- Install Stable Diffusion (recommending automatic1111 webui, will be using it for the rest of this effortpost).
- Get a model (used above is Babes (hash 121ec74ddc)).
- Generate some basic images with prompts. Use danbooru tags, for example Maya was 'best quality, 1girl, large breasts, wide hips, hoodie, shy, black hair, hair over one eye'. Pick best one.
- Open up your favourite image editor (it's paint.net) and make some alterations (like paint hoodie red, add simple bun with brush).
- Copy your image into img2img tab (inpainting).
- Alter tags to reflect your changes (so ie. change 'hoodie' to 'red hoodie', add 'single hair bun'). You can increase/decrease focus on tag with (tag:value), base value is 1.0.
- Cover select areas with impainting brush. CHOOSE ONE FROM BELOW:
a) cover elements you want changed (like hair bun) then select 'Inpaint masked' mask mode
b) cover elements you want unchanged (like face) and select 'Inpaint not masked' - Start generating images until happy with current changes or you feel you're stuck with subpar results. Play with Denoising strength (how much of the original image is retained).
- Repeat 4-8 until done.
- Send image to extras tab, upscale it.
- Post it on bald man with glasses site.
Examples:
