—
NVIDIA Cosmos — Image to Video (nvidia/cosmos-3-super/image-to-video)
Cosmos3 is a collection of Omnimodal world models capable of generating dynamic, high-quality video, image, audio, and action commands from combinations of text, image, video, and action trajectory inputs.
Input
Add your inputs and customize the generation settings.
Result & Pricing
Video output — your generated video will appear here.
Your video will appear here
Choose your inputs and customize your settings.
Model pricing (tax excluded) per billed second
| Specification | Webot INR |
|---|---|
| Rate | ₹5.99 |
Calculate credits in the input panel for your selected settings.
Prices, estimates and credit deductions exclude tax. Any applicable tax is charged only when you buy plans or credits.
Billed duration rounds up to whole seconds. Disable agentic generation. Intermediate candidate billing cannot be verified. Disable prompt expansion for this video workflow until its separate fee is documented. Prices are converted to INR. Credits use your current plan rate.
- Prices shown are estimates, not final charges. Final billing details will be available in Reports after generation is complete.
API
WEBOT API · Image to Video (nvidia/cosmos-3-super/image-to-video)
Generate from your server with your WEBOT account credits.
Use your 120-character WEBOT customer key (a–z, A–Z, 0–9). Send
Authorization: Bearer YOUR_KEY. Keys are secrets: call from a server, never from public frontend code. Results come from verified webhooks and remain saved if your stream disconnects.1. Configure your server
WEBOT_BASE_URL=https://webot.online
WEBOT_API_KEY=<your-server-side-WEBOT-key>This base URL is resolved from verified domain mappings and your direct creator reseller. Forwarded Host headers are not trusted. New custom reseller domains need DNS, TLS and platform verification before activation.
2. Quote, then generate
| Method | Endpoint | Purpose |
|---|---|---|
| POST | https://webot.online/api/v1/models/creative_6c030d1b9bc94677b6bd1a62/quote | Validate input and calculate credits (no deduction). |
| POST | https://webot.online/api/v1/models/creative_6c030d1b9bc94677b6bd1a62/requests | Reserve quoted credits and queue an asynchronous job. |
| GET | https://webot.online/api/v1/models/creative_6c030d1b9bc94677b6bd1a62/requests/{request_id}/result | Read your saved result: 202 pending, 200 terminal. |
| GET | https://webot.online/api/v1/models/creative_6c030d1b9bc94677b6bd1a62/requests/{request_id}/events | Authenticated SSE: generation / terminal events. |
Send JSON {"input":{…}} to quote. Send the same input plus integer quoted_credits to generate, with Authorization: Bearer … and Idempotency-Key. Use a unique 16–128 character key per new job; persist it and reuse the same key and input on any uncertain response. Never retry with a new key: that creates a second chargeable job.
Credits are reserved only after validation and a durable request is saved. A changed quote or insufficient balance returns 409. Credits exclude purchase taxes. The quoted amount is an initial reservation. After generation, the final charge replaces it using confirmed usage. Any verified unused allocation becomes available automatically; additional usage may require more credits. Missing billing stays pending, including for failed jobs.
3. Input example
Save this as input.json. Replace sample prompts, files and media URLs with your own. Schemas do not guarantee provider acceptance of the content.
Supported generation options
- Disable agentic generation. Intermediate candidate billing cannot be verified.
- Disable prompt expansion for this video workflow until its separate fee is documented.
{
"prompt": "Audit sample text",
"image_url": "https://example.com/source.png",
"enable_agentic_generation": false,
"enable_prompt_expansion": false,
"num_frames": 189,
"agentic_max_iterations": 2,
"guidance_scale": 6,
"num_inference_steps": 28,
"agentic_early_stop": true,
"sync_mode": false,
"negative_prompt": "The 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 moir\u00e9 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.",
"enable_safety_checker": true,
"image_size": {
"height": 480,
"width": 832
},
"agentic_samples_per_iteration": 2,
"frames_per_second": 24
}4. Language examples
# Save the model's example input as input.json, then replace prompts/media URLs.
# Requires curl and jq. Store NONCE once for this job; reuse it on any retry.
export NONCE="$(openssl rand -hex 16)"
BASE="${WEBOT_BASE_URL%/}"
jq -n --slurpfile input input.json '{input:$input[0]}' > quote-body.json
curl --fail-with-body -sS "$BASE/api/v1/models/creative_6c030d1b9bc94677b6bd1a62/quote" \
-H "Authorization: Bearer $WEBOT_API_KEY" -H 'Content-Type: application/json' \
--data-binary @quote-body.json > quote.json
CREDITS=$(jq -er '.quote.credits_consumed' quote.json)
jq --argjson credits "$CREDITS" '. + {quoted_credits:$credits}' quote-body.json > request-body.json
curl --fail-with-body -sS "$BASE/api/v1/models/creative_6c030d1b9bc94677b6bd1a62/requests" \
-H "Authorization: Bearer $WEBOT_API_KEY" -H "Idempotency-Key: $NONCE" \
-H 'Content-Type: application/json' --data-binary @request-body.json > job.json
# Subscribe; after a nonterminal disconnect, reconnect this SAME URL.
curl --fail-with-body -N "$(jq -er '.events_url' job.json)" -H "Authorization: Bearer $WEBOT_API_KEY"
# Read saved output after terminal notification (never call provider status/results).
curl --fail-with-body -sS "$(jq -er '.result_url' job.json)" -H "Authorization: Bearer $WEBOT_API_KEY"// Node.js 18+. Save your customized input in input.json.
