Frame Replacer ✂️

FrameReplacer is the frame-accurate edit node. Where FrameExtender splices whole clips together, this one swaps out a specific run of frames inside a video: frame 40 through 55 becomes a new sequence you feed in, everything else stays untouched. It's what you reach for when a re-render left a glitched face for a few frames, you want to swap in a better take of one shot, or you need to overwrite a segment of a video with new content at an exact timestamp.

How it works

You give it source_video (the video to edit) and replacement_frames (what goes in), then point at the spot with target_frame (0-based index) and say how many frames to replace with replace_count. It slices the source around that window and drops the replacement in. Two flags control the edges of the behavior:

• loop_replacement (default on) - if your replacement batch has fewer frames than replace_count, it tiles the batch to fill the gap, like a stamp. Turn it off and the replacement just ends early.
• preserve_length (default on) - keeps the output the same length as the source. This matters more than it sounds: with it on, if the replacement overshoots the end of the video it gets capped; with it off, the length changes to match what you actually fed in. Keep it on unless you're deliberately re-timing the clip.

The boundary problem is the same one any video editor solves with a dissolve: a hard cut where your replacement meets the original frames can pop. blend_edges (default off) fixes that by crossfading the join, and blend_strength (0.0–1.0) controls how heavy the fade is. With a small blend_strength like 0.3–0.5 you get a soft landing instead of a visible seam.

The inputs that matter

• target_frame and replace_count - the window. This is the whole job.
• blend_edges / blend_strength - flip the first on when you see a seam at the boundaries.
• preserve_length - leave on unless you specifically want the video to grow or shrink.

The rest (width, height, interpolation, method, condition, multiple_of) is the pack's shared advanced-resize system. Same caveat as the rest of this pack: defaults are 512×512 with condition: always, so set width/height to your real resolution and switch condition to if different unless you want an unexpected downscale. Output is modified_video, an IMAGE tensor you can feed to SaveImage/VideoCombine or keep processing.

Use cases

• Fix glitches: rerun one bad section through a rediffusion or inpainting pass, then patch the clean frames back in.
• Frame-accurate timing: swap a single frame to fix a flicker or a double-exposure.
• Localized edits: replace just the segment you changed instead of re-rendering the whole video.

Installing

Manager: search "ComfyUI-FrameUtilitys". Or:

cd ComfyUI/custom_nodes
git clone https://github.com/lum3on/ComfyUI-FrameUtilitys

Restart. No models, no extra dependencies - the pack's requirements.txt is comments only, and everything runs on ComfyUI's bundled torch/numpy.

Troubleshooting

• "Frame index out of bounds" - target_frame + replace_count exceeds the video length. Shorten replace_count or check your frame math; the readme's own advice is to verify replace_count doesn't overshoot available frames.
• Replacement isn't showing up - check preserve_length; if it's on and your replacement overshoots, the tail gets cut to keep the length constant.
• Visible seam at the join - enable blend_edges and nudge blend_strength up until the pop disappears.