LTX2.5 IC Pixel Spatial Upscale LoRA (CRT AutoDL)
Bigger frames, learned detail
- model
- MODEL
- latent_downscale_factor
Spatial upscaling in the LTX world has a specific vocabulary, and this node is the one with the mouthful of a name: LTX2.5 IC Pixel Spatial Upscale LoRA (CRT AutoDL). It loads the LTX-2.5 22B pixel spatial upscaler IC-LoRA and applies it to your model, giving you a learned 2x spatial upscale - bigger frames with detail the model knows how to fill in, rather than just stretched pixels.
"Pixel spatial" is the key phrase, and it's how this differs from its siblings. The plain Spatial Upscaler node works in latent space before decode. This one is a pixel-space path implemented as an IC LoRA - the model itself performs the spatial upscale as a conditioning task, which is a different (and sometimes better-behaved) route to the same goal, depending on what your workflow is doing. It's part of the IC-LoRA family's "upscaler" tier, where Lightricks shipped adapters that fold a whole pipeline stage into a LoRA.
How it works
Standard AutoDL LoRA mechanics: model in, strength_model (default 1.0) in, patched MODEL out. Auto-download on first run from the pack's LTX2.5 repo. There's no downscale-factor output on this one - just model in, model out - which makes it the most fire-and-forget of the LTX upscaling options: apply, then run your upscale pass.
Where to use it: when your workflow already does pixel-space upscaling (after VAE decode) and you want the model's learned detail rather than a generic resize or a separate pixel upscaler model. If you're doing latent-space upscaling instead, the Spatial Upscaler node is the matching piece.
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 (a few GB).
Common issues
- Latent vs pixel confusion - "pixel spatial" means it participates in pixel-space upscaling. Wiring it into a latent upscale stage is a type mismatch that'll just do nothing useful. Pair with the stage it's named for.
- Weak effect alone - like all IC upscalers, it works as part of an upscale pass, not as a magic switch. Combine with the appropriate upscale pipeline step.
- Update breakage - red socket or
NaN→ right-click → Fix node (recreate).
Inputs (2)
| Name | Type | Default | Description |
|---|---|---|---|
| model | MODEL | — | |
| strength_model | FLOAT | 1.00-100–100 | — |
Outputs (2)
| Name | Type | Description |
|---|---|---|
| MODEL | MODEL | — |
| latent_downscale_factor | FLOAT | — |