import { readFileSync } from 'node:fs';
import { randomUUID } from 'node:crypto';
const base = process.env.WEBOT_BASE_URL.replace(/\/$/, '');
const key = process.env.WEBOT_API_KEY;
const input = JSON.parse(readFileSync('input.json', 'utf8'));
const nonce = process.env.WEBOT_JOB_NONCE || randomUUID(); // Persist before submission.
async function post(path, body, nonce) {
const response = await fetch(base + path, {method:'POST', headers:{
Authorization:`Bearer ${key}`, 'Content-Type':'application/json',
...(nonce ? {'Idempotency-Key':nonce} : {})}, body:JSON.stringify(body)});
const json = await response.json();
if (!response.ok) throw new Error(JSON.stringify(json));
return json;
}
const {quote} = await post('/api/v1/models/creative_6c030d1b9bc94677b6bd1a62/quote', {input});
const job = await post('/api/v1/models/creative_6c030d1b9bc94677b6bd1a62/requests', {input, quoted_credits:quote.credits_consumed}, nonce);
console.log(job); // Stable WEBOT request_id, events_url, result_url.
// Stream saved webhook state. Reconnect ONLY the event stream, not the generation.
while (!job.terminal || job.financial_pending) {
const stream = await fetch(job.events_url, {headers:{Authorization:`Bearer ${key}`}});
if (!stream.ok) throw new Error(`Event stream HTTP ${stream.status}`);
let buffer = '';
for await (const chunk of stream.body.pipeThrough(new TextDecoderStream())) {
buffer += chunk;
let end;
while ((end = buffer.indexOf('\n\n')) !== -1) {
const event = buffer.slice(0, end); buffer = buffer.slice(end + 2);
const data = event.split('\n').find(line => line.startsWith('data: '));
if (!data) continue;
const state = JSON.parse(data.slice(6));
if (state.error && !state.status) throw new Error(JSON.stringify(state.error));
if (state.terminal && !state.financial_pending) { console.log(state.output ?? state.error); process.exit(0); }
}
}
await new Promise(resolve => setTimeout(resolve, 1500));
}
console.log(job.output ?? job.error);// Node.js 18+. Save your customized input in input.json.
import { readFileSync } from 'node:fs';
import { randomUUID } from 'node:crypto';
const base = process.env.WEBOT_BASE_URL!.replace(/\/$/, '');
const key = process.env.WEBOT_API_KEY!;
const input = JSON.parse(readFileSync('input.json', 'utf8'));
const nonce = process.env.WEBOT_JOB_NONCE || randomUUID(); // Persist before submission.
async function post(path: string, body: unknown, nonce?: string) {
const response = await fetch(base + path, {method:'POST', headers:{
Authorization:`Bearer ${key}`, 'Content-Type':'application/json',
...(nonce ? {'Idempotency-Key':nonce} : {})}, body:JSON.stringify(body)});
const json = await response.json();
if (!response.ok) throw new Error(JSON.stringify(json));
return json;
}
const {quote} = await post('/api/v1/models/creative_6c030d1b9bc94677b6bd1a62/quote', {input});
const job = await post('/api/v1/models/creative_6c030d1b9bc94677b6bd1a62/requests', {input, quoted_credits:quote.credits_consumed}, nonce);
console.log(job); // Stable WEBOT request_id, events_url, result_url.
// Stream saved webhook state. Reconnect ONLY the event stream, not the generation.
while (!job.terminal || job.financial_pending) {
const stream = await fetch(job.events_url, {headers:{Authorization:`Bearer ${key}`}});
if (!stream.ok) throw new Error(`Event stream HTTP ${stream.status}`);
let buffer = '';
for await (const chunk of stream.body.pipeThrough(new TextDecoderStream())) {
buffer += chunk;
let end;
while ((end = buffer.indexOf('\n\n')) !== -1) {
const event = buffer.slice(0, end); buffer = buffer.slice(end + 2);
const data = event.split('\n').find(line => line.startsWith('data: '));
if (!data) continue;
const state = JSON.parse(data.slice(6));
if (state.error && !state.status) throw new Error(JSON.stringify(state.error));
if (state.terminal && !state.financial_pending) { console.log(state.output ?? state.error); process.exit(0); }
}
}
await new Promise(resolve => setTimeout(resolve, 1500));
}
console.log(job.output ?? job.error);# Python 3.9+, pip install requests
import json, os, time, uuid, requests
base = os.environ['WEBOT_BASE_URL'].rstrip('/')
headers = {'Authorization': 'Bearer ' + os.environ['WEBOT_API_KEY']}
with open('input.json', encoding='utf-8') as file:
input_data = json.load(file)
nonce = os.environ.get('WEBOT_JOB_NONCE') or str(uuid.uuid4()) # Persist before submitting.
