polym-eval-generate-gpt-image

name	polym-eval-generate-gpt-image
description	Generate or edit images with GPT Image via ByteDance AIDP. Use for gpt-image-1 text-to-image, multi-image edits, and masks.

GPT Image Generation & Editing

Generate or edit images using gpt-image-1 via ByteDance AIDP.

Two Modes

Mode	Description	Endpoint
generate (default)	Text-to-image	/images/generations
edit	Edit/composite reference images with a prompt	/images/edits

Pre-flight Check (IMPORTANT - Do this BEFORE generating)

Read the .env file and verify AIDP_API_KEY is set. If missing, ask the user for the key before proceeding.

Quick Start

# Text-to-image
python skills/polym-eval-generate-gpt-image/scripts/generate.py \
  --prompt "A serene mountain landscape at sunset"

# Image editing
python skills/polym-eval-generate-gpt-image/scripts/generate.py \
  --mode edit \
  --prompt "给水獭穿上宇航员服装" \
  --images photo.jpg

# Image editing with mask
python skills/polym-eval-generate-gpt-image/scripts/generate.py \
  --mode edit \
  --prompt "Replace background with a beach" \
  --images photo.jpg --mask mask.png

Required Configuration

# .env
AIDP_API_KEY=your_aidp_key

# Optional overrides
GPT_IMAGE_MODEL=gpt-image-1   # default model

Available Models

Model ID	Status	Notes
gpt-image-1	✅ Available (~21s)	Default
gpt-image-2	❌ Not yet available	Server error on AIDP

Complete Parameter Reference

Flag	Default	Description
--prompt / -p	required	Text prompt
--mode	generate	generate or edit
--model / -m	gpt-image-1	Model ID
--size / -s	1024x1024	1024x1024, 1536x1024, 1024x1536, auto
--quality / -q	low	low, medium, high, auto
--images / -i	—	Reference image paths/URLs (edit mode, required)
--mask	—	Mask image path (edit mode, optional)
--output / -o	generated/{timestamp}.png	Output file path
--n	1	Number of images to generate

AIDP API Details

Generate

POST https://aidp.bytedance.net/api/modelhub/online/v2/crawl/openai/images/generations?ak={ak}

Headers: Content-Type: application/json, api-key: {ak}

{
  "model": "gpt-image-1",
  "prompt": "A cute cat",
  "n": 1,
  "size": "1024x1024",
  "quality": "low"
}

Edit

POST https://aidp.bytedance.net/gpt/openapi/online/v2/crawl/openai/images/edits?ak={ak}

Headers: api-key: {ak} (multipart/form-data, no Content-Type header)

Form fields: image[] (one or more files), prompt, model, quality, size, n, mask (optional)

Response (both modes)

{
  "created": 1745742988,
  "data": [{"b64_json": "<base64 encoded image>"}],
  "usage": {"total_tokens": 284, "input_tokens": 12, "output_tokens": 272}
}

Programmatic Usage

from generate import generate_image, edit_image

# Generate
result = generate_image(
    prompt="A beautiful sunset",
    output_file="generated/sunset.png",
    model="gpt-image-1",
    size="1024x1024",
    quality="low",
)

# Edit
result = edit_image(
    prompt="给水獭穿上宇航员服装",
    image_paths=["photo.jpg"],
    output_file="generated/edited.png",
    model="gpt-image-1",
    quality="high",
    mask_path=None,
)

# Returns: {"gen_success": bool, "gen_image_path": str, "gen_url": str}

Dependencies

Pillow
requests
python-dotenv

Troubleshooting

no model permission: gpt-image-1

The AIDP AK does not have permission for this model. Apply on the ModelHub platform.

gpt-image-2 server error

gpt-image-2 is not yet available on AIDP. Use gpt-image-1.

Edit mode: image not updating

Provide a more specific prompt. For targeted edits, use a --mask image to indicate the region to modify.