Highlights for 2025-10-31

Less than 2 weeks since last release, here’s a service-pack style update with a lot of fixes and improvements:

  • Reorganization of Reference Models into Base, Quantized, Distilled and Community sections for easier navigation
    and introduction of optimized pre-quantized variants for many popular models - use this as your quick start!
  • New models:
    HunyuanImage 2.1 capable of 2K images natively, HunyuanImage 3.0 large unified multimodal autoregressive model,
    ChronoEdit that re-purposes temporal consistency of generation for image editing
    Pony 7 based on AuraFlow architecture, Kandinsky 5 10s video models
  • New offline mode to use previously downloaded models without internet connection
  • Optimizations to WAN-2.2 given its popularity
    plus addition of native VAE Upscaler and optimized pre-quantized variants
  • New SOTA model loader using Run:ai streamer
  • Updates to rocm and xpu backends
  • Fixes, fixes, fixes… too many to list here!