Plan for LTX 2 AIO NSFW?
The newly released LTX 2 seems to be faster and better than wan2.2, also with audio. Hopefully there is plan to make the AIO NSFW version.
phr00t doesn't have to do that, also, they haven't updated Hunyuan Video AIO
I've been experimenting a bunch with LTX2. I find it very promising. I do want to make an "AIO" for it, but it is a bit more complicated, since the checkpoint includes two VAEs (audio and video). There also isn't much for NSFW LORAs to mix in. If I do find something useful to put together, I will make a repository to upload it.
I've been experimenting a bunch with LTX2. I find it very promising. I do want to make an "AIO" for it, but it is a bit more complicated, since the checkpoint includes two VAEs (audio and video). There also isn't much for NSFW LORAs to mix in. If I do find something useful to put together, I will make a repository to upload it.
If the VAE is the complicating factor maybe you create an AIO version but load the VAE separately, by adjusting the workflow? Then in the meantime you can figure out how to include two VAEs.
I haven't had any luck mixing in the text encoder either, at least with ComfyUI nodes. I don't think an AIO is going to be that practical, since at best you'd still need a separate VAE audio loading node and separate file for it. I was able to make diffusion model merges successfully, so I might settle for that. It won't be an "all in one", but I'm hoping I can at least mix and match LORAs to make a better diffusion model. I am messing with distill strengths but struggling to get something worth uploading yet.
Better diffusion model would be great, we can manually load the text encoders and VAEs.
Thank you for the Qwen-Image-Edit-Rapid-AIO and WAN2.2-14B-Rapid-AllInOne versions you have been sharing for free on here :)
I have been checking often for new versions really appreciate the dedication and work you put into the new versions and sharing them for free.
Appreciate the effort for this, also for Qwen and WAN Rapids released before. Thanks.
lets' see if you can make a ltx-2 version that is fast, long, good looking - and not complete crap in every other aspect.
it went straight back to Sd1.5 days in anatomy, physics, world understanding and so on. Even if you prompt pamper it to the insane amount it requires. ("don't prompt abstract concepts like 'sad', prompt every single face muscle movement instead" because our model is too stupid to know what sad means. Or anything physics related, like things touching other things without everything glitch-exploding...).
LTX-2 is even worse than HV1.5 at that, but at least it has certain technical aspects that are indeed game changing.
however i add my thanks for this model. It was (and is for a quick job) a great alternative to base wan2.2)
I have moved on more or less to Wan2.2 remix, it is not as capable in nsfw stuff, but it remains fully compatible to wan loras and SVI Pro. And it's sooo much more lively and gap filling than base Wan2.2 (which was a big + of thse aio models too). Remix+svi pro It is actually my real game changer in the recent weeks. A video as long and longer as ltx-2 can do (if much slower ) but man is it less glitchy and prompt demanding for longer videos. It just works. .