ComfyUI Node
Krea2 SVDQuant W4A4 Loader
Loads a W4A4 + low-rank (SVDQuant) Krea2 checkpoint. Self-contained: no separate base model needed. The status output tells you whether the fast int4 kernel is in play.
Krea2 SVDQuant W4A4 Loader
- model
- status
◄model_name▾►
◄vram_managementauto►
CategoryKrea2/SVDQuant
Inputs (2)
| Name | Type | Default | Description |
|---|---|---|---|
| model_name | COMBO | A checkpoint from quantize_krea2.py --format svdq (it carries *.svdq_l1/*.svdq_l2 tensors). The --format w4a4 / int8 / fp8 checkpoints have no branch and load with the stock UNETLoader instead. | |
| vram_managementopt | COMBO | auto | Leave on "auto" unless something is wrong. "auto" lets ComfyUI manage this model the way it manages any other, which on an ordinary launch means the dynamic-VRAM streaming patcher -- that is what keeps iteration time flat once the model stops fitting in VRAM. "classic" pins it to the older patcher, where a model that no longer fits falls back to streaming weights per module per step and an iteration goes from ~1 s to 30-100 s. KREA2_DISABLE_DYNAMIC=1 forces classic for a whole server run. |
Outputs (2)
| Name | Type | Description |
|---|---|---|
| model | MODEL | Wire this to a KSampler. |
| status | STRING | Rank, variant, size and which kernel the quantized layers will actually dispatch to. Read this if generation is slow. |