acking-you/codex-bridge/gpt2api-image-generator

GPT2API Image Generator

Overview

Use this skill for direct image-generation and direct image-edit requests. It calls the configured GPT2API public image endpoints, stores generated images under .run/artifacts/generated/, and returns local artifact paths that can be sent with reply-current --image.

Do not answer image-generation requests with text only unless generation fails. Do not use GPT2API session APIs for this skill.

Config

The private config lives at:

.bridge-private/gpt2api-image.json

That directory is gitignored. The required fields are:

• base_url: GPT2API public API root, for example https://ackingliu.top/api/gpt2api
• api_key: bearer token for GPT2API

Optional fields:

• model: defaults to gpt-image-2
• size: defaults to 1024x1024
• n: defaults to 1
• response_format: defaults to b64_json
• timeout_seconds: defaults to 300
• endpoint_path: defaults to /images/generations
• edit_endpoint_path: defaults to /images/edits

Workflow

Direct Generation

1. Treat the user's visual request as the prompt. Keep concrete style, subject,

text, ratio, and size requirements if they were given.

1. Generate the image:

python3 skills/gpt2api-image-generator/generate_gpt2api_image.py \
  --prompt "一只穿宇航服的猫，月球表面，电影感光照"

Direct Edit

1. Recover or use the input image file path from the current request context.
2. Treat the user's requested change as the prompt. Keep size requirements if

they were given.

1. Edit the image:

python3 skills/gpt2api-image-generator/generate_gpt2api_image.py \
  --image .run/artifacts/input.png \
  --prompt "保留人物姿势，把背景改成雨夜霓虹街道" \
  --size 1024x1536

Reply

1. Parse the JSON output. For one image, use path; for multiple images, use

each value in paths.

1. Send each generated image with reply-current:

python3 skills/reply-current/reply_current.py \
  --context-file /abs/path/from/developer_instructions/contexts/group_111.json \
  --image .run/artifacts/generated/gpt2api_20260428T120000Z_1_abcd1234.png

Rules

• This skill exposes only two bot-level concepts: direct generation

(prompt + optional size/n) and direct edit (image + prompt + optional size/n).

• Use /images/generations for direct generation and /images/edits for

direct edit. Do not use /sessions, /messages, or product session APIs.

• Always send the generated image artifact itself. A text-only "done" reply is

not enough.

• If the user asks for more than one image, pass --n, capped by the script.
• If the user specifies a square or exact size, pass --size WIDTHxHEIGHT.
• If the user asks to edit an existing image, pass that image with --image.
• Keep the prompt focused on the requested image. Do not prepend chat markers,

routing notes, or commentary.

• Do not expose the API key in replies, logs, or copied command output.