Workflow : Long Video with Custom Audio
I2V & T2V - Long Extended Video with Custom Audio
Create longer shot with extension groups that extended the video incrementally to a longer length.
In theory infinite length, but the model will forget what it rendered in past windows, so realistically for minute or two shots.
Its not a simple or clean workflow, so might not be for everyone ;-)
(and basically extending 2-3,4 or 5 videos in a row, is heavy on a computer, so might not work for all)
Limitations
High experimental, and results may vary. There is a slight degrading of quality over time as well as introduced changes, and that might come apparent in the subject's identity (unless you create a character lora).
The extensions also have no idea what happened in extended groups in in far past (lets say you are generating extended video in group 3, it has no idea what went on in group 1).
Its less ideal for anything where you go to one place, and return to same place later. More suitable if the subject is constantly moving to new places, like say a person walking through the city, or if the subject is staying in same place using the IC Static Camera Lora (for example a scene with dialog, singer or similar where the camera is not moving)
Probably mostly usable if you are generating multiple clips to combine in video editor, and need one or two longer shots ;-)
But if you are in the experimental corner, feel free to play around : https://huggingface.co/RuneXX/LTX-2.3-Workflows/
i tried using this workflow and it followed the prompt i gave it but it just produced a very fuzzy video after the first few seconds. You where unable to make out the content. all your older workflows worked well
Probably wrong model then? dev model? Its made for distilled model. Unless you also enable the distilled lora (right below the main model loader, to be used if you use dev model)
(and look over the models so you dont have some LTX-2.0 in there. LTX-2.3 needs its own new video vae, audio vae etc etc)
That being said, its a bit experimental at the moment, will clean it up and simplify it a bit, but was the first draft (and a bit spaghetti under the hood)
yer that was it thanks for all your hard work
NEW
I2V & T2V Long Length Video - Single Pass loop extender
New much lighter and simpler workflow for long video generations with custom audio.
And some improvements and fixes, with some inspirations from Kijai.
For prompting, prompt something like "static camera. Focus on main subject" or something like that if you want a singer video.
(might come some loop improvements later on if added to Comfy)
Will update the 2-pass long video workflow asap ;-) Should work far better after the update
This is all very cool, but the actual lip sync on the uploaded audio only works on one model: input_scaled_v3 fp8 - everything else doesn't work. I really need a dev fp6 model that works with lip sync, and a gguf model because the sound it generates is terrible.
for low resolution, I suggest to upscale image from last image loop to keep quality. then resize again according setup width x height
Or combine all video loop first, then save to latent and preview video, if the result ok then load the latent and scale using scaling latent model, and trim when scale latent. I think this ways make more stable for low VRAM.
I2V & T2V Long Length
Video - Single Pass loop extender. the upscaler is not used, right? cmiiw
At single pass no upscaling true ;-)