# Cosmos 3 Super API

> Text-to-image generation with prompt expansion and an optional agentic candidate-comparison mode

Creates 1 to 4 images from a `prompt` of up to 4096 characters. `image_size` edges are clamped to 512-1280 px. `enable_agentic_generation` produces several candidates and refines the prompt between rounds, with up to 3 iterations. Also has `negative_prompt`, `guidance_scale` up to 20, `num_inference_steps` up to 50, and jpeg or png output.

- Model id: `nvidia/cosmos-3-super/text-to-image`
- Made by: NVIDIA
- Category: Text to Image
- Output: image
- Price per run: from $0.04 / run of output image (cheapest run $0.04 (40 API credits))
- Run: `POST https://api.doitong.com/v1/run/nvidia/cosmos-3-super/text-to-image`
- Quote: `POST https://api.doitong.com/v1/quote/nvidia/cosmos-3-super/text-to-image`
- Docs page: https://doitong.com/developers/models/nvidia/cosmos-3-super/text-to-image

## Pricing

Price: **from $0.04 / run of output image** (the lowest tier; the quote prices your exact input). Price per run: the input files are measured before the run (length, resolution, frame rate). The quote measures the same files and is exactly what the run charges.

The full example input below costs $0.04 = 40 API credits (1 run × $0.04).

Prices are the upstream price in USD, paid from the API balance in API credits (1000 API credits = 1 USD). `POST https://api.doitong.com/v1/quote/nvidia/cosmos-3-super/text-to-image` with the same JSON body returns the exact price for that input; the quote is exactly what a run charges. Invalid input is never charged and failed or canceled requests are refunded automatically.

## Input parameters

Send either the canonical name or the native field name (not both). Canonical names are the same across models.

| Parameter | Canonical name | Type | Required | Default | Allowed values / range | Description |
|---|---|---|---|---|---|---|
| `prompt` |  | string | yes |  | 1 – 4096 characters | Text description of the desired output image. |
| `negative_prompt` |  | string | no | `` | ≤ 1024 characters | Negative guidance identifying unwanted colors, objects, or artifacts for generation to avoid. |
| `enable_prompt_expansion` |  | boolean | no | `false` |  | Pre-generation prompt enhancement through OpenRouter, producing the dense structured-JSON representation used to train Cosmos3 when enabled. If enhancement fails, image generation proceeds with the unmodified prompt. |
| `enable_agentic_generation` |  | boolean | no | `false` |  | Automatic prompt-refinement loop aimed at closer adherence to the original request. Each round generates multiple image candidates and compares them, with prompt revisions between rounds. |
| `agentic_max_iterations` |  | integer | no | `2` | 1 – 3 | Refinement round limit for generation with agentic mode enabled. |
| `agentic_samples_per_iteration` |  | integer | no | `2` | 1 – 3 | Per-iteration image candidate count for the agentic loop, determining how many images compete in evaluation before the best proceeds to prompt rewriting. |
| `agentic_early_stop` |  | boolean | no | `true` |  | Early-stop control that treats a strong image-to-prompt match as sufficient to finish without further iterations. |
| `image_size` |  | object {width, height} \| enum (string) | no | `square_hd` | `square_hd`, `square`, `portrait_4_3`, `portrait_16_9`, `landscape_4_3`, `landscape_16_9` | Set the output image dimensions; clamping keeps each edge between 512 and 1280px, with edge lengths restricted to multiples of 16. |
| `num_inference_steps` |  | integer | no | `28` | 1 – 50 | Denoising step count, trading longer processing for better quality as the count increases. |
| `guidance_scale` |  | number | no | `4` | 0 – 20 | Strength of classifier-free guidance. |
| `num_images` | `num_outputs` | integer | no | `1` | 1 – 4 | Image count requested for generation. |
| `seed` |  | integer \| null | no |  |  | Seed for repeatable image generation. The output image remains unchanged across runs unless the seed, prompt, or model version changes. |
| `enable_safety_checker` |  | boolean | no | `true` |  | Content moderation switch covering both prompt input and image output. Accounts must be authorized to bypass moderation; otherwise checks remain mandatory regardless of this setting. Any image deemed unsafe is replaced with a black image. |
| `output_format` |  | enum (string) | no | `jpeg` | `jpeg`, `png` | File format for the resulting image. |