def post(path, body, extra=None):
response = requests.post(base + path, headers={**headers, **(extra or {})}, json=body, timeout=180)
response.raise_for_status()
return response.json()
quote = post('/api/v1/models/creative_6c030d1b9bc94677b6bd1a62/quote', {'input': input_data})['quote']
job = post('/api/v1/models/creative_6c030d1b9bc94677b6bd1a62/requests', {'input': input_data, 'quoted_credits': quote['credits_consumed']}, {'Idempotency-Key': nonce})
state = job
while not state['terminal'] or state.get('financial_pending', False):
# Reconnect while nonterminal; this reads WEBOT's database, never a provider status API.
with requests.get(job['events_url'], headers=headers, stream=True, timeout=(10, 40)) as stream:
stream.raise_for_status()
for line in stream.iter_lines(decode_unicode=True):
if line and line.startswith('data: '):
state = json.loads(line[6:])
if 'status' not in state: raise RuntimeError(state)
if state['terminal'] and not state.get('financial_pending', False): break
if not state['terminal'] or state.get('financial_pending', False): time.sleep(1.5)
print(state['output'] if state['status'] == 'completed' else state['error'])<?php // PHP 8.1+ with cURL
$base = rtrim(getenv('WEBOT_BASE_URL'), '/');
$input = json_decode(file_get_contents('input.json'), false, 64, JSON_THROW_ON_ERROR);
$nonce = getenv('WEBOT_JOB_NONCE') ?: bin2hex(random_bytes(16)); // Persist this before submission.
function post(string $path, array $body, ?string $nonce = null): array {
global $base;
$headers = ['Authorization: Bearer '.getenv('WEBOT_API_KEY'), 'Content-Type: application/json'];
if ($nonce) $headers[] = 'Idempotency-Key: '.$nonce;
$ch = curl_init($base.$path);
curl_setopt_array($ch, [CURLOPT_RETURNTRANSFER=>true, CURLOPT_POST=>true,
CURLOPT_TIMEOUT=>180, CURLOPT_HTTPHEADER=>$headers,
CURLOPT_POSTFIELDS=>json_encode($body, JSON_THROW_ON_ERROR)]);
$raw = curl_exec($ch); $status = curl_getinfo($ch, CURLINFO_HTTP_CODE); curl_close($ch);
if ($raw === false || $status < 200 || $status >= 300) throw new RuntimeException('WEBOT HTTP '.$status);
return json_decode($raw, true, 64, JSON_THROW_ON_ERROR);
}
$quote = post('/api/v1/models/creative_6c030d1b9bc94677b6bd1a62/quote', ['input'=>$input])['quote'];
$job = post('/api/v1/models/creative_6c030d1b9bc94677b6bd1a62/requests', ['input'=>$input, 'quoted_credits'=>$quote['credits_consumed']], $nonce);
echo json_encode($job, JSON_PRETTY_PRINT); // Connect to events_url with the same Bearer header.// Go 1.20+; save as main.go.
package main
import ("bytes"; "encoding/json"; "fmt"; "io"; "net/http"; "os"; "strings"; "time"; "crypto/rand"; "encoding/hex")
func main() {
base := strings.TrimRight(os.Getenv("WEBOT_BASE_URL"), "/")
raw, err := os.ReadFile("input.json"); if err != nil { panic(err) }
nonce := os.Getenv("WEBOT_JOB_NONCE")
if nonce == "" { b := make([]byte,16); if _,err:=rand.Read(b);err!=nil {panic(err)}; nonce=hex.EncodeToString(b) }
// Persist nonce before the submission; reuse on ambiguous responses.
client := &http.Client{Timeout:180*time.Second}
post := func(path string, body any, id string) map[string]any {
data, err := json.Marshal(body); if err != nil {panic(err)}
req, err := http.NewRequest("POST",base+path,bytes.NewReader(data)); if err!=nil {panic(err)}
req.Header.Set("Authorization","Bearer "+os.Getenv("WEBOT_API_KEY")); req.Header.Set("Content-Type","application/json")
if id!="" {req.Header.Set("Idempotency-Key",id)}
res,err:=client.Do(req); if err!=nil {panic(err)}; defer res.Body.Close()
data,err=io.ReadAll(res.Body); if err!=nil {panic(err)}
if res.StatusCode<200||res.StatusCode>=300 {panic(string(data))}
var result map[string]any; if err=json.Unmarshal(data,&result);err!=nil {panic(err)}; return result
}
input:=json.RawMessage(raw)
quote:=post("/api/v1/models/creative_6c030d1b9bc94677b6bd1a62/quote",map[string]any{"input":input},"")["quote"].(map[string]any)
job:=post("/api/v1/models/creative_6c030d1b9bc94677b6bd1a62/requests",map[string]any{"input":input,"quoted_credits":quote["credits_consumed"]},nonce)
fmt.Println(job) // Stream events_url with the same Authorization header.
