Save Prompt Travel file

"Prompt travel" is the trick where every frame of an AnimateDiff video gets its own prompt and the video quietly morphs through all of them - a building un-building, a woman aging backward, a scene melting from city to desert. The hard part was never writing the per-frame prompts. It's keeping them around in a form you can actually reuse. That's the whole reason Save Prompt Travel file exists: it's the write-half of a tiny two-node pair from Hiero207's personal pack, and it does one thing - take your batch of per-frame prompts and drop them into a plain text file.

It's a sink node, pure and simple. You feed it a STRING of prompts, tell it where to write, and it writes them out. No frame numbers get baked in - by design, the loader half of the pair adds those at load time, so you can start a video at any point in a collection and have the prompts count from frame 1. The author built this specifically for AnimateDiff workflows, and the README describes the source of the prompts honestly: he passes each frame of a reference video through a WD14 Tagger node and saves the tag output. It's a one-way trip out of the graph.

You only set two of the three inputs in practice:

• prompts - the big string of per-frame prompts, one per line, typically coming out of a WD14 Tagger or just typed into a multiline box.
• save_directory - where the file lands (default is a placeholder save/dir, so set a real path).
• file_name - the filename (default prompts, which matches the loader's own default, handy).

There are no outputs. Wire nothing out of it; the "result" is a file on disk.

The typical loop: tag a reference video frame-by-frame with WD14, collect the prompt lines, save them here, and later load them back with Load Prompt Travel file feeding FizzNodes' Batch Prompt Schedule - which the author calls out as working great with this pair and is the node that actually interpolates the prompts into an AnimateDiff-Evolved conditioning timeline. One honest caveat from the KB: AnimateDiff is legacy territory now, superseded for general video work by Wan and friends. Per-frame prompt scheduling is still how a lot of SD1.5-style animation is driven, so the file format remains useful even as the models move on.

Installation is as boring as it gets, which is a compliment. This is one of the rare packs with zero dependencies - no requirements.txt, no model downloads, just standard-library file IO. ComfyUI Manager → search Hiero-Nodes (the pack title), or clone it:

cd ComfyUI/custom_nodes
git clone https://github.com/Hiero207/ComfyUI-Hiero-Nodes

then restart ComfyUI. One caveat: at the time of writing the pack's GitHub URL returns a 404, so the repo may have been renamed or pulled - if the clone fails, Manager's registry is the more reliable path.

The only thing that trips people up is expecting to see something happen. There's no output preview, no "saved!" toast - check your directory for the file. Paths default to something relative to where ComfyUI runs, so make sure the folder exists. And don't go looking for frame numbers in the saved text; that's the loader's job, and it's why the pair works so well together.