33 poses in total, and each pose has a preview. All the previews are generated using the corresponding canny.
The previews are anime style, but most should be fine with generating realistic images.
The scales are not fixed (2:3, 3:4, 1:2), but it is not strictly required to generate image with the same scale. Set the resize mode to resize and fill if you need.
The recomanded weight is 0.4~0.8. It depands on whether you want the pose to be more similar to the canny or you want to generate a character with a different body type.
一共33个姿势，每个姿势都有一张preview，preview都是用对应编号的canny生成
Preview是2d，但理论上写实也没问题
canny图的比例并不固定（有2:3，3:4，1:2），但并不严格要求生成同一尺寸的图，如有需要在ControlNet的缩放模式中选择填充即可，大部分情况下都能适用
控制权重推荐0.6，但如果生成的姿势一直有偏差可以尝试0.8，如果需要生成的角色体型与canny图相差较大则可以尝试0.4
喜欢的记得点赞评论哦
Description
FAQ
Comments (6)
Awesome! Thank you ❤️👍
Please tell me where to download the model? You only have pictures here.
Canny is not an actual 'model'. It's a control type of ControlNet. You need to install the ControlNet module first.
really good, thank you!
Very good poses that do exactly what they are supposed to on realistic images. Thank you for sharing your work. Please tell me how you made the poses in the first place.
I first used some photos of real models to generate anime image by using anime style ckpt. Anime objects always have a much sharper edge in color, which makes it very easy for the canny model to draw the outlines of the characters.Then i download the outline image and wipe out the details like eyes, mouths, shadow, etc. As you've already seen, when you send the processed images to canny model, it comes out with great results.