Image Generation

POST /v1/images/generations

Generate one or more images from a text prompt. Compatible (with extensions) with the OpenAI Images API.

Headers

Request body

Examples

curl -X POST https://api.quantized.us/v1/images/generations \
  -H "Authorization: Bearer sk-quantized-YOUR-KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "dall-e-3",
    "prompt": "A whimsical watercolor of a sleepy hedgehog under a mushroom",
    "size": "1024x1024",
    "quality": "standard"
  }'

curl -X POST https://api.quantized.us/v1/images/generations \
  -H "Authorization: Bearer sk-quantized-YOUR-KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-image-1",
    "prompt": "A logo on a fully transparent background, single hedgehog character",
    "quality": "medium",
    "background": "transparent",
    "output_format": "png"
  }'

import base64
from openai import OpenAI

client = OpenAI(
    api_key="sk-quantized-YOUR-KEY",
    base_url="https://api.quantized.us/v1",
)

resp = client.images.generate(
    model="dall-e-3",
    prompt="A whimsical watercolor of a sleepy hedgehog",
    size="1024x1024",
    quality="standard",
)
img_bytes = base64.b64decode(resp.data[0].b64_json)
with open("hedgehog.png", "wb") as f:
    f.write(img_bytes)

import base64
import httpx

response = httpx.post(
    "https://api.quantized.us/v1/images/generations",
    headers={"Authorization": "Bearer sk-quantized-YOUR-KEY"},
    json={
        "model": "gpt-image-1",
        "prompt": "An overhead view of an open notebook with a pen",
        "size": "1024x1024",
        "quality": "low",
    },
    timeout=180,
)
data = response.json()
img_bytes = base64.b64decode(data["data"][0]["b64_json"])
print(f"image: {len(img_bytes)} bytes, watermark: {data['data'][0]['watermark']}")

Response

{
  "created": 1777975004,
  "model": "dall-e-3",
  "data": [
    {
      "b64_json": "iVBORw0KGgoAAAANSUhEUgAA...",
      "revised_prompt": "A whimsical watercolor illustration ...",
      "seed": null,
      "watermark": "none",
      "flagged": false
    }
  ],
  "usage": {
    "images": 1,
    "input_tokens": null,
    "output_tokens": null,
    "total_tokens": null,
    "megapixels": null,
    "credits_used": 640000,
    "credits_remaining": 99360000
  }
}

Response fields

Watermarking

Different providers/models embed different provenance metadata. The watermark enum in each data[] entry lets clients disclose this to end users if needed.

Per-model constraints

Each model has stricter limits than the serializer’s generic [1, 10] range for n and unrestricted size. The router does not enforce these — they’re handled by upstream:

Discovery via /v1/models

Image-gen-capable models are listed in GET /v1/models alongside chat models. Filter for image generation by:

• supported_features contains "image_generation" (recommended, semantic):

  image_models = [m for m in models if "image_generation" in m.get("supported_features", [])]
  

• output_modality.image == true — image models declare image output; LLMs and embedding models do not.

The catalog also exposes per-model pricing in cost (JSONB — shape varies by model: per_image, per_image_by_size, per_image_by_size_quality, or per_token).

Providers

Routing is resolved per-model from the catalog. The default for image_generation is OpenAI, but if you request amazon.nova-canvas-v1:0 (once active) the router automatically picks the bedrock provider — no X-Quantized-Provider header needed.

# Once Bedrock is active, this routes through Bedrock automatically:
curl -X POST https://api.quantized.us/v1/images/generations \
  -H "Authorization: Bearer sk-quantized-YOUR-KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "amazon.nova-canvas-v1:0", "prompt": "a hedgehog"}'

Errors