V2.0 HuncusVid edition INFO : I finally deleted the model, after deeper testing (going beyond the first batch of low res quick test) it turned out the model was way too blurry, it wasn't as sharp as I expected when going higher resolution... Nothing lost, even if I liked the initial results, it was just not good enough in the end

V1.0 INFO below

A merge of Loras with Hunyuan fp8 cfgdistill model, mostly for testing and feedback, very likely I will delete this model soon, unless it turn out amazing (which I doubt, but maybe it has some good points)

• The goal is to use this model with various loras, and see IF there is any point in this model, I merged Move Enhancer V3 and V1.1 and also Str3etLIF3, plus couples of other loras at lower weight (none of the loras have been merge above .60 weight, and the whole is merged at 50/50 with Hunyuan model 3times, meaning the actual percentage of these loras is quite minimal but it still give some decent results (at least I hope)

• Seems to accept fast lora, seems to accept Teacache (but I didn't tested really, just noticed it was not crashing), probably accept wavespeed

• Probably not as great than base Hunyuan, if you add lots of loras, since it has already have a few loras merged in (that don't mean you can't add loras, just be more selective)

Demo images made with no added loras (except the fast lora at 0.6, just to get quicker results, but it should give better results without the fast lora and increasing steps)

Edit : Now SOME demo images use loras, just check the details...

Inspired by https://civarchive.com/models/1237378/huncusvid?modelVersionId=1394503 HuncusVid merged model (also thanks to his author for providing the merging template, you can find it on his model page)

My first intention was to upload a merge of Hunyuan custom (this one) merged with Skyreels t2v model, but after testing, the model was too unstable, producing way too much blurry garbage (and sometimes decent stuff) also it wasn't accepting loras as well, and was not working well with fast lora (but it worked with teacache), so I decided to upload this Hunyuan only (plus loras) intermediate model instead, which produce more consistent results and is more gentle with additional loras and stuff...

So far it seems to work plug and play replacing other hunyuan models in my cvarious workflow, no clip or vae included, it use the usual clip and vae of other Hunyuan models

It is less ambitious than HuncusVid model, far less number of loras merged into the hunyuan.