Wan2.2 Skinny/Fat Slider LoRA
V1
Adjust body fat effortlessly:
Negative (-2 to 0): Slim to bony (use prompt to help, lower than -2 to starts to degrade the image but I got some results even at -4)
Positive (0 to 3): from big to huge, (use prompt to help, bigger than 3 works in some images)
Note: May slightly alter composition; avoid extreme settings for best results, test until you understand the impact of each value.
Note 2: You can add bigger values to High noise than low noise, usually I use same on both, but it works.
Description
First version
FAQ
Comments (23)
What script do you use to map out 2 lora values ? ( im assuming you merge 2 )
And do you train on images for this one ?
Just curious to try as well
Can you do de-age lora?
Nice try FBI
Wan does a good job of generating people of varied age all on its own. Age loras are not necessary.
This would be so much better if it were trained on real images and not generated images. Nice work though!
Sliders don't use images for training and don't add new information to the model
@terrariyum Very confidently incorrect
@playtime_ai can you explain? There are definitely methods to train sliders that do you not use images for training - you can see Ostris's ai-toolkit recent slider training video for an example. The LORA creator used ai-toolkit for this training, so presumably they didn't use images, real or ai generated for this training.
@CallMeMaybe if they didn't, then the resulting videos would like like every other wan video... Like real people. They don't. Just because there is a method to make a slider without input data does not mean that this one was trained that way. This looks like it was trained on synthetic data
@playtime_ai gotcha, thanks for explaining!
@CallMeMaybe It was not trained on synthetic data. It was trained using the Ostris method.
@alcaitiff Fair enough. Maybe the method just produces outputs like that. I'll look into it for myself.
@playtime_ai the idea is to push the boundaries, normal results we do with prompts. And small str values do very natural results.
@playtime_ai Very confidently incorrect
:P
bruh
what?
@alcaitiff I second alcaitiff's "what?" and add a "your point?"
Have you considered a penis size/thickness slider?
Would love to see that since WAN is notorious for thin ones and doesn't play the best with attempting to prompt girth.
Closest I've seen to this is character loras like futa and the shrek character loras.
Great! Thank you very much! I hope to come up with a lora that can make the skin fairer or darker
Going to negatives (skinny) tend to have the long torso side-effect. Any idea how to avoid it? Thank you
Very awesome slider! Please make them for any versions of Flux and Qwen! I use this to generate the initial images at -1 or -1.5 then have the others add details with low denoise because the others can't generate them thin enough.
Addendum: It looks like Flux Dev is covered.
Details
Files
Wan2.2_skinny_fat_slider_v1_high_noise.safetensors
Mirrors
Wan2.2_skinny_fat_slider_v1_high_noise.safetensors
Wan2.2_skinny_fat_slider_v1_high_noise.safetensors
skinny_HI.safetensors
Wan2.2_skinny_fat_slider_v1_high_noise.safetensors
Wan2.2_skinny_fat_slider_high_noise.safetensors
Wan2.2_skinny_fat_slider_v1_high_noise.safetensors
Wan2.2_skinny_fat_slider_v1_high_noise.safetensors
Skinny-Fat-Slider_t2v_High-Noise.safetensors
Wan2.2_skinny_fat_slider_v1_high_noise.safetensors
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.