}// Java 11+, Jackson databind dependency; class Main in Main.java.
import java.net.URI;
import java.net.http.*;
import java.nio.file.*;
import java.time.Duration;
import java.util.UUID;
import com.fasterxml.jackson.databind.*;
import com.fasterxml.jackson.databind.node.ObjectNode;
public class Main {
static final ObjectMapper json = new ObjectMapper();
static final HttpClient client = HttpClient.newHttpClient();
static final String base = System.getenv("WEBOT_BASE_URL").replaceAll("/$", "");
static JsonNode post(String path, JsonNode body, String nonce) throws Exception {
var b = HttpRequest.newBuilder(URI.create(base+path)).timeout(Duration.ofSeconds(180))
.header("Authorization","Bearer "+System.getenv("WEBOT_API_KEY")).header("Content-Type","application/json")
.POST(HttpRequest.BodyPublishers.ofString(json.writeValueAsString(body)));
if(nonce!=null)b.header("Idempotency-Key",nonce);
var r=client.send(b.build(),HttpResponse.BodyHandlers.ofString());
if(r.statusCode()<200||r.statusCode()>=300)throw new IllegalStateException(r.body());
return json.readTree(r.body());
}
public static void main(String[] args) throws Exception {
var input=json.readTree(Files.readString(Path.of("input.json")));
ObjectNode body=json.createObjectNode();body.set("input",input);
var quote=post("/api/v1/models/creative_6c030d1b9bc94677b6bd1a62/quote",body,null).get("quote");
body.put("quoted_credits",quote.get("credits_consumed").asInt());
String nonce=System.getenv("WEBOT_JOB_NONCE");if(nonce==null)nonce=UUID.randomUUID().toString(); // Persist before POST.
var job=post("/api/v1/models/creative_6c030d1b9bc94677b6bd1a62/requests",body,nonce);System.out.println(job);
// Subscribe to job.events_url with Authorization; use BodyHandlers.ofLines().
}
}// .NET 8 console app, top-level statements.
using System.Net.Http;
using System.Text;
using System.Text.Json.Nodes;
var client = new HttpClient { Timeout = TimeSpan.FromSeconds(180) };
var root = Environment.GetEnvironmentVariable("WEBOT_BASE_URL")!.TrimEnd('/');
var key = Environment.GetEnvironmentVariable("WEBOT_API_KEY");
async Task<JsonNode> Post(string path, JsonNode body, string? nonce = null) {
using var req = new HttpRequestMessage(HttpMethod.Post, root + path);
req.Headers.Add("Authorization", "Bearer " + key);
if (nonce != null) req.Headers.Add("Idempotency-Key", nonce);
req.Content = new StringContent(body.ToJsonString(), Encoding.UTF8, "application/json");
using var res = await client.SendAsync(req);
var raw = await res.Content.ReadAsStringAsync();res.EnsureSuccessStatusCode();
return JsonNode.Parse(raw)!;
}
var input = JsonNode.Parse(await File.ReadAllTextAsync("input.json"));
var body = new JsonObject { ["input"] = input };
var quote = await Post("/api/v1/models/creative_6c030d1b9bc94677b6bd1a62/quote", body);
body["quoted_credits"] = quote["quote"]!["credits_consumed"]!.GetValue<int>();
var nonce = Environment.GetEnvironmentVariable("WEBOT_JOB_NONCE") ?? Guid.NewGuid().ToString(); // Persist before submitting.
var job = await Post("/api/v1/models/creative_6c030d1b9bc94677b6bd1a62/requests", body, nonce);
Console.WriteLine(job); // Subscribe to events_url with the same Bearer header and ResponseHeadersRead.# Ruby 3+, standard library only.
require 'net/http'
require 'json'
require 'securerandom'
base = ENV.fetch('WEBOT_BASE_URL').sub(%r{/$}, '')
post = lambda do |path, body, nonce = nil|
uri = URI(base + path)
req = Net::HTTP::Post.new(uri)
req['Authorization'] = 'Bearer ' + ENV.fetch('WEBOT_API_KEY')
req['Content-Type'] = 'application/json'
req['Idempotency-Key'] = nonce if nonce
req.body = JSON.generate(body)
res = Net::HTTP.start(uri.host, uri.port, use_ssl: true, read_timeout: 180) { |http| http.request(req) }
raise res.body unless res.is_a?(Net::HTTPSuccess)
JSON.parse(res.body)
end
input = JSON.parse(File.read('input.json'))
quote = post.call('/api/v1/models/creative_6c030d1b9bc94677b6bd1a62/quote', {'input' => input})['quote']
nonce = ENV['WEBOT_JOB_NONCE'] || SecureRandom.uuid # Persist this before submission.
job = post.call('/api/v1/models/creative_6c030d1b9bc94677b6bd1a62/requests', {'input' => input, 'quoted_credits' => quote['credits_consumed']}, nonce)
puts JSON.pretty_generate(job) # Subscribe to events_url with the same Authorization header.// Swift 5.9 server-side/CLI only. Do not embed customer keys in shipped apps.
