Poldion Self-HostedDeutsch
Part of Poldion Self-Hosted

ImageGen

Images and videos on your own GPU: ComfyUI with ready-made workflows, a phone-friendly web app with users and an OpenAI-compatible image API.

  • ComfyUI
  • Python
  • FastAPI
  • AMD ROCm
  • Qwen-Image
  • Wan 2.2
  • OpenAI Images API
ImageGen web app on a phone with workflow selection and personal gallery

From idea to image

Pick a workflow, type a short idea, done. "Expand" turns it into a detailed prompt via the language model. Variables such as [day|night] and series create many variants at once.

Every user only sees their own images. With a personal API key, other programs use ImageGen too, for example Tabletop Forge or Open WebUI.

The screenshot shows the German interface – the app itself can be switched to English.

Made for game nights

A dedicated D&D tab creates illustrations in the style of the rulebooks, round tokens with a transparent edge (also from your own picture), battle maps and location maps in VTT style.

Example made with ImageGen: Scene
Scene
Example made with ImageGen: Token
Token
Example made with ImageGen: Battle map
Battle map

Features

Ready-made workflows

Anime, realistic, drawn, editing, same character in a new scene, text and image to video.

D&D styles

Illustration, location, token, battle map and location map – with a fixed style and matching size.

Prompt expansion

Short idea → detailed English prompt, with rules per workflow.

Variables & series

Nested variables rolled evenly, series of any length that can be stopped at any time.

Users & API

Setup wizard, permissions per user, ZIP download, API keys, OpenAI-compatible.

Memory under control

Really frees ComfyUI's RAM so language and image models fit side by side.

Working together

ImageGen gives Tabletop Forge tokens, portraits, maps and scene images. The OllamaProxy helps with prompt expansion and attaches ImageGen as a module: it shows ComfyUI's memory and frees it when there is nothing to do.

Quick start

git clone https://github.com/Poldion/ImageGen.git && cd ImageGen
mkdir -p ~/imagegen/{models,output,input,custom_nodes,extras,comfy-user,webapp-data,cache}
cp imagegen.env.example ~/imagegen/imagegen.env   # adjust paths, PUID/PGID, GPU groups
docker compose --env-file ~/imagegen/imagegen.env up -d --build

Then open http://<host>:8190/ and create the admin account. You download the models yourself; MODELLE.md lists which file goes where.

License

MIT for the web app and workflows, GPL-3.0 for the ComfyUI extensions. Models are not included, several of them are licensed for non-commercial use only. The repository contains no adult content. Your own extensions can be added and are enabled per user.