⚠️ Google Colab & Low-VRAM Users (Important)

## ⚠️ Google Colab & Low-VRAM Users (Important) IDM-VTON is based on SDXL and requires **high VRAM** for stable inference. ### Known limitations - Google Colab T4 / P100 often crashes with `SIGKILL (OOM)` - 12–16 GB VRAM is usually **not sufficient** for 768×1024 inference - Mixed precision alone may not prevent OOM ### Recommendations - Prefer A100 / L4 / RTX 3090+ GPUs - Reduce resolution (e.g., 384×512) for testing only - Use batch size = 1 - Expect degraded quality at low resolutions If you encounter crashes on Colab, this is a **hardware limitation**, not a code bug.