import Foundation
#if canImport(FoundationNetworking)
import FoundationNetworking
#endif
@main struct Main {
static func main() async throws {
let env=ProcessInfo.processInfo.environment
let base=env["WEBOT_BASE_URL"]!.trimmingCharacters(in: CharacterSet(charactersIn:"/"))
let input=try JSONSerialization.jsonObject(with: Data(contentsOf:URL(fileURLWithPath:"input.json")))
func post(_ path:String,_ body:[String:Any],_ nonce:String?=nil) async throws -> [String:Any] {
var req=URLRequest(url:URL(string:base+path)!);req.httpMethod="POST";req.timeoutInterval=180
req.setValue("Bearer "+env["WEBOT_API_KEY"]!,forHTTPHeaderField:"Authorization")
req.setValue("application/json",forHTTPHeaderField:"Content-Type")
if let nonce {req.setValue(nonce,forHTTPHeaderField:"Idempotency-Key")}
req.httpBody=try JSONSerialization.data(withJSONObject:body)
let (data,res)=try await URLSession.shared.data(for:req)
guard let http=res as? HTTPURLResponse,(200..<300).contains(http.statusCode) else {throw NSError(domain:"WEBOT",code:1)}
return try JSONSerialization.jsonObject(with:data) as! [String:Any]
}
let quote=try await post("/api/v1/models/creative_6c030d1b9bc94677b6bd1a62/quote",["input":input])["quote"] as! [String:Any]
let nonce=env["WEBOT_JOB_NONCE"] ?? UUID().uuidString // Persist before submission.
let job=try await post("/api/v1/models/creative_6c030d1b9bc94677b6bd1a62/requests",["input":input,"quoted_credits":quote["credits_consumed"]!],nonce)
print(job) // Subscribe to events_url with the same Authorization header.
}
}// Server-side Kotlin/JVM 17+, Gradle dependencies: OkHttp 4.12, org.json 20240303.
import okhttp3.*
import okhttp3.MediaType.Companion.toMediaType
import okhttp3.RequestBody.Companion.toRequestBody
import org.json.JSONObject
import java.io.File
import java.util.UUID
import java.util.concurrent.TimeUnit
fun main() {
val base=System.getenv("WEBOT_BASE_URL").trimEnd('/')
val client=OkHttpClient.Builder().readTimeout(180,TimeUnit.SECONDS).build()
fun post(path:String,body:JSONObject,nonce:String?=null):JSONObject {
val b=Request.Builder().url(base+path).header("Authorization","Bearer "+System.getenv("WEBOT_API_KEY"))
.post(body.toString().toRequestBody("application/json".toMediaType()))
if(nonce!=null)b.header("Idempotency-Key",nonce)
return client.newCall(b.build()).execute().use {res ->
val raw=res.body!!.string();check(res.isSuccessful){raw};JSONObject(raw)
}
}
val body=JSONObject().put("input",JSONObject(File("input.json").readText()))
val quote=post("/api/v1/models/creative_6c030d1b9bc94677b6bd1a62/quote",body).getJSONObject("quote")
body.put("quoted_credits",quote.getInt("credits_consumed"))
val nonce=System.getenv("WEBOT_JOB_NONCE") ?: UUID.randomUUID().toString() // Persist before submission.
val job=post("/api/v1/models/creative_6c030d1b9bc94677b6bd1a62/requests",body,nonce)
println(job) // Subscribe to events_url with the same Authorization header.
}5. Receive all results
The response includes a stable WEBOT request_id, events_url and result_url. Subscribe with a Bearer-authenticated HTTP stream. Nonterminal streams end after 25 seconds; reconnect to the same URL after at least 1.5 seconds. Current saved state is replayed immediately, including completion that happened while disconnected. A terminal event contains the result envelope. The JSON terminal flag describes generation completion. Keep reconnecting while financial_pending is true; stop after the terminal event confirms financial settlement. Never poll provider status/results. Native browser EventSource cannot send a Bearer header; keep the key on your server.
Completed output contains the full model result, including every generated media/document item, not just the first URL. Failed output is null with error and refund_state. Results are strictly limited to the key owner, including on reseller domains. Model-independent aliases: /api/v1/requests/{request_id}/result and /events. Revoking a key blocks subsequent reads and streams; saved results remain accessible with another valid key of the same account.
6. Schemas and parameter defaults
Required fields are marked with *. Required text must contain non-whitespace content. Defaults apply when an optional parameter is omitted; null, false, 0 and an empty array are distinct values. “No documented default” means WEBOT does not invent one. Examples are sample inputs, not defaults. WEBOT generation constraints can require an explicit value even when the provider publishes an automatic default.
