Grab
These paintings are based on a popular photo from a wrestling match sent to me by a friend.
Using two control nets, Depth (depth_midas) and SoftEdge (softedge_pidinet), I adjusted the weights to around 0.5 to let the figures move.
In order to capture the grab gesture and limb configuration, I took the images to Photoshop Beta and used generative layers for inpainting. This was more efficient than using Stable Diffusion. Photoshop's alternative ideas for filling parts could eliminate the need for creating hands and feet, which AI engines tend to struggle with.
Tech: Text prompts with Stable Diffusion: 60 steps, Euler a, CFG 5, 512x512, Virile Fantasy v1.1, Denoising 0.5. ADetailer with prompts face_yolov8n.pt. CN0 Depth depth_midas. CN1 SoftEdge softedge_pidinet. Hires 1.5x (768x768), 30 steps, 4x_foolhardy_Remacri. Post: Topaz Gigapixel HQ 4x (3072x3072). Adobe Lightroom. Adobe Photoshop AI Generative Layers Inpainting (limb fix).
Images
- 12 variations
- Instagram set
- Facebook post
- Twitter thrread
