AD_latent_history

This is the smallest node in the pack, and it's also sitting in the pack's own Deprecated folder. It exists to do exactly one thing, per its own node description: "Pack sampled latent tail to HISTORY." That's it. It's the companion piece to AD_Latent_Diffusion_Keyframe - after a sampler produces a segment of a video, this node would take the resulting latent and package its trailing frames into a HISTORY object, which then fed back into AD_Latent_Diffusion_Keyframe's history input to carry continuity into the next segment.

Since both halves of that pairing live in Deprecated, treat this as a historical reference rather than something to build a new workflow around - the pack's active video pipeline is the LTX_video branch, where AD_LTX_vae_combine gives you a last_image output specifically designed for chaining one generation segment into the next, without needing a separate history-packing node at all.

What it did

Took a sampled latent and repackaged part of it (the "tail" - presumably the last N frames of that segment's output) into a history object, ready to be consumed by a segment-relay node downstream. No image or conditioning handling here, just latent bookkeeping.

The inputs and outputs, for the record

Required: latent (LATENT) - the sampled output you want to pack. Output: history (HISTORY) - a single custom-typed object, consumed by nodes that accept a history input (in this pack's current set, that's AD_Latent_Diffusion_Keyframe, also deprecated).

Installing the pack (if you're maintaining an old workflow)

Through ComfyUI Manager: search "ComfyUI-Apt_Preset". Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/cardenluo/ComfyUI-Apt_Preset.git

Then run install.bat (Windows) or pip install -r requirements.txt (Linux/Mac) per the pack's README, and restart ComfyUI. As with its sibling node, "deprecated" here means moved out of the active menu category, not stripped from the codebase - installing the pack still gives you access to it if an old workflow needs it.

What to actually use today

If you're doing segment-by-segment video generation with continuity between segments, the LTX_video branch is the current path: build your config with AD_LTX_load_model, sample with AD_LTX_sampler, and decode with AD_LTX_vae_combine - the last_image output from that last node is your continuity hook for the next segment, taking over the job this node used to do at the latent level.

Common issues & troubleshooting

There's nothing meaningfully node-specific to troubleshoot here beyond the deprecation itself - a one-input, one-output packing node has very little surface area to break. If history isn't behaving the way you expect, or the node throws on a latent shape it doesn't like, the more productive move is migrating the workflow to the LTX_video branch rather than debugging a node the author has already moved past.