Repository navigation
Image Studio v0.2.10 (Hotfix: Model Re-Download & VRAM Cache)
Image Studio v0.2.10 (Hotfix: Model Re-Download & VRAM Cache)
Fixes & Improvements in v0.2.10:
- Prevent Unnecessary Model Re-Downloads:
- Existing installs that already have weights downloaded (e.g.
qwen3vl_8b_int8_convrot.safetensors) now automatically preserve the standard INT8 profile instead of falling through to the 8GB profile. - Profile ordering in
models.jsonset to default standard INT8 (12GB+) first. - Added robust detection of existing models across all candidate directories before picking a profile.
- Fresh installs on <= 8GB GPUs will automatically select the 8GB quant profile; existing installs on 12GB+ VRAM or prior versions never re-download.
- Existing installs that already have weights downloaded (e.g.
- VRAM Detection Caching:
- Cached GPU VRAM query (
nvidia-smi) in aOnceLockso it executes only once at startup instead of running on every UI render frame.
- Cached GPU VRAM query (
- Canonical Windows Path Separators:
- Fixed mixed slash formatting in default candidate path joins (
ComfyUI-Shared/models->ComfyUI-Shared\models).
- Fixed mixed slash formatting in default candidate path joins (
Download:
Download ImageStudio.exe below and run!