## Example input

Required parameters only:

```json
{
  "prompt": "A red fox trots through fresh snow at sunrise, soft golden light, shallow depth of field"
}
```

A fuller example:

```json
{
  "prompt": "A red fox trots through fresh snow at sunrise, soft golden light, shallow depth of field"
}
```

## Run it

Set `DOITONG_API_KEY` to a key from https://doitong.com/developers/dashboard#keys. `?wait=20` returns the finished request in the same call when it is ready within 20 seconds; otherwise poll `GET /requests/{id}` (it also accepts `?wait=`), or pass `webhook_url`.

### curl

```bash
# Run the model; wait up to 20 s for the finished result in the same call
curl -X POST "https://api.doitong.com/v1/run/nvidia/cosmos-3-super/text-to-image?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"
  }'

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

### JavaScript

```javascript
const API = "https://api.doitong.com/v1";
const headers = {
  "Authorization": "Bearer " + process.env.DOITONG_API_KEY,
  "Content-Type": "application/json",
};

// Run the model; wait up to 20 s for the finished result in the same call
let res = await fetch(API + "/run/nvidia/cosmos-3-super/text-to-image?wait=20", {
  method: "POST",
  headers,
  body: JSON.stringify({
    "prompt": "A red fox trots through fresh snow at sunrise, soft golden light, shallow depth of field"
  }),
});
let request = await res.json();
if (!res.ok) throw new Error(request.error.code + ": " + request.error.message);

// Not finished yet: long-poll until it is
while (request.status === "queued" || request.status === "running") {
  res = await fetch(API + "/requests/" + request.id + "?wait=20", { headers });
  request = await res.json();
}
if (request.status !== "succeeded") throw new Error(JSON.stringify(request.error));
console.log(request.outputs.map((o) => o.url), request.cost.credits + " API credits ($" + request.cost.usd + ")");
```

### Python

```python
import os
import requests

API = "https://api.doitong.com/v1"
headers = {"Authorization": "Bearer " + os.environ["DOITONG_API_KEY"]}

payload = {
    "prompt": "A red fox trots through fresh snow at sunrise, soft golden light, shallow depth of field",
}

# Run the model; wait up to 20 s for the finished result in the same call
r = requests.post(API + "/run/nvidia/cosmos-3-super/text-to-image", params={"wait": 20}, json=payload, headers=headers)
request = r.json()
if not r.ok:
    raise RuntimeError(request["error"]["code"] + ": " + request["error"]["message"])

# Not finished yet: long-poll until it is
while request["status"] in ("queued", "running"):
    request = requests.get(API + "/requests/" + request["id"], params={"wait": 20}, headers=headers).json()
if request["status"] != "succeeded":
    raise RuntimeError(request["error"])
print([o["url"] for o in request["outputs"]], request["cost"]["credits"], "API credits", "$" + request["cost"]["usd"])
```

### MCP

The MCP server at `https://api.doitong.com/v1/mcp` exposes the tools `search_models`, `get_model`, `quote`, `run`, `get_request`, `cancel_request`, `list_requests`, `create_upload` and `get_account`. A `run` call for this model:

```json
{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "run",
    "arguments": {
      "model_id": "nvidia/cosmos-3-super/text-to-image",
      "input": {
        "prompt": "A red fox trots through fresh snow at sunrise, soft golden light, shallow depth of field"
      },
      "wait_seconds": 20
    }
  }
}
```

## Response

Every call returns a request object. `status` is one of `queued`, `running`, `succeeded`, `failed`, `canceled`. `outputs` lists the generated files with URLs on our storage (kept forever, or 7 days with `?storage=temporary`; `storage` says which); `output` is the model's own result object with the same URLs.

