NL LTXV Img To Video Inplace At Index

Standard LTXV image-to-video conditions on the first frame: you feed an image, the model animates forward from it. That's fine for most shots, but it's also the whole game, and LTX users have known since the 0.9.x days that conditioning on a middle or end frame gives dramatically more interesting results - the model "guesses" less, and motion gets anchored at both ends. This node is the pack's answer: it takes an LTXV-encoded image and places it at an arbitrary pixel-frame index inside an existing video latent, so you can anchor your clip on frame 0, frame 40, the last frame, or anywhere else.

The longer name is doing a lot of work. "Inplace" is the key word: instead of building a latent from scratch, it works on a latent you already have (typically from a text-to-video or previous stage), preserves any chained noise masks riding on it, and injects the encoded image at your chosen index. That's the mechanism that makes middle-and-end-frame conditioning possible without tearing up the latent structure.

The settings that matter

• vae - the LTXV VAE used to encode your image into the latent.
• image - the frame to inject. latent - the video latent you're modifying.
• frame_idx - where to put it. 0 is the first frame, -1 is the last, anything else is that pixel-space index. This is the control that makes the node interesting.
• strength - conditioning strength, 0–1. At 1.0 the target frames are fully conditioned on the image; dial it down to let the model drift from the anchor.
• LTXV preprocessing controls: crf - compression applied before encoding (0 disables; the LTXV community's well-worn advice is CRF in the 30–40 range for the best motion/quality balance), blur_radius - a Gaussian blur after preprocessing (0 disables), interpolation - the resize algorithm (lanczos default, or bislerp and friends), and crop - disabled or center.

Output: a single latent, ready for the sampler.

How it fits

LTXV's selling point has always been speed - "blazing fast" iteration compared to Wan - and I2V is its community-preferred mode. This node leans into the end-frame trick that long-time LTX users swear by: anchor the start and the end, and the middle stops drifting. Start with frame_idx = 0 and strength = 1 to confirm the basics, then try -1 and watch how much more controlled the ending becomes.

Install

Standard pack install:

cd ComfyUI/custom_nodes
git clone https://github.com/NOLABEL-VFX/ComfyUI-NL_Nodes

or ComfyUI Manager → "ComfyUI-NL_Nodes", restart. It depends on LTXV's own ComfyUI integration (the VAE/LATENT types come from there), so have ComfyUI-LTXVideo installed and working first - this node is a wrapper on top of that, not a replacement. The bypass toggle returns the latent untouched, which is a handy A/B switch to prove the injection is doing what you think.