Open Vision Banana: open recipe turns FLUX.2 image generator into unified perception model for segmentation, depth, normals
NielsRogge · x · 2026-11-16
Open Vision Banana, highlighted by HuggingPapers, is a fully open recipe that turns an image generator into a unified perception model capable of segmentation, depth estimation, and normals prediction. It is built on Black Forest Labs' FLUX.2 and released openly on Hugging Face. HF's Niels Rogge amplified it as really cool work.
Related event: Open Vision Banana Open-Sources Recipe to Give Image Generators Vision(2 posts)→
More from Multimodal
- AI-generated megastructures: 'It began as research and ended as ritual' — piotrbinkowski · 2026-11-16
- PixVerse adds Seedance 2.5 Draft Mode: draft in 480p, upgrade picks to 4K — Aiden_Tech_Ai · 2026-11-16
- New image model nails painterly illustration style so well it looks hand-drawn — gizakdag · 2026-11-16
- Bicycle glissando: an unusual MIDI editor built in Antigravity with Gemini 4 Carbon — alexanderchen · 2026-11-16
- Synthesia launches Syren, an AI agent that turns one prompt into on-brand video — synthesiaIO · 2026-11-16
- Open Vision Banana: Flux 2 Klein finetunes give generative models vision, 7 tasks in one — linoy_tsaban · 2026-11-16