gazeCOM: Open-Source Gaze- and VLM-Driven Outpainting on Top of ComfyUI
grebenshyo · reddit · 2026-09-08
A Reddit developer shared gazeCOM, an open-source, locally-run app for interactive and agent-driven outpainting built on ComfyUI.
Generation is organized around spatial points: drivers supply inputs forming a saliency distribution whose weighted center of mass positions a 1024×1024 generation frame; results are pasted back onto the canvas, turning outpainting into a navigable process. Two driver types:
- Interactive/procedural drivers convert gaze tracking, hand gestures, cursor movement, camera CV, or autonomous roaming into spatial points accumulating into a live heatmap;
- Vision-model drivers let a local VLM (the author uses Gemma) inspect the image and return an operative canvas coordinate rather than just a description.
Other features: automatic iterative generation and compositing, expandable canvas, weighted mutable pools for prompts and ComfyUI workflows, per-prompt LLM enhancement, visual memory and decision history, automatic workflow discovery with local overrides. The author candidly notes limits — local Gemma doesn't reason well about artistic decisions, so results are underwhelming and development is paused — and shares exported settings from the demo video for reproducibility.
More from coding & agent
- Agent verified 'success' after double-voting: a lesson in harness vs model bugs — TejasKumar_ · 2026-10-26
- diginomica: agentic automation hype is overdone, and 'The Hackening' of enterprise security has begun — jonerp · 2026-10-26
- 'Vibe verification': ~60% of AI-written formal specs prove the wrong property — basedjensen · 2026-10-26
- Devin launches Security suite powered by Claude Mythos with $100M+ credits — DevinAI · 2026-10-26
- HackBench injects reward hacks into agent trajectories to stress-test monitors — DanielKhashabi · 2026-10-26
- Hugging Face ships hf CLI updates: deduped cache saves 19GB, faster buckets, retryable GPU jobs — yb2698 · 2026-10-26