LTX2.5 Temporal Upscaler (CRT AutoDL)
The half that doubles the frame count
- model
LTX2.5 Temporal Upscaler (CRT AutoDL) is the time-axis half of LTX2.5's upscaling pair. Where the spatial upscaler doubles resolution, this one doubles the frame count - it's a temporal upscaler for the latent space, loaded as a LATENT_UPSCALE_MODEL and auto-downloaded on first run into ComfyUI/models/latent_upscale_models/.
Think of it as interpolation that lives inside the latent representation instead of being bolted on after decode. Your sampler might generate at a modest frame count for speed, and the temporal upscaler doubles the effective duration/frame rate as a separate pass - smoothing motion between frames rather than leaving you with the stutter of a low-fps render. In the LTX2.5 stack it's the sibling of Spatial Upscaler: use both for a clean 2x in every dimension, or just this one when your bottleneck is motion smoothness, not resolution.
How it works
Like its spatial sibling, this is a pre-decode pass: it operates on the latent tensor, so it's cheaper than frame-interpolating pixels afterward, and the LATENT_UPSCALE_MODEL output plugs into the same latent-upscale machinery. It's the bf16 repack of the official LTX2.5 temporal upscaler, downloaded from the same UntMods/CRT_Nodes_25Gate repo that supplies the video/audio VAEs - so the CRT stack keeps all its LTX2.5 pieces from consistent sources.
No inputs to configure. Add it, let the download run, and wire the output where your workflow expects a latent upscale model for the temporal stage.
Installing it
Standard CRT install:
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. The download is small (a few hundred MB); the pack's heavy requirements install is the slower part.
Common issues
- Frame count didn't change - the temporal upscaler only does its thing if the workflow actually runs a temporal upscale pass. Check that you're feeding the latent into the upscale stage, not just loading the node.
- Spatial vs temporal mix-ups - this one is for duration/frame rate. If you wanted bigger frames, that's the Spatial Upscaler.
- Update breakage - red socket or
NaN→ right-click → Fix node (recreate).
Inputs (0)
No inputs
Outputs (1)
| Name | Type | Description |
|---|---|---|
| model | LATENT_UPSCALE_MODEL | — |