SAM3 + LTX-2.5 alpha-gen trick hits 88.7% hair matte accuracy in ComfyUI, open source
Last_T_Rex · reddit · 2026-10-29
Developer LastTRex open-sourced a ComfyUI node that fixes LTX-2.5's alpha-gen IC-LoRA tendency to matte everyone in frame, enabling single-subject extraction from crowds with native 4K hair-level mattes.
The pipeline:
- SAM3 picks the subject from one click
- Other people are inpainted out of a copy of the plate
- LTX alpha-gen runs on that copy so only one person remains
- 4K plates get a second LTX pass on native 1920x1088 tiles along edges
- Overlap mode: three LTX passes (one attention mask per person) plus a Depth Anything vote resolve occlusion
On a 247-frame shot it achieves 88.7% edge detail match vs 45.0% for SAM3+ViTMatte and 29.4% for raw SAM3 masks. Runs on 12GB VRAM and 16GB RAM with weight streaming; open weights only, GPL-3.0, available in ComfyUI Manager.
Related event: Open-source ComfyUI node delivers 4K hair roto with SAM3 and LTX-2(2 posts)→
More from Multimodal
- AReA: training-free attention method pushes Qwen Image and Flux to 6K — linoy_tsaban · 2026-10-29
- 115-Prompt Test: Abliterated Text Encoders Barely Affect Image/Video Generation — afinalsin · 2026-10-29
- Can Open-Source Models Generate Comics With LLM-Written Speech Bubbles? — Someoneqqqq · 2026-10-29
- Flova 1.5 auto-generates missing character assets from your script for consistent AI video — socialwithaayan · 2026-10-29
- maya2 TTS Model Trends on Hugging Face with Streaming and Voice Cloning — maya-research · 2026-10-29
- Cozy AI video: a humanoid robot reading by the window on a rainy day — merkelmusic · 2026-10-29