LTX2.5 IC Outpaint LoRA (CRT AutoDL)

The clip ends too soon, or the shot is framed too tight - either way, you want the scene to keep going beyond the edge of the frame. LTX2.5 IC Outpaint LoRA (CRT AutoDL) is the adapter for exactly that: it loads the LTX-2.3 22B IC-LoRA Outpaint and applies it to your model so the generator can extend the picture beyond its original borders.

Outpainting is one of those jobs that's miserable without a model trained for it - you're asking the diffusion model to invent pixels that don't exist yet and have them blend seamlessly with what's already there. The IC-LoRA approach makes it a conditioning task: the existing frame becomes the reference, and the LoRA teaches the model to continue it outward in a coherent way rather than hallucinating a disconnected edge. In the LTX world, outpainting is one of the stock behaviors that the June IC-LoRA collection formalized, and this node is the "just apply it" version.

How it works

Standard CRT LoRA mechanics: you bring a model in and set strength_model (default 1.0). The node auto-downloads the LoRA from its HF repo on first run, applies it with the given strength, and returns the patched MODEL. There's no downscale-factor output here - outpainting doesn't carry the resolution math that the control-net LoRA does - so it's the simpler of the IC family: model in, patched model out.

It's an LTX-2.3 adapter that the CRT 2.5 stack reuses, which is worth knowing only because it means the outpainting behavior follows the same IC-LoRA conventions across generations. If you're doing frame extension, this is the node to reach for instead of trying to hack it with padding and inpainting.

Installing it

Part of CRT-Nodes:

cd ComfyUI/custom_nodes
git clone https://github.com/PGCRT/CRT-Nodes.git
pip install -r requirements.txt

or ComfyUI Manager → search CRT-Nodes, restart. First run downloads the LoRA from the oumoumad repack; watch the console progress bar.

Common issues

• Outpaint looks smeared - strength_model too high can overdrive the adapter. Drop it toward 0.5–0.8 and let the base model carry more of the continuation.
• Wrong conditioning - outpainting works by conditioning on the existing frame region. If your workflow isn't giving the model the "what's already there" context, the LoRA has nothing to extend from.
• Update breakage - red socket or NaN → right-click → Fix node (recreate).