KY_FFmpeg: Image Sequence to Video

Where KY_SaveVideo takes an IMAGE tensor batch straight out of your ComfyUI graph and writes a video, this node starts from the other end: frames that are already sitting on disk as individual image files. Point it at a folder, and it runs a proper ffmpeg encode with the kind of controls you'd expect from actually driving ffmpeg by hand - codec choice, rate-control mode, colorspace, pixel format - rather than a couple of simplified sliders.

When you'd reach for this instead of KY_SaveVideo

If your frames came from this pack's own KY_FFmpegVideoToImages (extracting frames from a source video), from an external tool, or from a batch render you already saved to a folder with KY_SaveImageSequence, they're on disk, not sitting in VRAM as a tensor. Re-loading them all back into ComfyUI as an IMAGE batch just to hand them to KY_SaveVideo wastes memory for no reason - this node skips that round-trip and lets ffmpeg read the files directly.

Inputs that matter

• input_dir / output_path - the source frame folder and the destination video file. No defaults worth keeping; the schema ships Windows-style placeholder paths (C:/path/to/...) that make it obvious you're expected to override both.
• input_fps - a long dropdown of standard rates from 2 up to 59.94+, matching the source frame rate your images were extracted or rendered at.
• encoder_name - libx264 (default, safe, CPU, broadly compatible), h264_nvenc/hevc_nvenc (GPU-accelerated, but only work if ffmpeg was built with NVENC support and you actually have an NVIDIA GPU available to it), libx265, prores_ks, libvpx-vp9.
• quality_type - CRF, QP, or Bitrate. This changes what quality_value actually means, which is the trap worth flagging up front (see Common issues).
• colorspace_choice (BT.709 HD / BT.2020 UHD-HDR / Unspecified) and pix_fmt - for matching whatever your downstream player or editor expects; leave at the HD default unless you know you need otherwise.

Optionally, audio_in_path lets you mux an audio track in alongside the video, with audio_codec (aac/copy) and audio_shortest (default true - trims the output to the shorter of the video or audio stream, rather than padding or looping to match).

One output: video_path, the finished file's location. It's marked an output node, so it writes to disk as its side effect.

Installing it

ComfyUI Manager: search ComfyUI-KYNode, install, restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/yorkane/ComfyUI-KYNode

Restart ComfyUI. Like the rest of the KY_FFmpeg* family, this needs ffmpeg itself installed and reachable on your system PATH - the pack's own README doesn't call this dependency out explicitly, so it's easy to miss until the node fails on first use. Run ffmpeg -version from wherever ComfyUI's process actually runs to confirm before you troubleshoot anything else.

Common issues

quality_value means a genuinely different thing depending on quality_type. A value of 18 under CRF is a well-known "visually lossless" x264 setting; that same 18 under QP or Bitrate mode is a completely different number with a completely different effect. Check which mode you're in before assuming a familiar CRF number will behave the same way.

h264_nvenc/hevc_nvenc will fail - often with a cryptic ffmpeg error rather than a clear "no GPU found" message - if your ffmpeg build wasn't compiled with NVENC support, or if there's no compatible NVIDIA GPU visible to the process. Run ffmpeg -encoders | grep nvenc first if you're not sure your setup actually supports it; if it comes back empty, stick with libx264.

audio_shortest defaulting to true means a longer audio track than your video gets silently cut short, not looped or padded - if your final video's audio ends early and you didn't expect that, this is the setting to check.

And since the node writes to a real file path, make sure the destination directory in output_path actually exists - the schema doesn't document whether it auto-creates missing folders, so pre-creating the destination is the safe move.