Input schema · fields, constraints and defaults
| Parameter | Type | Required | Default when omitted | Description / constraints |
|---|---|---|---|---|
seed | integer | null | No | No documented default | The same seed and prompt given to the same model version will produce the same video every time. |
num_frames | integer | No | 189 | Number of frames to generate. More frames yield a longer video. minimum: 5 · maximum: 189 |
agentic_max_iterations | integer | No | 2 | Maximum agentic prompt stages when agentic generation is enabled. minimum: 1 · maximum: 3 |
enable_agentic_generation | boolean | No | false | Enable the iterative Cosmos agentic loop: prompt upsampling, candidate video generation, VLM critique of sampled frames, and prompt rewrite. Each candidate is a full render, so this is substantially slower and costlier than a single generation. const: false |
enable_prompt_expansion | boolean | No | false | If true, the Cosmos3-Nano Reasoner (a VLM that sees the first frame) rewrites the prompt into the dense caption Cosmos3 was trained on. The app starts a local Reasoner by default, or uses COSMOS_PROMPT_UPSAMPLER_BASE_URL when configured. Falls back to the raw prompt if expansion fails. const: false |
guidance_scale | number | No | 6 | Classifier-free guidance scale. Higher values increase prompt adherence at the cost of diversity. minimum: 0 · maximum: 20 |
num_inference_steps | integer | No | 28 | Number of denoising steps. More steps yield higher quality but take longer. minimum: 1 · maximum: 50 |
agentic_early_stop | boolean | No | true | Stop the agentic loop early when the critic score clears the strict quality threshold. |
image_url | string | Yes * | No documented default | URL of the conditioning first-frame image for the video. |
sync_mode | boolean | No | false | If `True`, the video is returned as a data URI and the output data won't be available in the request history. |
prompt | string | Yes * | No documented default | Text prompt describing the motion and scene of the video to generate. minLength: 1 · maxLength: 4096 |
negative_prompt | string | No | "The 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." | Content to steer the generation away from (artifacts, unwanted motion). Defaults to NVIDIA's recommended i2v negative prompt; pass an empty string to disable. maxLength: 2048 |
enable_safety_checker | boolean | No | true | Enable content moderation for the input prompt and image. Disabling it requires account authorization; unauthorized requests are always checked. |
image_size | object | string | No | {"height":480,"width":832} | The size of the generated video. The request is clamped and snapped to the nearest supported NVIDIA tier (256p/480p/720p) and aspect ratio. |
image_size.variant0.width | integer | No | 512 | The width of the generated image. maximum: 14142 · exclusiveMinimum: 0 |
image_size.variant0.height | integer | No | 512 | The height of the generated image. maximum: 14142 · exclusiveMinimum: 0 |
agentic_samples_per_iteration | integer | No | 2 | Candidate videos to generate and judge per agentic iteration. The best candidate advances to the next rewrite stage. minimum: 1 · maximum: 3 |
frames_per_second | integer | No | 24 | Frames per second of the output video. minimum: 4 · maximum: 60 |
{
"title": "ImageToVideoInput",
"x-webot-order-properties": [
"prompt",
"image_url",
"negative_prompt",
"enable_prompt_expansion",
"enable_agentic_generation",
"agentic_max_iterations",
"agentic_samples_per_iteration",
"agentic_early_stop",
"image_size",
"num_frames",
"frames_per_second",
"num_inference_steps",
"guidance_scale",
"seed",
"enable_safety_checker",
"sync_mode"
],
"properties": {
"seed": {
"title": "Seed",
"description": "The same seed and prompt given to the same model version will produce the same video every time.",
"anyOf": [
{
"type": "integer"
},
{
"type": "null"
}
]
},
"num_frames": {
"title": "Num Frames",
"minimum": 5,
"maximum": 189,
"default": 189,
"description": "Number of frames to generate. More frames yield a longer video.",
"type": "integer"
},
"agentic_max_iterations": {
"title": "Agentic Max Iterations",
"minimum": 1,
"maximum": 3,
"default": 2,
"description": "Maximum agentic prompt stages when agentic generation is enabled.",
"type": "integer"
},
"enable_agentic_generation": {
"title": "Enable Agentic Generation",
"description": "Enable the iterative Cosmos agentic loop: prompt upsampling, candidate video generation, VLM critique of sampled frames, and prompt rewrite. Each candidate is a full render, so this is substantially slower and costlier than a single generation.",
"default": false,
"type": "boolean",
"const": false
},
"enable_prompt_expansion": {
"title": "Enable Prompt Expansion",
"description": "If true, the Cosmos3-Nano Reasoner (a VLM that sees the first frame) rewrites the prompt into the dense caption Cosmos3 was trained on. The app starts a local Reasoner by default, or uses COSMOS_PROMPT_UPSAMPLER_BASE_URL when configured. Falls back to the raw prompt if expansion fails.",
"default": false,
"type": "boolean",
"const": false
},
"guidance_scale": {
"title": "Guidance scale (CFG)",
"minimum": 0,
"maximum": 20,
"default": 6,
"description": "Classifier-free guidance scale. Higher values increase prompt adherence at the cost of diversity.",
"type": "number"
},
"num_inference_steps": {
"title": "Num Inference Steps",
"minimum": 1,
"maximum": 50,
"default": 28,
"description": "Number of denoising steps. More steps yield higher quality but take longer.",
"type": "integer"
},
"agentic_early_stop": {
"title": "Agentic Early Stop",
"description": "Stop the agentic loop early when the critic score clears the strict quality threshold.",
"default": true,
"type": "boolean"
},
"image_url": {
"title": "Image Url",
"description": "URL of the conditioning first-frame image for the video.",
"type": "string",
"examples": [
"https://storage.googleapis.com/falserverless/example_inputs/hunyuan_i2v.jpg"
]
},
"sync_mode": {
"title": "Sync Mode",
"description": "If `True`, the video is returned as a data URI and the output data won't be available in the request history.",
"default": false,
"type": "boolean"
},
"prompt": {
"title": "Prompt",
"maxLength": 4096,
"description": "Text prompt describing the motion and scene of the video to generate.",
"minLength": 1,
"type": "string",
"examples": [
"The camera slowly pushes in as the subject turns their head toward the light, hair drifting in a gentle breeze, dust motes floating through warm afternoon sun."
