UPDATE (03/19/23) : v1.1 seems to work much better with 768x768 resolution
This model allows you to generate some nice feet (soles specifically) in anime style.
Since the pose is very specific, I STRONGLY recommend you to use the following prompt structure :
1girl, sitting, feet up, soles, YOUR DETAILS, <lora>
Example : 1girl, sitting, feet up, soles, brown hair, blue eyes, white bikini, bedroom, <lora:feetPoseAnime_v11:0.6>
I suggest a weight between 0.6 and 0.8. It should work on most anime models (I got solid results with Anything v4.5).
Sample images setting :
Euler a
20 steps
768x768
Hires :
realESRGAN_4xplus_anime_6B
upscale by 1,5
denoising strength: 0,1
Keep in mind that AI always struggles with feet so you'll still get missing or extra toes, but overall it should provide good results.
Looking forward to see what you will generate with this, don't hesitate to share your images !
Description
Larger dataset training :
more detailed feet
better success rate with 768x768 resolution
FAQ
Comments (21)
a new version looks so good, what a blessing
仅仅对于正方形有泛用性
对于highheel和legwear这类的tags有强烈的至畸作用
What is the model of the first LoRA image?
Anything v4.5
@Notaledeji Hello, I use Anything-v4.5, but I can't make images in the same style as yours, could you explain to me how you got this result in Anything-v4.5?
@Juansigma Do you generate with 768×768 resolution ? That's what I do + Hires. And if it's not Anything v4.5, then it's AOM3 (Abyss Orange)
@Notaledeji Which Hires? Are many
@Notaledeji Which Hires? Are many
@Juansigma Check the description, I put all my settings 👍
Is it possible to add a label to distinguish between bent and unbent knees?
为什么这种手脚指很少出错呢?
Why are these fingers and toes rarely wrong?
Very curious to see what did you put in you're Datasets for Having such a High precision of creating this pose & Feet. Did you just use like the same image but just resized, cropped ? like less than 5 images ? Can be very helpful to know.
this pose is great! And the feet mix really well with other models!
I'd love to be able to do other poses, such as standing over you, lying down, different angles, etc.
Does it work with custom realistic models?
Thanks for this LoRa works really good !
This lora actually works better for me than the realistic one on realistic models. I found that including the realistic one but prompting the TE weight to 0 (IE: <lora:feetpose-000006:0.4:0.0>) cleans up the too many/too few toes.
ss
玉足是吧,我直接暴风吸入
I love this LoRa tyty
if i have more than 1 lora how do i make it reference this lora and no the other
Details
Files
animefeetv4.safetensors
Mirrors
animefeetv4.safetensors
animefeetv4.safetensors
animefeetv4 (1).safetensors
animefeetv4.safetensors
animefeetv4.safetensors
feetPoseAnime_v11.safetensors
feetPoseAnime_v11.safetensors
feetPoseAnime_v11.safetensors
animefeetv4.safetensors
feetPoseAnime_v11.safetensors
feetPoseAnime_v11.safetensors
animefeetv4.safetensors
feetPoseAnime_v11.safetensors
feetPoseAnime_v11.safetensors
feetPoseAnime_v11.safetensors
teefemina.safetensors
Available On (2 platforms)
Same model published on other platforms. May have additional downloads or version variants.





