🎬🔊AGSoft Replace Audio To Video

🎬🔊AGSoft Replace Audio To Video is the dubbing node. It takes a video and a new audio track and puts them together - and critically, it does it without re-encoding the video. FFmpeg stream-copies the video packets, so a big generated clip gets new sound in seconds instead of a full re-render. If you've ever re-encoded a 4K clip just to change the voiceover and watched the quality (and the encode time) crawl, you get why this matters.

This is the natural partner to the pack's audio tools: extract the original audio with Extract Audio From Video, generate or load a replacement, and swap it in here.

How it works

The video source is the optional video input (priority) or video_path. The voiceover is audio_path (a file, priority) or the audio AUDIO input. Then audio_mode decides how the two lengths reconcile - and this is the field that actually shapes your result:

• fit - trim the track or pad it with silence so it matches the video's length. The safe default for a voiceover that's roughly the right length.
• loop - repeat the track across the whole video. Perfect for a music bed.
• trim_to_audio - the opposite direction: cut the video down to the audio's length. Great when the sound is the boss (a song, a finished VO) and the picture just has to fit it.

You also get audio_volume (0–2 multiplier, 1.0 = original), audio_fade_in and audio_fade_out (0–60s) for the new track. output_name is the filename (keep the .mp4 extension as default), output_path the folder, empty = ComfyUI/output.

Outputs

• video_path (STRING) - the muxed result.
• duration_seconds, duration_timecode, file_size_mb.
• width, height, fps, frames_est - the original video's specs, passed through unchanged (which is the whole point of stream copy).

Install

Part of the comfyui-AGSoft pack. ComfyUI Manager → search "comfyui-AGSoft" → Install → restart, or:

cd ComfyUI/custom_nodes
git clone https://github.com/Art-xmaster/comfyui-AGSoft

Restart. No model downloads; the node drives the FFmpeg binary ComfyUI bundles (or a system ffmpeg on PATH).

Where people get burned: the "no re-encode" guarantee is about the video. The audio almost always gets re-encoded into the container (or at least remuxed), so don't expect bit-perfect sound - expect fast, clean muxing. And if you're replacing audio on a clip whose track is longer than the video, fit will cut it off; reach for loop or trim_to_audio depending on which side should win.