]
},
"negative_prompt": {
"title": "Negative Prompt",
"maxLength": 2048,
"description": "Content to steer the generation away from (artifacts, unwanted motion). Defaults to NVIDIA's recommended i2v negative prompt; pass an empty string to disable.",
"default": "The 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 moir\u00e9 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.",
"type": "string",
"examples": [
"The 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 moir\u00e9 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."
]
},
"enable_safety_checker": {
"title": "Enable Safety Checker",
"description": "Enable content moderation for the input prompt and image. Disabling it requires account authorization; unauthorized requests are always checked.",
"default": true,
"type": "boolean"
},
"image_size": {
"title": "Image Size",
"description": "The size of the generated video. The request is clamped and snapped to the nearest supported NVIDIA tier (256p/480p/720p) and aspect ratio.",
"examples": [
{
"height": 480,
"width": 832
}
],
"default": {
"height": 480,
"width": 832
},
"anyOf": [
{
"x-webot-order-properties": [
"width",
"height"
],
"type": "object",
"properties": {
"width": {
"exclusiveMinimum": 0,
"default": 512,
"maximum": 14142,
"type": "integer",
"title": "Width",
"description": "The width of the generated image."
},
"height": {
"exclusiveMinimum": 0,
"default": 512,
"maximum": 14142,
"type": "integer",
"title": "Height",
"description": "The height of the generated image."
}
},
"title": "ImageSize"
},
{
"enum": [
"square_hd",
"square",
"portrait_4_3",
"portrait_16_9",
"landscape_4_3",
"landscape_16_9"
],
"type": "string"
}
]
},
"agentic_samples_per_iteration": {
"title": "Agentic Samples Per Iteration",
"minimum": 1,
"maximum": 3,
"default": 2,
"description": "Candidate videos to generate and judge per agentic iteration. The best candidate advances to the next rewrite stage.",
"type": "integer"
},
"frames_per_second": {
"title": "Frames Per Second",
"minimum": 4,
"maximum": 60,
"default": 24,
"description": "Frames per second of the output video.",
"type": "integer"
}
},
"required": [
"prompt",
"image_url"
],
"type": "object",
"allOf": [
{
"not": {
"required": [
"enable_agentic_generation"
],
"properties": {
"enable_agentic_generation": {
"const": true
}
}
},
"x-webot-message": "Disable agentic generation. Intermediate candidate billing cannot be verified."
},
{
"not": {
"required": [
"enable_prompt_expansion"
],
"properties": {
"enable_prompt_expansion": {
"const": true
}
}
},
"x-webot-message": "Disable prompt expansion for this video workflow until its separate fee is documented."
