Nodes/AUN ComfyUI Nodes/AUN Clean VRAM
ComfyUI Node

AUN Clean VRAM

Pass-through node that frees GPU memory mid-workflow and returns its input unchanged. On execution it unloads all cached models and empties the CUDA cache, so place it where a heavy model is no longer needed (e.g. between KSampler and VAEDecode, mirroring the CleanVRAM (obvpm) position in MiniMax-H3 workflows). Downstream nodes lazily reload whatever they still need, with no external dependencies.

By loz2754·Created 9 months ago·Updated about 12 hours ago· 6
AUN Clean VRAM
  • anything
  • output
CategoryAUN Nodes/Utility

Inputs (1)

NameTypeDefaultDescription
anythingopt*Any value to pass through. Connect the output that should trigger the VRAM cleanup (e.g. the sampled LATENT).

Outputs (1)

NameTypeDescription
output*—