A workflow for adding sounds to existing videos, using my nsfw lora.
The workflow takes an input video and a text prompt, it outputs a video with the original frames and the newly generated audio.
Previews were generated with wan 2.2, audio added using this workflow.
Note: Changing JWMaskLikeImageSize's mask value to something like 0.1 could improve results, going too high could cause desyncing.
Description
NSFW (+furry) lora: https://civitai.com/models/2310920
Distil lora: https://huggingface.co/Lightricks/LTX-2/blob/main/ltx-2-19b-distilled-lora-384.safetensors
NVFP4 text encoder: https://huggingface.co/GitMylo/LTX-2-comfy_gemma_fp8_e4m3fn/blob/main/gemma_3_12B_it_nvfp4_uncalibrated.safetensors
LTX2 Dev fp4 checkpoint: https://huggingface.co/Lightricks/LTX-2/blob/main/ltx-2-19b-dev-fp4.safetensors
FAQ
Comments (17)
it always says invalid number of frames. 8 plus 1
Oh I didn't think of that. I'll probably find a fix and update it. You could use the vhs video loader, it has an ltx compatibility option. That's most likely how I'm fix this
I've replaced the vanilla comfy video loader with the vhs video loader
I cant get it to work, the audio is so bad.
Can it be used in wan2gp?
It should work in wan2gp but I have only tested it in ComfyUI.
Are you using the lora?
@mylo1337 i am using both loras yes
@Jdoe666 also if you're using it in wangp, does that work the same way as this workflow?
@mylo1337 its bad
@Jdoe666 How does wan2gp's ltx 2 foley work? Does it even support it?
Honestly the examples are bad, I getting the same generic (not actual blowjob slurping) moans and low quality ASMR clicking sounds without any loras.
sounds like a burning fire out in the woods
LTX 2.3 + Text to video suppory enytime soon?
2.3 version soon (Currently working on the dataset still) t2v depends on if that ends up working out lol. I have a decent idea for what to do for a higher chance of success, I would use muon, sadly it doesn't seem many tools support it. I'll just use a lower learning rate instead for now.
I'm also starting work on a trainer but the next lora will come first, since I'm not really working on the trainer that much.
@mylo1337 Love you. I'll test out the 2 first but looking forward for the 2.3
will this remove any already added audio? shouldn't be an issue either way since i can just add it back in post.
Yeah it removes added audio. Iirc it's not particularly well synchronized in a lot of cases, maybe with ltx 2.3 it's better, but I haven't tried that yet.
Any Update here, did anyone get this working well?