|
|
|||
| forums: groups: | |||
|
"Black Gunged gifs (Wan2.2)"
Love you, too
Smooth black gunge, comin' at ya. Or comin' at some lovely ladies, rather 
These are from the new version of the Wan video model, 2.2. I'm getting a *lot* better results from image-to-video for this one, and it's compatible with Wan2.1 custom loras, so I didn't have to re-train.
Image-to-video is fun because it unlocks "clean start" generations, with more variety of style than I'd been able to manage before. These are all extended from random clean images I'd made along the way, with Flux or even Stable Diffusion in some cases. Might have to get back into the image generating game again!
I haven't gotten the splattery stuff working as well, but for these smooth pours, they're going pretty well I think. Enjoy!
These are from the new version of the Wan video model, 2.2. I'm getting a *lot* better results from image-to-video for this one, and it's compatible with Wan2.1 custom loras, so I didn't have to re-train.
Image-to-video is fun because it unlocks "clean start" generations, with more variety of style than I'd been able to manage before. These are all extended from random clean images I'd made along the way, with Flux or even Stable Diffusion in some cases. Might have to get back into the image generating game again!
I haven't gotten the splattery stuff working as well, but for these smooth pours, they're going pretty well I think. Enjoy!
You've got some good results there. I appreciate how much effort it takes to train it.
I keep meaning to training a wan Lora but keep getting distracted by new image models coming out. I've had my best results with HiDream but Qwen-Image is looking like it's going to be amazing after doing a small test training. Running training on Qwen for the next week.
I did a quick test of Wan2.2 too and that looks very promising as just an image model. I think Once Qwen finishes training I'll run a larger Wan2.2 model.
Out of curiosity, how big is your training data set?
I keep meaning to training a wan Lora but keep getting distracted by new image models coming out. I've had my best results with HiDream but Qwen-Image is looking like it's going to be amazing after doing a small test training. Running training on Qwen for the next week.
I did a quick test of Wan2.2 too and that looks very promising as just an image model. I think Once Qwen finishes training I'll run a larger Wan2.2 model.
Out of curiosity, how big is your training data set?
messg said: Out of curiosity, how big is your training data set?
Not big. I think it was 40 images or so. Honestly it was a pretty tacky attempt, it just worked well enough to give results I was happy with, so I've kept it.
Good to hear your optimism about Qwen, I'm curious to try that out now too . . .
wammypinupart said: Smooth black gunge, comin' at ya. Or comin' at some lovely ladies, rather
These are from the new version of the Wan video model, 2.2. I'm getting a *lot* better results from image-to-video for this one, and it's compatible with Wan2.1 custom loras, so I didn't have to re-train.
Image-to-video is fun because it unlocks "clean start" generations, with more variety of style than I'd been able to manage before. These are all extended from random clean images I'd made along the way, with Flux or even Stable Diffusion in some cases. Might have to get back into the image generating game again!
I haven't gotten the splattery stuff working as well, but for these smooth pours, they're going pretty well I think. Enjoy!
These are from the new version of the Wan video model, 2.2. I'm getting a *lot* better results from image-to-video for this one, and it's compatible with Wan2.1 custom loras, so I didn't have to re-train.
Image-to-video is fun because it unlocks "clean start" generations, with more variety of style than I'd been able to manage before. These are all extended from random clean images I'd made along the way, with Flux or even Stable Diffusion in some cases. Might have to get back into the image generating game again!
I haven't gotten the splattery stuff working as well, but for these smooth pours, they're going pretty well I think. Enjoy!
what exactly did you call this fluid when prompting the AI? a lot of times it will object to covering a person in 'oil' but will gladly use chocolate syrup for example.
the real GoOfBaLL said:
what exactly did you call this fluid when prompting the AI? a lot of times it will object to covering a person in 'oil' but will gladly use chocolate syrup for example.
what exactly did you call this fluid when prompting the AI? a lot of times it will object to covering a person in 'oil' but will gladly use chocolate syrup for example.
I'm making these with Wan Video installed on a private computer, so it doesn't object to anything
I think I just called it "smooth black gunge."
9/19/25, 4:25am: No-bump reply
Design & Code ©1998-2025 Loverbuns, LLC 18 U.S.C. 2257 Record-Keeping Requirements Compliance Statement
Epoch Billing Support Log In