Media Bridge (FFMPEGA)

ComfyUI thinks in IMAGE tensors - big batched arrays of frames living in GPU memory. FFMPEGA's Agent thinks in file paths - strings that ffmpeg reads from disk. Those two worlds don't talk, and that's the entire reason Media Bridge (FFMPEGA) exists. It's a bidirectional adapter: encode a tensor batch into a temp video file, or decode a video file back into a tensor batch plus audio. One node, a mode dropdown, both directions.

This is the node you reach for when a workflow has one foot in each world. Say you generated frames with a video model, want to run them through FFMPEGA's editor or agent, then come back to ComfyUI's tensor-based nodes with the result. Bridge handles the seams so you don't have to hand-assemble temp files and remember where you put them.

The inputs that matter

• mode - the direction. images_to_path encodes an IMAGE tensor into a temp video file; path_to_images decodes a video path back into IMAGE + AUDIO. That's the whole node, really.
• images - used in images_to_path mode: the frames to encode. Any IMAGE source works.
• video_path - used in path_to_images mode: the file to decode.
• fps - encoding frame rate (default 24), images_to_path only. Ignored in the other direction, where FPS is read from the file.
• audio - optional, images_to_path only: mux this AUDIO into the encoded video. Leave it out and you get a silent temp file.

Outputs are populated in both modes for predictable wiring: video_path (temp file in one direction, empty string in the other), images, audio, plus fps and frame_count, which are always filled regardless of mode - handy metadata if you're using this as a pass-through measurement point.

Installing it

It ships in ComfyUI-FFMPEGA, so it comes with the pack: ComfyUI Manager → search ComfyUI-FFMPEGA → Install, or:

cd ComfyUI/custom_nodes
git clone https://github.com/AEmotionStudio/ComfyUI-FFMPEGA.git
cd ComfyUI-FFMPEGA
pip install -r requirements.txt

Restart ComfyUI. FFMPEG on PATH is required (the encode/decode is real ffmpeg work, so ffmpeg -version should respond).

Where people get burned

Two things to keep straight. First, this writes a temp video - it's a bridge, not a save node. If you want a permanent file, run the result through Save Video (FFMPEGA), which copies it into ComfyUI's output directory without re-encoding. Second, images_to_path gives you a silent video unless you connect audio; if the downstream node is supposed to have sound and it doesn't, that's why. And don't confuse the bridge's 24 fps default with your actual content - if your frames were generated at 30 or 16, set fps explicitly or your "video" plays at the wrong speed and the decoded duration lies to you.