(Deno) LTX Sequencer

Guide-image workflows are how people get LTX and Wan to actually obey a storyboard: you insert reference images as keyframes into the latent timeline, and the sampler does its best to hit them in sequence. Doing that by hand means fiddly per-frame math and a graph full of tiny conditioning nodes. (Deno) LTX Sequencer is the cleaned-up version: feed it a batch of guide images, tell it where in the timeline each one lands, and it returns sequenced positive/negative conditioning plus the latent with guide frames appended.

How it works

The node is tuned for the batch output of the pack's (Deno) Multi Image Loader - connect one and it even auto-fills num_images from how many images the loader is actually feeding, which removes the most common "I changed the image count and the sequence broke" failure. For each guide image you set an insertion point and a strength:

• insert_mode - positions entered as frames or seconds. Seconds mode converts using frame_rate (24 default), so you can think in time instead of frame numbers.
• Per image: insert_frame_N / insert_second_N and strength_N - strength is deliberately the only thing that gets to break out into manual control, per the author's design.

The bypass toggle is the quiet hero: flip it and positive, negative, and latent pass through unchanged, which gives you a one-click A/B between "with guide sequence" and "plain generation" for the same graph. There's also strength_sync, which keeps guide strengths synced across multiple sequencer nodes when you use the frontend sync tool - handy for multi-shot workflows where every shot should treat its references with the same weight.

Outputs are positive and negative CONDITIONING (guide-image timing inserted) and the latent timeline with guide frames appended - straight into your sampler.

Inputs to care about

• positive / negative - your base conditioning before guides.
• vae - used to encode the guide/reference images into LTX latent space.
• latent - the LTX latent timeline receiving the keyframes.
• multi_input - the batch of guide images.
• num_images, insert_mode, frame_rate, then per-slot positions and strengths (up to 18 slots).

Install and context

cd ComfyUI/custom_nodes
git clone https://github.com/Deno2026/comfyui-deno-custom-nodes.git
# restart ComfyUI

It sits on top of the official ComfyUI-LTXVideo stack, so that plus a 2.3 model is the prerequisite. The workflow idea is borrowed from WhatDreamsCost's LTX approach, reshaped for daily use - if you've seen those multi-image LTX workflows and wanted the same power without rebuilding the plumbing, this is that.

The trap to know: this is a sequence tool, not a length tool. The latent timeline and its frame budget come from your empty-latent/sampler setup upstream - the sequencer inserts guides into what exists; it doesn't extend the clip. If your references land after the clip ends, nothing visibly "breaks," the frames just don't get reached, which confuses people no end. Check your latent's frame count against the last insertion point before you blame the node.