Hi folks, this is CCS.
Today I’m sharing a Long Length Video workflow for LTX-2 in ComfyUI, built around the latest updated IAMCCS_nodes The whole point is simple: you generate a solid first segment, then you extend it iteratively (multi-segment) while keeping continuity, overlap, and timing under control—without turning your graph into spaghetti.
Before we start: if you want the deep dive on the brand-new nodes (what every widget truly does, why some options exist, and the more technical diagrams), I already made a dedicated post for that. If you want to level up your skills and actually understand the “why” behind the pipeline (instead of just copy/paste), support me and unlock the supporter posts. That support is a real contribution to my work—it literally buys me time to keep building, testing, documenting, and sharing workflows I use as a filmmaker in real projects, then releasing them back to the open-source image/video generation community. More updates are coming, and I’ll keep pushing them out as fast as I can.
Now, the workflow.
What you need installed (important)
You’ll want:
-
IAMCCS_nodes (updated) for the LTX-2 extension nodes and helpers described below.
-
IAMCCS_annotate, because the workflow contains annotated tips: if you don’t have Annotate installed, you’ll still be able to run the graph, but you’ll miss the embedded notes that explain how to tweak it efficiently.
Core idea in one sentence:
You’re building a long video by chaining segments, and each new segment is blended into the previous one using overlap frames (crossfade / seam management), while extracting the correct start/guide frames for the next iteration. This is exactly what the IAMCCS LTX-2 Extension Module and Get Images From Batch are built for.
Step-by-step
Step 1 — Load your LTX-2 base + VAE setup
At the top of the workflow you’ll see the usual LTX-2 stack: checkpoint/model loaders and the VAE pieces. In this graph there are also switchers (rgthree Any Switch) to keep the pipeline flexible (swap CLIP/VAE/model variants quickly).
Step 2 — (Optional) GGUF path acceleration
If you’re using a GGUF UNet route, the workflow includes IAMCCS_GGUF_accelerator, which is meant to reduce the painful CPU↔GPU patch movement overhead (especially when LoRA patches are involved). It doesn’t magically change your sampling quality—this is runtime/throughput hygiene.
Step 3 — Stage your model / LoRA logic (cleanly)
You’ll notice staged logic in the workflow (model stage 1 vs stage 2). The goal is consistency and control: you can structure when and how LoRA influence is applied across the pipeline without constantly rewiring the entire graph. In this workflow you’ll see dedicated IAMCCS LTX-2 nodes for model/LoRA I/O handling as well. 13d_IAMCCS_LTX2_LONG_LENGHT_VID…
Step 4 — Generate the first segment (the “seed” segment)
This is your segment 1: you run LTX-2 normally, decode to images, and you have your initial “source_images” batch. Everything after that is extension.
In the shared workflow, segment creation and packaging is repeated in a structured way, so you can preview intermediate videos too (CreateVideo nodes appear along the chain).
Step 5 — Extract the right frames for extension (Get Images From Batch)
This is where IAMCCS starts saving you time.
IAMCCS_LTX2_GetImageFromBatch is used to grab the exact portion of frames you want (most often: “from_end” to take the tail frames that will guide the next segment). In the workflow file you can see it configured in a classic extension-friendly mode (e.g., from_end with a count).
Why this matters: LTX-2 extension lives or dies on “what you feed forward.” If you feed the wrong slice, you’ll get weird seam timing, sudden resets, or continuity breaks. The node is built specifically for long-length iterative flows, and includes safety patterns that match stable native workflows.
Step 6 — Merge segments (Extension Module)
This is the heart of the long-length workflow.
IAMCCS_LTX2_ExtensionModule takes:
-
source_images (your accumulated video so far)
-
new_images (the newly generated segment)
-
an overlap_frames value
…and blends them into:
-
extended_images (the new merged total)
-
start_images (the guide frames for the next pass)
In the workflow, you can see multiple ExtensionModule instances, each one extending the chain.
A few practical notes (kept lightweight on purpose):
-
The overlap is where you control seam visibility (hard cut vs crossfade). The module supports different blend modes, including smoother filmic/perceptual options, but the workflow keeps it straightforward for stability.
-
There are built-in safety options to mimic proven “native stable workflow” behavior when you hit frame-count constraints. That is not theoretical—this is exactly the kind of thing that saves you from the classic encode errors in long runs.
Step 7 — Keep frame math consistent (seconds ↔ frames)
Long workflows become painful when duration and frame count drift.
IAMCCS includes a frontend sync system that keeps seconds and frame length aligned with FPS, and can also snap to LTX-2 constraints when needed. This is part of the “don’t let tiny math errors destroy a 40-minute render” philosophy.
Step 8 — AutoLink helps keep the graph readable
You’ll see IAMCCS AutoLink nodes in the workflow. They’re inspired by the Set/Get concept (KJ style), but adapted for IAMCCS usage to keep long graphs manageable. These are frontend-driven and designed to reduce cable mess, especially in iterative/loop-like sections.
Step 9 — Audio stitching (optional, but included)
If you’re producing segment videos and combining audio across iterations, you’ll see audio concat steps (AudioConcat) in the workflow so that the final assembly can carry forward a continuous track.
Step 10 — Final output video
At the end you output a final CreateVideo using the extended image batch and the stitched audio, with your chosen FPS.
What to do if you want cleaner seams (quick guidance)
If you notice visible seams, exposure shift, or a “restart feeling” at the join, that’s exactly why the Extension Module includes options like color/exposure matching and seam search modes. I’m not going deep here because that’s supporter-post territory, but the tools are literally built for those issues.
Closing
That’s it: a long-length LTX-2 pipeline that stays controllable, readable, and extendable—without you manually slicing frames and rewiring connections every time you add a segment.
If you want the full technical breakdown of the new nodes (including deeper reasoning, diagrams, and how to squeeze the best results out of each widget), check the dedicated node post— CLICK HERE:
https://www.patreon.com/posts/ltx-2-i2v-long-149232500?utm_medium=clipboard_copy&utm_source=copyLink&utm_campaign=postshare_creator&utm_content=join_link
and if you want access to those deeper posts, consider supporting. It’s the most direct way to fund ongoing development and documentation, and it’s what lets me keep sharing real production-tested workflows with the wider open-source community.
Follow for updates.