ComfyUI-Kompressor v1.0 ships INT8/W4A8 ConvRot quantization for FLUX.2, Wan 2.2, and more
a3tinu · reddit · 2026-09-21
A new open-source ComfyUI node pack, ComfyUI-Kompressor v1.0, brings INT8 ConvRot and W4A8 ConvRot quantization to media generation models, with model inspection and architecture-aware handling.
Auto-detected architectures include Krea 2, Z-Image, FLUX.2 Klein 4B/9B, Wan 2.2, Anima, LTX-2.5, MiniMax H3, and the new Qwen Image 2.1. First public release—bug reports and feedback welcome. GitHub: github.com/a3nu/ComfyUI-Kompressor.
More from Multimodal
- AI-Generated UGC Jewelry Ad Looks Real: No Actress, No Camera, Just a Prompt — aftahi_ai · 2026-09-21
- Higgsfield Genjutsu Videos Go Viral as 'Fake Life Maxxing' Meme — SimplyAnnisa · 2026-09-21
- H3 video quality struggles on a 5070 Ti: smudged faces regardless of workflow — ApplicationCreepy987 · 2026-09-21
- Will the community distill a 4-step LoRA for Qwen 2.1 Klein? — Friendly-Fig-6015 · 2026-09-21
- Qwen-Image-2.1 GGUF Quantized Model Trends on Hugging Face for ComfyUI — abenzerps · 2026-09-21
- Xpeng's XGEN Debuts World Model XGEN-JING Built on MiniMax H3, Weights Released — AdinaYakup · 2026-09-21