LTXVAddGuideMulti

Keyframe conditioning is what turns a video model from "generate something plausible" into "generate this, hitting these images at these moments." LTX-Video got frame/keyframe conditioning back in its 0.9.5 update, and it's the backbone of controlled I2V, first-frame/last-frame chaining, and any shot where you need the clip to pass through specific images. LTXVAddGuideMulti lets you drop several of those guide images into one node instead of stacking a chain of single-guide nodes.

It's from kijai's KJNodes. The "Multi" is the point: one node, many guides, each landing at the frame index you choose, each with its own strength.

How it works

It injects guide images into your conditioning at specified frame indices with per-guide strengths, then returns updated positive/negative conditioning and the latent for your sampler. Say "this image at frame 0, that image at frame 40, this other one at the end," and it encodes each through the VAE and anchors the generation to hit them, blending between them for the frames in between.

The number of guides is driven by a dynamic combo (num_guides), which spawns the right number of guide slots in the node's UI as you increase it. That's a newer ComfyUI UI feature, which brings the one hard requirement below.

The inputs and outputs that matter

• positive and negative - your conditioning. The node folds the guides into these and passes them back out.
• vae - needed to encode the guide images into latent space.
• latent - the video latent being conditioned.
• num_guides - a dynamic combo that sets how many guides you're adding; raising it reveals that many guide slots, where each guide takes an image, a frame index, and a strength (the description's own framing: guide images at specified frame indices with strengths).

Outputs: positive, negative (updated conditioning) and latent, all wired onward into your LTX sampler.

Installing it

You need a reasonably current ComfyUI for this one. The node uses DynamicCombo, which per its description requires ComfyUI 0.8.1 and frontend 1.33.4 or later. Update first if you're behind.

• ComfyUI Manager - search KJNodes for ComfyUI, install, restart. Update ComfyUI itself too if it's old.
• Manual:

  cd ComfyUI/custom_nodes
  git clone https://github.com/kijai/ComfyUI-KJNodes
  pip install -r ComfyUI-KJNodes/requirements.txt
  

  then restart.

No extra model download; it works with your existing LTX-Video setup.

Common issues

The node loads broken, or num_guides won't expand. You're on an older ComfyUI or frontend. The dynamic-combo widget needs ComfyUI 0.8.1 and frontend 1.33.4+. Update both, restart, and the guide slots will behave.

A guide doesn't land where I expected. Check your frame indices against the model's actual frame count, and remember LTX (like other video models) compresses time in latent space, so indices map to the compressed timeline. If a keyframe hits at the wrong moment, the index is usually the thing to adjust.

Guides fight each other / blending looks muddy. Turn down the strengths, or space the frame indices further apart. Two strong guides close together give the model contradictory targets and the in-between frames get mushy. Fewer, well-placed, moderate-strength guides beat a pile of maxed-out ones.

Missing VAE. The guide images have to be encoded, so the vae input is required. If it's unconnected, wire in the same VAE your LTX workflow uses.