```json
{
  "id": "req_8k2m4q7x1z9c3v5b6n0p2r4t",
  "object": "request",
  "model": "nvidia/cosmos-3-super/text-to-image",
  "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": 40,
    "usd": "0.04",
    "charged": 40,
    "refunded": 0
  },
  "input": {
    "prompt": "A red fox trots through fresh snow at sunrise, soft golden light, shallow depth of field"
  },
  "output": {
    "images": [
      {
        "url": "https://content.doitong.com/dev-api/2026/10/20/req_8k2m4q7x1z9c3v5b6n0p2r4t/0-output.jpeg",
        "content_type": "image/jpeg",
        "file_name": "0-output.jpeg",
        "file_size": 1843200,
        "width": 1024,
        "height": 768
      }
    ],
    "seed": 42,
    "has_nsfw_concepts": [
      false
    ]
  },
  "outputs": [
    {
      "type": "image",
      "url": "https://content.doitong.com/dev-api/2026/10/20/req_8k2m4q7x1z9c3v5b6n0p2r4t/0-output.jpeg",
      "content_type": "image/jpeg",
      "file_size": 1843200,
      "width": 1024,
      "height": 768,
      "stored": true
    }
  ],
  "error": null,
  "metrics": {
    "inference_seconds": 6.8
  },
  "urls": {
    "self": "https://api.doitong.com/v1/requests/req_8k2m4q7x1z9c3v5b6n0p2r4t",
    "cancel": "https://api.doitong.com/v1/requests/req_8k2m4q7x1z9c3v5b6n0p2r4t/cancel"
  }
}
```

### Output fields (`output`)

| Field | Type | Description |
|---|---|---|
| `images` | array of object {url, content_type, file_name, file_size, width, height} | Images produced by the model. |
| `seed` | integer | Seed value applied during generation. |
| `has_nsfw_concepts` | array of boolean | Safety-checker flag status for each generated image. |

## Related models

- [Cosmos 3 Super Image to Video](https://doitong.com/developers/models/nvidia/cosmos-3-super/image-to-video.md) (`nvidia/cosmos-3-super/image-to-video`): from $0.05 / second of output video
- [FLUX.1 \[dev\]](https://doitong.com/developers/models/flux/dev.md) (`flux/dev`): $0.025 / megapixel
- [FLUX.1 \[schnell\]](https://doitong.com/developers/models/flux/schnell.md) (`flux/schnell`): $0.003 / megapixel
- [FLUX.2 \[dev\]](https://doitong.com/developers/models/flux-2.md) (`flux-2`): $0.012 / megapixel
- [FLUX.2 \[pro\]](https://doitong.com/developers/models/flux-2-pro.md) (`flux-2-pro`): $0.015 / megapixel of output image
- [GPT Image 2.5 Text to Image](https://doitong.com/developers/models/openai/gpt-image-2.5/flare/text-to-image.md) (`openai/gpt-image-2.5/flare/text-to-image`): from $0.008 / image
- [Ideogram V4.5](https://doitong.com/developers/models/ideogram/v4.5.md) (`ideogram/v4.5`): from $0.03 / image
- [Nano Banana 2](https://doitong.com/developers/models/nano-banana-2.md) (`nano-banana-2`): from $0.06 / image

## Links

- HTML page: https://doitong.com/developers/models/nvidia/cosmos-3-super/text-to-image
- All models and guides for AI agents: https://doitong.com/developers/llms.txt
- OpenAPI 3.1 with this model's typed request body: https://doitong.com/developers/openapi.json?model=nvidia/cosmos-3-super/text-to-image
- Errors: https://doitong.com/developers/docs/errors
- Model details as JSON: https://api.doitong.com/v1/models/nvidia/cosmos-3-super/text-to-image