IAMCCS Cine LTX Keyframer (legacy alias)

LTX Video's keyframe conditioning is how you force a fast video model to hit specific beats: you encode stills, slot them into the latent at chosen frames, and mark them so the sampler treats them as anchors rather than noise. IAMCCS Cine LTX Keyframer (legacy alias) does that insertion - up to 50 images, each with its own frame and strength - and hands back the latent ready for sampling.

What it does

Inputs:

• vae - the LTX video VAE used to encode the stills.
• latent - the video latent you're inserting into (usually an EmptyLTXVLatentVideo).
• multi_input - the batched IMAGE from the Cine Reference Board (IAMCCS_WDC_MultiImageLoader), which has already normalized everything to one size.
• num_images - how many of the per-image widget sets you want to see. The node has slots for 50; this hides the ones you're not using.

Then, per image (up to 50): insert_frame_N - the frame (in pixel space) where that still gets inserted - and strength_N - how hard the image pins that position (1.0 = full lock, lower = looser suggestion). The tooltip calls these "frame ... in pixel space," which means real frame numbers, not latent slots.

The output is a single latent with the images encoded in and the noise mask updated, so the sampler knows those positions are conditioned rather than free.

How it works

It's the LTX-native keyframe mechanism wrapped in a multi-image batch: encode each still with the video VAE, place it at its frame index in the latent, and mark the mask so the sampler keys off it. The strength per image is the knob that decides whether a still is a hard cut (1.0) or just a nudge in the right direction (0.3). The IAMCCS_WDC_* family is the legacy-compat port of the Cine LTX nodes, so the important thing for you is: this is the node old IAMCCS Cine workflows wire their reference stills into.

Where it fits

Reference Board → Keyframer → sampler (with your LTX conditioning) → VAE decode. If you also want the stills to shape the text conditioning - not just the latent - you want the sibling IAMCCS_WDC_LTXSequencer instead, which does both. The Keyframer is the latent-only version.

Installing it

Part of IAMCCS-nodes. ComfyUI Manager (search "IAMCCS") or:

cd ComfyUI/custom_nodes
git clone https://github.com/IAMCCS/IAMCCS-nodes.git

Restart after. It needs a working LTX Video setup (model + video VAE) in ComfyUI - the node itself adds no model files.

Common issues

The usual suspects: a multi_input batch that isn't size-consistent (fix it at the Reference Board), or frame numbers that overshoot the latent's length - the node clamps on the negative side but an insert past the end is a silent no-op, so double-check insert_frame_N against your total frame count. And remember num_images only controls visible widgets; leave it at 1 until you actually need more.