@psspsspsspssspss Is there a workflow for this "inpainting with spatial masking"? I've looked all over civ, in comfy's samples and some search results, and couldn't find a single example, only video extension. I tried reverse engineering the basics and it just returned the same video, only audio would change significantly when applying the usual techniques, video seems to accept just x2 upscale.

It's not a 2.3 thing by any chance, is it?

@firemanbrakeneck 

It's not specific to 2.3 no, you use these two nodes from the official LTX nodes:

Create masks for the area you want to inpaint, then feed them to "🅛🅣🅧 LTXV Preprocess Masks" then feed those masks to the spatial mask input of "🅛🅣🅧 LTXV Set Audio Video Mask By Time" ... set the end time past the length of the video you are inpainting over. Check mask_video to true.

@psspsspsspssspss I'm not even there. I tried their official workflow (https://github.com/Lightricks/ComfyUI-LTXVideo/blob/master/example_workflows/2.0/LTX-2_V2V_Detailer.json) , almost as is, only some swaps due to limited vram: load unet Q5KM + vae ltx2 bf16 + dual clip gemma fp8 / 19b embeddings connector, with the detailer and distill loras.

All it does is make the text in the video a little noisy (possibly because the video is downscaled). Nothing that could be classified as V2V.

@NRDX Do you use another node to generate the mask on an input video? For example, the one from Kijai that let's you use control points? Or do you use something else for segmentation?

I found your repo on huggingface, there's definitely a lot more info on your model card there. I notice you have other links in your profile here, have you considered adding the url to your user account from HF to your CivitAI profile? For anyone else, you can just search Models on huggingface.co for LTX-Loras