Load Video Path FFmpeg (Topaz)

Same loader, different door. TopazLoadVideoFFmpegPath is the sibling of TopazLoadVideoFFmpeg in the ComfyUI-TopazVideoAI pack, and it exists for one reason: the dropdown version only scans ComfyUI/input, and not everything you want to enhance lives there. This one takes a raw file path - drag in a clip from anywhere on disk, paste a full path, and it decodes the frames with Topaz's own bundled ffmpeg instead of whatever ffmpeg happens to be on your PATH.

It's the loading half of the same two-step flow: this node turns a video into an image batch, the pack's TopazVideoAI node runs the actual tvai_up / tvai_fi upscaling and frame-interpolation filters, and VHS_VideoCombine writes the result. If you'd rather keep your source videos outside ComfyUI's input folder - your own footage library, a scratch dir, files pulled by an earlier step of the pipeline - this is the one you reach for.

How it works

The internals are identical to the dropdown version: both funnel through the same _do_load_video helper, spawn ffmpeg, parse the stream header for resolution/fps/alpha, and decode to float32 [0,1] tensors with a lazy-loaded audio track. The differences are in the front end. Where the other node validates against ComfyUI's input directory, this one checks os.path.isfile and raises a clear video is not a valid path error if the file isn't there. The video field is a free-text STRING with a X://insert/path/here.mp4 placeholder, and quote marks around the path are stripped for you. The pack also adds a custom /topaz/view_video server endpoint so you can preview arbitrary paths - ComfyUI's built-in /view only serves input/output/temp, which is exactly the limitation this node sidesteps.

The inputs that matter

• video - the full path to a video file. Absolute paths are safest; it handles spaces and wrapped quotes. Anything in video_extensions works (mp4, mov, mkv, webm, gif, avi, flv, ts, m4v, mpg, mpeg).
• force_rate - target fps, 0 keeps source. Match your model's native fps.
• custom_width / custom_height - resize before downstream nodes; 0/0 means untouched.
• max_duration / start_time - seconds, 0 off. The port from VHS exists because this pack keeps these in seconds, which makes splicing clips by time much less error-prone than frame math.
• meta_batch (optional) - a VHS_BatchManager for VHS-driven batches.

Outputs

IMAGE, frame_count, mask, audio, video_info - the exact same contract as VHS LoadVideoPath, so anything that accepts a VHS video pipe will happily take this. Wire IMAGE into the pack's TopazVideoAI node. mask is the alpha channel inverted (all zeros if no alpha), audio is lazy-loaded, and video_info carries source + loaded fps, duration, and dimensions for the rest of the graph.

Install

Same as the whole pack: Topaz Video installed and licensed first (no model files ship with this repo - they come from your Topaz install), then either search "TopazVideoAI" in ComfyUI Manager or:

cd ComfyUI/custom_nodes
git clone https://github.com/sh570655308/ComfyUI-TopazVideoAI.git

Restart ComfyUI, and once, from a shell in the Topaz install folder, run .\login (Windows) so licensing checks pass later.

Gotchas

The path node's failure modes are mostly about the path. video is not a valid path means exactly what it says - check drive letters, backslashes on Windows, and trailing spaces. And the two pack-level traps still apply: after upgrading from old "Topaz Video AI", delete stale TVAI_MODEL_DIR / TVAI_MODEL_DATA_DIR environment variables or you'll chase Model not found errors in the upscale node; and keep in mind the loader prefers Topaz's ffmpeg but falls back to system ffmpeg if it can't find one, which decodes fine but won't run the tvai_up / tvai_fi filters the other node needs.