The i2v_nsfw version of v7 has some undesirable effects such as inconsistent facial features of the characters, distorted faces, and numerous noise points in the picture.
The i2v_nsfw version of v7 has some undesirable effects such as inconsistent facial features of the characters, distorted faces, and numerous noise points in the picture.
I'm still tweaking the NSFW version to hopefully improve it, but I'd recommend using base V7 and picking/choosing specific LORAs for the best results. The "merge" NSFW is mainly for ease of use (e.g. "jack of all trades, master of none"). With that said, I usually only have 1 or 2 frames of noise at the start and it clears up pretty quickly. Make sure you are not doing anything else in your workflow that might be degrading quality.
Base v7, with excellent results. The prompt compliance and character consistency are good, especially the facial balance has been achieved. Thank you. Looking forward to a better NSFW version.
i have no idea how you merge loras and these video models anyway, and I hope you can atleast give us a list of what loras you merged. dont have to give the full names of them if you cant, but atleast say what they do. problem is, if we use this model and then by chance add a lora you merged or another lora that does the same thing, the weight will be overpowered, thus decreasing output quality.
your work is always appreciated.
If possible, please specify each nsfw_lora that is being merged. This way, during the trial period, we can avoid using the merged lora or having overly high weights, which might cause undesirable effects. Thank you very much.
What's the easiest way to connect multiple loras to your example workflow?
What's the easiest way to connect multiple loras to your example workflow?
You can try using the Power LoRa Loader as a node.
i'd also be interested if the "Jack of all trades" mix is as usual a exclusively a Jil of all trades mix. I think that also comes down to just have some form of list.
Dang.. I was hoping to use this as an image model. The first few frames being scuffed sorta kills that idea.
Dang.. I was hoping to use this as an image model. The first few frames being scuffed sorta kills that idea.
It is pretty easy to just drop the first 1-3 frames you generate, which are mostly static anyway. I'll be happy once I figure out how to clear it up, though.
T2V is working for me without bad frames. Just need to cut down the gen time.