Image Batch Blend

Image Batch Blend is the dissolve button for AI video. You've got two batches of frames - say, the same shot rendered with two different prompts, or a before/after style pass - and you want them to melt into each other frame by frame. This node takes image1 and image2, mixes them at a blend_factor, and hands back one synchronized sequence. It's the glue that sits between "I generated two versions" and "I want them to transition without exporting to a video editor."

Mechanically it's about as honest as compositing gets: per-pixel math on tensors, zero AI. Which is exactly right for a transition tool. If image1 and image2 have different batch lengths, it truncates to the shorter one - the frames that exist get blended, the rest are ignored. And it normalizes channel counts on the fly, so an RGBA batch and a grayscale batch can meet in the middle (alpha gets dropped, grayscale gets repeated to 3 channels).

The inputs

• image1, image2 - the two batches. Same resolution makes life easy; it won't resize for you.
• blend_factor - 0 to 1, default 0.5. In normal mode this is the crossfade position: 0 = all image1, 1 = all image2, 0.5 = 50/50.
• blend_mode - normal, multiply, screen, overlay, soft_light, add, subtract. These are the classic blend modes from Photoshop/CapCut, implemented as straight per-channel math.

Output is a single blended_images batch you can feed straight to VHS_VideoCombine.

The quirk that will trip you up

Same trap as the pack's other blend node, so hear me once: blend_factor only does anything in normal mode. Switch to multiply and the factor is ignored - the node runs full-strength multiply on every frame, no mixing ratio involved. So if you reach for blend_factor expecting to dial down a screen or overlay, it just won't. For fade transitions, use normal and animate the factor. For "I want these two looks composited," pick a mode and leave factor alone.

Where it fits in a workflow

The pack's example workflow (Video fast change.json) chains it after the push-pull lens and overlay blend, batching frames up with ImageBatch and ImageScale before the final combine. A pattern that works well: generate the same prompt at two different styles, blend them at a slowly-changing factor to get a continuous style drift, and render. That's a genuinely nice effect this node makes trivial - animate blend_factor with a ramp node and you've got a crossfade across the whole clip without touching a timeline.

Installing it

One repo, all seven SwiftCut nodes, standard install - ComfyUI Manager (search "ComfyUI_SwiftCut") or:

cd ComfyUI/custom_nodes
git clone https://github.com/yichengup/ComfyUI_SwiftCut

Restart and it's under YC_VideoCutHelper → Image. Dependencies: torch, Pillow, numpy. No models, no downloads, no hidden requirements. The pack is barely known on reddit, but for "I have two batches and want them mixed," this is a zero-friction answer.