SeedVR2 batch upscale (avoid offloading model)
A critical bug is crippling batch processing for one of the hottest new AI upscalers.
Deep Dive
Users of the popular SeedVR2 upscaler in ComfyUI are reporting a major performance bug during batch processing. The model is being offloaded from VRAM after processing each individual image, forcing a complete reload for the next one. This creates massive slowdowns, turning what should be a quick batch job into an agonizingly slow process. The community is actively seeking a fix, as the workaround currently involves complex manual workflow adjustments.
Why It Matters
This bug makes professional-grade batch upscaling practically unusable, blocking workflows for artists and content creators.