kadevin/ilab-conjure

面向 GPT-image-2 的 AI 图片生成 WebUI 工作台，支持 Codex Responses 与 OpenAI 兼容 API 接入，内置公用图库、多类型 Chip 快捷引用、提示词模板、多任务并发和本地队列管理。An AI image generation WebUI workbench for GPT-image-2 with Codex Responses and OpenAI-compatible API support, shared gallery references, multi-type quick chips, prompt templates, concurrent tasks, and local queue management.

iLab CONJURE – a local, multi‑model AI image‑generation workstation

What it is – iLab CONJURE is a Python‑based desktop/web application that lets you generate, edit and organise AI‑created images. It bundles a FastAPI web UI, a command‑line client and a thin update‑helper, and it can talk to several image‑generation back‑ends:

GPT‑Image (including the 2, 2.5 Flare and Sunburst variants)
Google Gemini (Nano Banana Pro/2.1/2/Lite)
Codex image and response endpoints (via local OAuth or an OpenAI‑compatible proxy)

All models are accessed through a single, extensible model catalogue, so you can switch providers without leaving the UI.

Key features (as described in the README)

Category	Highlights
Model support	GPT‑Image & Gemini in one directory; automatic discovery of remote model names; OpenAI‑compatible API or direct OAuth for Codex.
Task handling	Concurrent generation, local queue, cancel‑and‑keep‑partial results, per‑task retry, multi‑image output, failure handling.
History & library	SQLite‑backed paginated history, searchable/filterable grid‑/list view, tagging, bulk export (ZIP of images ± prompts).
Prompt editor	Chip system – `@` image chips, `#` colour chips, `~` prompt‑snippet chips; template library; auto‑lock of generation parameters.
Image editor	Layer‑based editor (Konva), supports inserting reference images, scaling, free‑transform, erasing, transparent‑background detection.
Web UI	Responsive FastAPI UI, dark/light/system theme, language selector (10+ locales), LAN sharing mode, no login required for trusted networks.
CLI	`codex_image generate …` for scripting, supports prompts, reference images, masks, dry‑run.
Packaging	Portable zip (bundled CPython + deps) and signed macOS DMG with auto‑update helper; data kept in `data/` and migrated on upgrade.
Configuration	API supplier cards (base URL, key, model, concurrency limit), network proxy settings, per‑request timeout/retry, backup & restore of chips, templates, gallery, settings.
Security / privacy	No automatic GitHub calls, explicit opt‑in for LAN sharing, OAuth mode warned as unstable, AGPL‑3.0 license.

Quick start (from the README)

# clone & set up a virtual‑env
git clone https://github.com/kadevin/ilab-conjure.git
cd ilab-conjure
python3 -m venv .venv
.venv/bin/python -m pip install --require-hashes -r requirements-webui.txt

# launch the UI (macOS example)
open "Start WebUI.command"
# or manually:
.venv/bin/python -m codex_image.webui.server codex_image.webui.app:app \
    --port 8787 --no-access-log

Then open http://127.0.0.1:8787/ in a browser. The UI guides you to add an API supplier (OpenAI‑compatible endpoint or Codex OAuth) and start generating images.

Command‑line usage example

.venv/bin/python -m codex_image generate \
    --prompt "A clean product photo of a ceramic mug" \
    --out output/mug.png

Run -‑help for the full list of options (reference images, masks, dry‑run, etc.).

Installation packages

Standard macOS DMG – Apple Silicon (arm64) or Intel (x64).
Standard Windows ZIP – contains an .exe launcher.
Portable zip – “extract‑and‑run” style, includes its own CPython and dependencies; useful for debugging or for a ComfyUI‑like experience.
All packages embed a signed latest.json manifest; the macOS app can check for updates from the menu bar.

License

The code is released under GNU AGPL‑v3‑only. If you modify the software and provide it as a network service you must also make the source of those modifications available under the same license.

Who might find it useful?

Artists or designers who want a self‑hosted, privacy‑first image‑generation UI.
Teams that need a shared LAN workstation with a common gallery, history and prompt templates.
Developers looking for a CLI‑driven automation layer on top of GPT‑Image, Gemini or Codex.
Anyone who prefers a single UI for multiple providers rather than juggling separate web portals.

Where to get help / custom work

The author invites contact via WeChat for AI‑coding advice, workflow consulting or bespoke development (e.g., internal automation dashboards, AI‑powered chat agents, etc.).