sustegnida, limita il sistem la dumonda e tscherna la stgalimada NVIDIA la pli datiers — 256p, 480p u 720p — e la proporziun sustegnida.
num_framesopziunal
Tipinteger
Standard189
5 – 189
Dumber da frames per la generaziun; augmentar quel producescha in video pli lung.
frames_per_secondopziunal
Tipinteger
Standard24
4 – 60
Ritmus da frames dal video d'output, mesirà en frames per secunda.
num_inference_stepsopziunal
Tipinteger
Standard28
1 – 50
Dumber da pass da denoisaziun, che baratta in'elavuraziun pli lunga per ina meglra qualitad cun l'augment dal dumber.
guidance_scaleopziunal
Tipnumber
Standard6
0 – 20
Furma da la guida senza classificatur.
seedopziunal
Tipinteger | null
Seed che controlla la reproducibilitad: in prompt e seed identics produceschan in video identic, sch’il model resta la medema versiun.
enable_safety_checkeropziunal
Tipboolean
Standardtrue
Controlla da moderaziun per il maletg e prompt furnids, cun disactivaziun permessa mo per contos autorisads. Dumondas senza autorisaziun vegnan controlladas independentamain da la configuraziun.
obligatori Num canonic, er acceptà

Ils parameters da datotecas acceptan ina URL publica https://, ina URI data: fin a 10 MB u il file_url d'ina chargiada. Datotecas

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"
  }
}
Tias datotecas, tia decisiunMintga outputs[].url è ina copia en nossa memorisaziun: conservada per adina (pajà per GB-mes) u 7 dis gratuitamain, tenor la configuraziun dal conto u ?storage=.
La medema structura per mintga modeltype, url, content_type, width, height e duration, tant per maletgs sco per video e audio. output è l'object da resultat dal model sez.
Champs da l'output
video
object {url, content_type, file_name, file_size, width, height, fps, duration, num_frames}Video producì dal model.
seed
integerValur da seed applitgada durant la generaziun.

Dapli models: Maletg a video

Per agents AI: questa pagina sco Markdown è disponibla sut https://doitong.com/rm/developers/models/nvidia/cosmos-3-super/image-to-video.md. Tut ils models e las guidas: https://doitong.com/developers/llms.txt