LTXV Multi Concat (N Frames) (beta)

If you've used the regular LTXV Multi Concat, you know the core trick: inject reference images straight into the LTX latent at specific frames, inpainting-style, no attention tokens. The beta sibling keeps that whole mechanism and adds one thing on top - a smooth_strength toggle that tries to fix the biggest visible artifact of the plain version. That artifact is the "harsh slideshow": when you pin keyframes at strength 1.0, each keyframe sits frozen while everything around it animates, and the transitions can read as cuts rather than motion.

The tooltip on smooth_strength says exactly what it does: it interpolates the latents and decays strength between keyframes to prevent those harsh slideshows. Instead of a hard on/off at each anchor frame, the influence of each keyframe fades across the frames leading up to it, so the model transitions into and out of each anchor gradually. If that's the failure mode you've been seeing - and with multi-keyframe LTX it's the most common one - this is the node to reach for.

How it works

Everything about the inputs, positions, and modes is inherited from the non-beta version, so the setup knowledge transfers directly. Same images, same latent, same mode/fps trio for interpreting positions (frames, seconds, percentage), same indices list override, same optional strengths. The one new input is smooth_strength (boolean, default true), which sits among the required inputs.

Under the hood the beta logic interpolates the latent content and the mask strength between anchor frames - a smoothing pass over the latent sequence - before the sampler runs. That's why it's beta: interpolating latents in LTX's compressed space is heuristically motivated rather than formally guaranteed, so results can differ from the strict version in ways you'll want to eyeball. On long videos or many keyframes the extra interpolation also costs a bit of compute.

The inputs that matter

• smooth_strength (boolean, default true) - the whole reason this node exists. On = interpolated, gradual transitions; off = behaves like the original concat.
• images, latent, mode, fps, strength - identical to LTXVMultiConcat.
• indices / positions / strengths - the same keyframe placement controls.

Outputs are positive, negative, and latent, exactly like the parent node - swap it into the same workflow position.

Install

Manager → search AnotherUtils, or:

cd ComfyUI/custom_nodes
git clone https://github.com/marcoc2/ComfyUI-AnotherUtils.git

Restart. Same conditional loading as its sibling: it only appears on ComfyUI builds with the new video API and native LTXV support. If you don't see it, update ComfyUI before suspecting the pack.

Common issues

Since this is a beta of a node you probably already know, the practical advice is comparison-first: run the same keyframes through both versions and judge for yourself whether smoothing helped. The classic trap is leaving smooth_strength off (the default is true, so you have to actively turn it off) and then assuming you're getting the beta behavior - check the toggle. And remember the whole concat family's shared quirks: indices override positions entirely, 0-based frame numbers, and strength 1.0 means "fully fixed." The smoothing interacts with that last one - high strength plus smoothing is usually the sweet spot, but it's worth testing both ways.