You'll need to download ComfyUI-LTX-Reframe nodes from my repo:
https://github.com/exportAnything/ComfyUI-LTX-Reframe
The node pack includes:
LTX Reframe Layout- Selects a source video, previews and serializes its placement, optionally selects an LTX-compatible frame range, and outputs padded frames, masks, audio, FPS, video, target dimensions, and placement metadata.
LTX Reframe Target Fit- Center-fits an image batch to the exact target canvas dimensions.
LTX Reframe Rectilinear Correction- Applies a deterministic, mask-aware barrel-to-rectilinear correction to a generated frame batch.
LTX Reframe Custom Audio Loader- Loads optional custom audio only when its lazy workflow branch executes.
The models needed for this workflow are linked in the repo.
***UPDATE*** Users experiencing magenta output is caused by substituting the workflow’s expected 3K LoRA with the 10K version; those weights are not interchangeable in this workflow. Thank you EKKIVOK for figuring this out.
Description
FAQ
Comments (12)
so this uses outpaint ic-lora at core, or?
ic-lora in-outpaint. If you use outpaint lora, you'll have to toggle the lora very high (>1.00 strength) and it still may not work correctly.
Not working magenta square in output even codex cannot fix the problem...
I'll investigate. If i had to guess right now, it might have something to do with the upscale stage set with a denoise of >0.1
@exportAnything My bad i used the same lora but trained at 10 000 steps instead of 3000....
this broke the entire output ! but that reveal a weak point the kornia importation maybe broken on a lot of config.
@EKKIVOK The workflow works perfectly fine for me with any old random video I have in my folders. Codex suggested a codec issue (browser/GPU video decoding). Otherwise, it could also be that you're using an older LTX VAE. It could literally be just about anything. The specific models you need for the workflow are listed and linked in the repo.
@EKKIVOK I'm just seeing this! I've noticed LoRA's trained with >3000 steps break a lot of different workflows of mine (especially skin lora). Thanks for investigating this!
@exportAnything
I have a few other recommendations (thanks for the credit in the description ^^).
There’s a node called “Stage 1 Half-Resolution Mask” that requires uploading something, but it’s not clear what users should put into it. Is it an image? A video? You didn’t mention that.
Another issue: you seem to be using Node 2.0, which is highly recommended to disable because it can create many nodes and cause compatibility problems (especially with the group I mentioned earlier). Not everyone knows that.
The sound part is also a bit unclear. What does “STEM off and on” mean? There’s a custom audio option too, but users will probably wonder:
How do I use it?
Can it be combined with STEM or not?
How does it work?
These are just some observations ^^ I’ve created a lot of workflows for the community and the feedback is often similar to this (and I’m asking these questions for myself too lol).
Great work anyway!
@EKKIVOK I always overestimate the clarity thinking that people will understand what everything is. Ill reupload with clearer instructions.
@exportAnything you're the best bro !
Where to download all necessery files from this WF ? loras etc ??
Go to the repo here: https://github.com/exportAnything/ComfyUI-LTX-Reframe/blob/main/MODEL_DOWNLOADS.md
There's a list and each item there, and they are linked to their official repository for download.