Video Transition Batch Merger (Buff)

This is the heavyweight of the pack, and it's the most ambitious node here by a wide margin. VideoTransitionBatchMerger takes multiple ComfyUI image batches - each one a clip - and merges them into a single timeline with a real transition between every pair, matching audio, a final frame count, and even a timeline summary string. It's a miniature video editor living inside a node.

The gap it fills is the one you hit the moment you generate a video in pieces - clip A, clip B, clip C - and need them cut together with a cross-dissolve instead of a hard jump cut. Without something like this, you're either leaving cuts in, or exporting each clip and cutting them in an external editor. This node keeps the whole thing inside ComfyUI and, because it's backend tensor math, it works through the API without browser-only extensions.

How it works

Two clips (image_batch_1, image_batch_2) is the minimum; crank inputcount and hit the node's "Update inputs" button to reveal image_batch_3+ and matching audio_N sockets, all the way to 64. Every socket also exists on the backend regardless of the UI, so raw Prompt API workflows can wire image_batch_9 directly. Between each pair, transition_mode decides the shape: overlap consumes handles from the outgoing and incoming clips (output shorter than the sum of inputs), non_overlap inserts generated frames between clips (output longer), and auto picks per transition - overlap for A/B roll moves, inserted frames for dips and camera artifacts.

Then pick your transition. There are ~40, from cross_dissolve, wipes, iris wipes, pushes and slides, to dip_to_black, morph_cut, whip_pan, light_leak, glitch, film_burn, even page_peel and cube_spin. default_transition and default_transition_frames set the baseline, and the per-pair transition_plan text field overrides individual boundaries with lines like 2:dip_to_white:8. easing shapes the curve, wipe_softness feathers wipe edges, effect_intensity boosts the bloom/leak/whip effects.

The audio story is the part that surprised me. Connect audio_1/audio_2 (ComfyUI AUDIO or legacy VHS_AUDIO both accepted) and it builds a single aligned track: clips without audio get silence, inserted transition frames get silence, and overlapping transitions trim and crossfade the overlapping audio handles. fps drives the whole timeline and audio alignment.

Inputs that matter

inputcount, the two image batches, fps (default 24), default_transition / default_transition_frames, transition_mode, and transition_plan for per-pair overrides. size_match (strict, first_batch, largest_area, smallest_area) plus resize_method (stretch, pad, crop) handle clips of different dimensions - strict errors if they differ, which is the safe starting point.

Outputs: merged images, aligned audio, fps, frame_count, and a timeline_info summary string for debugging.

Install and gotchas

From BuffMcBigHuge's pack - ComfyUI Manager (search "ComfyUI-Buff-Nodes") or:

cd ComfyUI/custom_nodes
git clone https://github.com/BuffMcBigHuge/ComfyUI-Buff-Nodes

Restart ComfyUI. No extra dependencies.

The gotchas are the usual video-tooling ones. strict size matching will throw on mismatched clips - that's a feature; set first_batch or pad if you want it to cope. fps matters for audio even if you don't care about video timing, since it aligns the audio timeline. Transition frames cost VRAM on long timelines; default_transition_frames of 0 gives a hard cut if you're memory-bound. And remember the "Update inputs" button - the UI-driven sockets don't appear until you press it. It's the node in this pack with the most moving parts, and the one that most clearly earns its keep.