LTX Video Context (TTP)

Zero clicks, zero impressions on this one so far, and it's not written up anywhere in the pack's README beyond a general mention that the examples/ folder has LTX workflow helpers - so this is a straight read of the node's own inputs, tooltips, and how it fits alongside its two siblings, LTXVFirstLastFrameControl_TTP and LTXVMiddleFrame_TTP.

What it's for. Video diffusion models generate a fixed-length clip at a time, and the standing community problem is stitching multiple clips into one continuous take without a visible seam where they meet. This node is a continuation helper: you give it previous_video (the frames of a clip you already generated), plus vae and a fresh latent for the next segment, and it conditions that new latent on the tail of the previous clip so the new generation picks up where the last one left off instead of starting cold.

How the math works, straight from the author's own tooltip. context_latent_frames (2–20, default 6, the author recommends 6) controls how much of the previous clip gets carried forward - but not in a 1:1 frame count. LTX's VAE compresses time as well as space, so the tooltip spells out the conversion: latent frame 0 corresponds to 1 original frame, and every latent frame after that corresponds to 8 original frames. At the recommended value of 6, that works out to 1 + (5 × 8) = 41 original frames of context, which the tooltip states is roughly 1.64 seconds at 25fps. If you want more or less carried-over context, that's the ratio to keep in mind rather than assuming it's linear.

context_strength (optional, 0.0–1.0, default 1.0) is the same fixed-vs-free dial you'll recognize from the first/last-frame nodes in this pack: 1.0, per the tooltip, means the context is "fully fixed" - the new generation is locked to match the previous clip's ending as closely as possible. Lower values give the sampler more latitude to diverge, which is worth doing deliberately if the previous clip's ending had a flaw you don't want carried forward and locked in.

Output: a single latent, ready for your sampler - same pattern as the pack's other LTX conditioning nodes.

Install. ComfyUI Manager, search "Comfyui_TTP_Toolset", or cd ComfyUI/custom_nodes && git clone https://github.com/TTPlanetPig/Comfyui_TTP_Toolset, then restart. No extra downloads for this node specifically - just the vae from whatever LTX checkpoint you're already using for the new segment, which should match the checkpoint that produced previous_video in the first place.

Common issues. Being honest about what I do and don't have here: this node has no measurable search traffic yet and I couldn't find it discussed anywhere in community threads, so there's no real-world failure report to relay. What I can offer is the same logic that applies to its siblings: context_strength=1.0 gives you the tightest continuity but the least room for the sampler to fix anything wrong with the source clip's final frames, so if a continuation looks unnaturally frozen or jerky right at the seam, try backing context_strength down before assuming something else is broken. And if the "1.64 seconds of context" default feels too short or too long for your use case, adjust context_latent_frames using the 1-then-8-per-frame conversion above rather than guessing at a frame count directly.