ComfyUI-Safe-Chunked-Image-Blend Defeats Freezes via Chunked Blending
ComfyUI-Safe-Chunked-Image-Blend is a new custom node for ComfyUI that gives users explicit control over how image batches are resized and blended. It replaces the standard blend to prevent hidden CPU resizes that can freeze upscaling and video projects.
Developer Xmarre who also made ComfyUI-Diffaid built this custom node to let you pick the compute device, process images in chunks, and log every step.
Explicit resize handling and chunked processing
Key safety and performance features
- Explicit resize policy: error, or resize to match.
- Run blending on CUDA even from CPU inputs.
- Process images one chunk at a time.
- Preallocated output avoids large temporary data.
- Optional CUDA sync per chunk for debugging.
- Detailed logs show shapes, devices, and steps.
Users upscaling images or videos to high resolutions will find this node helpful for avoiding freezes. It’s especially useful when blending a lower‑resolution original with an upscaled version, letting you move resizing and blending to the GPU. Anyone debugging batch workflows will appreciate seeing tensor shapes, devices, and memory use.
How it works and what to watch for
The node arose because the default blend path follows the device of the first image tensor, often triggering large CPU resizes without warning. If compute_device is set to CPU but CUDA is available, the node will automatically switch to CUDA for large resizes to avoid slowdowns. Remember that batch sizes and channel counts must match, and extremely large images may still run out of memory.
“it is a replacement-style image blend node for cases where large image/video tensors get unstable, slow, or freeze when a blend node silently resizes one input to match the other.” — Source: Reddit
Project Links
