gpt-image-1-mini

A text description of the required image. The more detailed the description, the better the expected result.
Character length limits:

• GPT-Image series models: maximum 32000 characters
• DALL-E-2: maximum 1000 characters
• DALL-E-3: maximum 4000 characters

Writing suggestions:

• Include the subject (e.g., “baby sea otter”), action (e.g., “playing”), scene (e.g., “in the ocean waves”)
• Specify the style (e.g., “watercolor style”), lighting (e.g., “soft light”), and color (e.g., “blue and white”)
• Avoid vague descriptions; use specific words to improve generation accuracy.

The generated image size varies depending on the model:

• GPT-Image series: 1024x1024, 1536x1024 (landscape), 1024x1536 (portrait), auto (automatic adaptation)
• DALL-E-2: 256x256, 512x512, 1024x1024
• DALL-E-3: 1024x1024, 1792x1024 (landscape), 1024x1792 (portrait)

Image background settings, applicable only to GPT-Image series models:

• transparent: Transparent background (PNG/WEBP format only supported)
• opaque: Opaque background
• auto: Automatic judgment (default, intelligently set according to prompt word scenario)

Note: When transparent is selected, output_format must be set to png or webp

Content moderation level, controlling the compliance of generated images:

• low: lenient moderation, allowing slight creative expression
• auto: automatic moderation (default), balancing compliance and creativity

Number of images generated:

• Value range: 1-10 (GPT-Image/DALL-E-2)
• DALL-E-3 only supports n=1

Note: The more images generated, the more tokens are consumed and the longer the generation time

The generated image quality. ‘auto’ (default) will automatically select the best quality for the given model.

• GPT-Image model: Supports high, medium, and low modes.
• DALL-E-3: Supports hd and standard image quality.
• DALL-E 2: Only standard can be selected.

Image compression level (GPT-Image only, default 100):

• Value range: 0-100 (percentage)
• 0: No compression (largest file size, best quality)
• 100: Maximum compression (smallest file size, quality may be compromised)

Only applicable to webp/jpeg formats, png format does not support compression

Image output format (GPT-Image only):

• png: Supports transparent backgrounds, lossless compression
• jpeg: Good compatibility, smaller file size
• webp: Balances compression and quality, supports transparency

When selecting transparent background, png or webp format must be used

Number of partial images generated (GPT-Image only):

• Value range: 0-3
• 0: Do not return partial images, return the complete image after generation (default)
• 1-3: Return partial image previews in stages, and finally return the complete image

Only effective when stream=true, suitable for scenarios that require fast preview

Enable streaming generation (GPT-Image only):

• false (default): Return all data at once after generation
• true: Return image data in stages as streaming events

Combined with the partial_images parameter, progressive image preview can be achieved

Original image to be edited (required):

• GPT-Image Series (gpt-image-1/1-mini/1.5):
  • Supported formats: PNG, WEBP, JPG
  • Single file size: ≤50MB
  • Quantity limit: up to 16 pieces
• DALL-E 2：
  • Supported format: only PNG (square)
  • Single image size: ≤4MB
  • Quantity limit: Only 1 piece

Note: When uploading multiple images, the mask only applies to the first image

Text description of the editing effect (required):

• Describe details such as the content, style, color, etc. to be replaced
• Character length limit:
  • GPT-Image Series: Up to 32,000 characters
  • DALL-E 2: Up to 1000 characters
• Writing Suggestion: Describe clearly and specifically in combination with the original image scene

gpt-image-1、gpt-image-1-mini、gpt-image-1.5

Mask Image (Optional):

• Function: Transparent area (alpha=0) indicates the position in the original image that needs to be edited
• Requirements:
  • Format: PNG only (supports transparency)
  • Size: ≤4MB (all models)
  • Dimensions: Must be exactly the same as the first original image
  • Quantity: Only 1 piece; when there are multiple original images, it only applies to the first one
• Applicable scenarios: Precisely locating the editing area, such as replacing a specific object in an image.

Quality level of the generated image:

• GPT-Image Series: Supports auto/high/medium/low
• DALL-E 2: Only supports standard

Size of the generated image:

• GPT-Image Series: 1024x1024, 1536x1024, 1024x1536, auto
• DALL-E 2：256x256、512x512、1024x1024

Number of edited images generated:

• Value range: 1-10
• The greater the quantity, the more Tokens are consumed, and the longer the generation time

Image background settings (only applicable to GPT-Image model):

• transparent: Transparent background (only supported in PNG/WEBP formats)
• opaque: opaque background
• auto: intelligently set based on the prompt and the original image

Output image format (only applicable to GPT-Image model):

• When choosing a transparent background, you need to use the png or webp format

Image Compression Level (only applicable to GPT-Image model):

• Value range: 0-100 (percentage)
• 0: No compression (largest file size, best quality)
• 100: Maximum compression (smallest file size, quality may be compromised)
• Only applicable to webp or jpeg formats; png format does not support compression

Partial Image Generation Quantity (only applicable to GPT-Image model):

• Value range: 0-3
• 0: Do not return partial images; return the complete image after generation is completed (default)
• 1-3: Return partial image previews in stages, and finally return the complete image
• Only takes effect when stream=true

Whether to enable streaming generation (only applicable to GPT-Image model):

• false (default): Return all at once after generation is completed
• true: Return image data in stages as streaming events