Iris Web UI

A Streamlit interface for the iris image generation binary. Exposes all generation settings in a browser-based UI with persistent task tracking.

Setup

Requires Python 3.10+ and the iris binary already compiled (see the main README for build instructions).

cd webui
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
streamlit run app.py

The app opens at http://localhost:8501. It auto-detects model directories in the project root.

Features

• All generation modes: text-to-image and image-to-image with up to 16 reference images
• Every CLI option exposed: dimensions, steps, seed, guidance, timestep schedule (sigmoid / linear / power curve), force-base, no-mmap
• Real-time progress: live progress bar with phase labels (loading → encoding → denoising steps → decode → save)
• Task persistence: tasks survive page reloads and server restarts — backed by SQLite with WAL journaling
• Cancel running tasks: stop a generation mid-flight from the UI
• Image gallery: browse completed generations with prompt, dimensions, timing, and seed metadata
• Gallery → reference: click "Use as ref" on any generated image to feed it back as input for a new generation — the mode switches to Image to Image automatically
• Save / Delete: download images or clean up directly from the gallery
• Format-agnostic uploads: any image format (JPEG, PNG, WebP, TIFF, BMP) is converted to PNG before passing to iris

Architecture

app.py      Streamlit UI — sidebar settings, prompt input, gallery, active tasks
store.py    SQLite persistence — task lifecycle (create → running → completed/failed)
runner.py   Background execution — spawns the binary in a thread, parses stderr for progress

The app launches iris as a subprocess per task. A monitor thread reads stderr with non-blocking I/O and maps the output phases to a 0–100% progress range:

Range	Phase
0–5%	Loading model
5–20%	Text encoding
20–90%	Denoising steps
90–97%	VAE decode
97–100%	Saving

On startup, store.init() checks for orphaned tasks (process no longer alive) and marks them as failed. The init call is cached via @st.cache_resource so Streamlit's auto-reloader doesn't re-trigger it.

File Layout

webui/
├── app.py              Main application
├── store.py            Task database
├── runner.py           Binary execution + progress parsing
├── requirements.txt    Python dependencies
├── .gitignore
├── data/               SQLite database (gitignored)
├── outputs/            Generated images (gitignored)
└── uploads/            Uploaded reference images (gitignored)