}
]
}Output (inside result.output) schema · fields, constraints and defaults
| Parameter | Type | Required | Default when omitted | Description / constraints |
|---|---|---|---|---|
video | object | Yes * | No documented default | The generated video. |
video.height | integer | null | No | No documented default | The height of the video |
video.file_size | integer | null | No | No documented default | The size of the file in bytes. |
video.file_name | string | null | No | No documented default | The name of the file. It will be auto-generated if not provided. |
video.content_type | string | null | No | No documented default | The mime type of the file. |
video.fps | number | null | No | No documented default | The FPS of the video |
video.num_frames | integer | null | No | No documented default | The number of frames in the video |
video.width | integer | null | No | No documented default | The width of the video |
video.url | string | Yes * | No documented default | The URL where the file can be downloaded from. |
video.duration | number | null | No | No documented default | The duration of the video |
seed | integer | Yes * | No documented default | The seed used for generation. |
{
"title": "VideoOutput",
"x-webot-order-properties": [
"video",
"seed"
],
"properties": {
"video": {
"title": "VideoFile",
"x-webot-order-properties": [
"url",
"content_type",
"file_name",
"file_size",
"width",
"height",
"fps",
"duration",
"num_frames"
],
"properties": {
"height": {
"title": "Height",
"description": "The height of the video",
"anyOf": [
{
"type": "integer"
},
{
"type": "null"
}
]
},
"file_size": {
"title": "File Size",
"description": "The size of the file in bytes.",
"examples": [
4404019
],
"anyOf": [
{
"type": "integer"
},
{
"type": "null"
}
]
},
"file_name": {
"title": "File Name",
"description": "The name of the file. It will be auto-generated if not provided.",
"examples": [
"z9RV14K95DvU.png"
],
"anyOf": [
{
"type": "string"
},
{
"type": "null"
}
]
},
"content_type": {
"title": "Content Type",
"description": "The mime type of the file.",
"examples": [
"image/png"
],
"anyOf": [
{
"type": "string"
},
{
"type": "null"
}
]
},
"fps": {
"title": "Fps",
"description": "The FPS of the video",
"anyOf": [
{
"type": "number"
},
{
"type": "null"
}
]
},
"num_frames": {
"title": "Num Frames",
"description": "The number of frames in the video",
"anyOf": [
{
"type": "integer"
},
{
"type": "null"
}
]
},
"width": {
"title": "Width",
"description": "The width of the video",
"anyOf": [
{
"type": "integer"
},
{
"type": "null"
}
]
},
"url": {
"title": "Url",
"description": "The URL where the file can be downloaded from.",
"type": "string"
},
"duration": {
"title": "Duration",
"description": "The duration of the video",
"anyOf": [
{
"type": "number"
},
{
"type": "null"
}
]
}
},
"required": [
"url"
],
"type": "object",
"description": "The generated video."
},
"seed": {
"title": "Seed",
"description": "The seed used for generation.",
"type": "integer",
"examples": [
1143
]
}
},
"required": [
"video",
"seed"
],
"type": "object"
}Errors and limits
401 invalid/expired/revoked key · 403 key scope or domain mismatch · 404 unavailable model or unowned request · 409 quote changed, balance, concurrency or idempotency conflict · 413 JSON exceeds 2 MB · 415 wrong content type · 422 invalid fields/media · 429 rate limit · 503 temporary/ambiguous submission. Respect Retry-After. Current limits per account: 12 quotes/submissions, 120 result reads and 8 stream connections per minute. One submission preparation at a time per account. Do not automatically regenerate failed jobs.
Your requests
NVIDIA Cosmos - Image to Video (nvidia/cosmos-3-super/image-to-video) · Your account only
Select Requests to load your history.
| Request ID | Created at (IST) | Status | Credits used | Refunded | Processing time | Report |
|---|
Your model analytics
NVIDIA Cosmos - Image to Video (nvidia/cosmos-3-super/image-to-video) · Request dates in IST
Select Analytics to load your metrics.
Request counts
Daily requests by their current status
Request outcomes
Completed, failed and pending requests
Credits used
Daily recorded charges before refunds
Processing time
Daily average for completed requests with recorded timing
Processing percentiles
Completed requests only · Time from submission to recorded completion
How to use NVIDIA Cosmos — Image to Video (nvidia/cosmos-3-super/image-to-video)
Give this model a picture to use as a starting point or visual guide. It creates a video based on that picture and the movement instructions supported here.
Required inputs
- Prompt: Your instructions to the model. Say what you want, include the important details and avoid contradictory directions. For a video, describe the action as well as the scene.
- Image Url: The input picture. Use a clear image in one of the accepted formats. If the form offers upload, select a file there; if it asks for a URL, use an accessible HTTPS file link.
Workflow
- Prepare your source material and fill the required inputs above. Review the supported settings for this model.
- Sign in to calculate the credit estimate. Review the estimate before choosing Generate.
- Keep this model page open while the request runs. Completed results arrive in the result area; use the available player or download controls to review the output.
When to use this workflow
- Bring a product picture or illustration to life.
- Try movement ideas using an existing picture.
Key settings
- Prompt
- Your instructions to the model. Say what you want, include the important details and avoid contradictory directions. For a video, describe the action as well as the scene. Minimum characters: 1. Maximum characters: 4096
- Image Url
- The input picture. Use a clear image in one of the accepted formats. If the form offers upload, select a file there; if it asks for a URL, use an accessible HTTPS file link.
- Negative Prompt
- Describe things you would prefer the model to avoid, such as unwanted blur or objects. This guides the result but is not a guarantee that those things will never appear. Maximum characters: 2048
- Enable Prompt Expansion
- Let the model expand or improve your written instructions before generation. This can add useful detail, but may also change the interpretation. Turn it off, where allowed, if you need your original wording followed more literally. Fixed value: false
- Enable Agentic Generation
- Controls for a supported multi-step or conversational workflow. Iterations are repeated attempts or steps, early stopping ends when its rule is met, and memory supplies prior context. More steps can add time and usage. Fixed value: false
- Agentic Max Iterations
- Controls for a supported multi-step or conversational workflow. Iterations are repeated attempts or steps, early stopping ends when its rule is met, and memory supplies prior context. More steps can add time and usage. Minimum: 1. Maximum: 3