P-Image

Model ID

Use with the Workers AI binding, the AI Gateway REST API, or the Unified API.

Calling the provider directly? Use the provider's model name instead, including with provider-native endpoints or the provider's own SDK. For example, Anthropic's name for anthropic/claude-opus-4.8 is claude-opus-4-8.

Third-party

Pruna's P-Image is an ultra-fast text-to-image model with automatic prompt enhancement and 2-stage refinement, combining exceptional speed with high-quality output and flexible aspect ratios.

Model Info
More information	link ↗
Pricing	Per image$0.005

Usage

const response = await env.AI.run(
  'pruna/p-image',
  {
    prompt: 'A majestic lion standing on a rocky cliff at sunset, photorealistic, 4k',
    aspect_ratio: '16:9',
  },
)
console.log(response)

curl https://api.cloudflare.com/client/v4/accounts/$CLOUDFLARE_ACCOUNT_ID/ai/run \
  --header "Authorization: Bearer $CLOUDFLARE_API_TOKEN" \
  --header "Content-Type: application/json" \
  --data '{
  "model": "pruna/p-image",
  "input": {
    "prompt": "A majestic lion standing on a rocky cliff at sunset, photorealistic, 4k",
    "aspect_ratio": "16:9"
  }
}'

{
  "state": "Completed",
  "result": {
    "image": "https://examples.aig.cloudflare.com/pruna/p-image/lion-at-sunset.jpg"
  },
  "gatewayMetadata": {
    "keySource": "Unified"
  }
}

Examples

Reading Nook — Square-format generation with prompt upsampling.

const response = await env.AI.run(
  'pruna/p-image',
  {
    prompt: 'A cozy reading nook by a rainy window, warm lighting, detailed illustration',
    aspect_ratio: '1:1',
    prompt_upsampling: true,
  },
)
console.log(response)

curl https://api.cloudflare.com/client/v4/accounts/$CLOUDFLARE_ACCOUNT_ID/ai/run \
  --header "Authorization: Bearer $CLOUDFLARE_API_TOKEN" \
  --header "Content-Type: application/json" \
  --data '{
  "model": "pruna/p-image",
  "input": {
    "prompt": "A cozy reading nook by a rainy window, warm lighting, detailed illustration",
    "aspect_ratio": "1:1",
    "prompt_upsampling": true
  }
}'

{
  "state": "Completed",
  "result": {
    "image": "https://examples.aig.cloudflare.com/pruna/p-image/reading-nook.jpg"
  },
  "gatewayMetadata": {
    "keySource": "Unified"
  }
}

Parameters

prompt

stringrequiredText description of the image to generate. The model automatically enhances prompts for better results.

aspect_ratio

stringrequireddefault: 16:9enum: 1:1, 16:9, 9:16, 4:3, 3:4, 3:2, 2:3, customAspect ratio for the image. Use "custom" with width/height for exact dimensions.

width

integerminimum: 256maximum: 1440multipleOf: 16Custom width in pixels (256-1440, multiple of 16). Only used when aspect_ratio="custom".

height

integerminimum: 256maximum: 1440multipleOf: 16Custom height in pixels (256-1440, multiple of 16). Only used when aspect_ratio="custom".

lora_weights

stringLoad LoRA weights. Supports HuggingFace URLs in the format huggingface.co/<owner>/<model-name>[/<file.safetensors>].

lora_scale

numberrequireddefault: 0.5minimum: -1maximum: 3How strongly the LoRA should be applied (-1 to 3).

hf_api_token

stringHuggingFace API token for accessing private LoRAs. This credential is forwarded verbatim to Pruna. It is only written to gateway request-body logs when the gateway-level collectLogPayload debug flag is explicitly enabled — it never appears in structured analytics logs.

prompt_upsampling

booleanrequireddefault: falseUpsample the prompt with an LLM for enhanced results.

seed

integerminimum: -9007199254740991maximum: 9007199254740991Random seed for reproducible generation.

disable_safety_checker

booleanrequireddefault: falseDisable safety checker for generated images.

image

stringformat: uriPresigned URL for the generated image.

API Schemas (Raw)

Input

Output