Add more LoRAs from Civitai
Add the images on the right. Start image: img2img, the result keeps its layout and colors. Reference: IP-Adapter, the result follows its style and subject.
Pick one of your People: the result keeps their face (Face ID). Use it instead of a Reference image, with a prompt that shows the face.
Add a control image on the right, or draw one. The result keeps its edges, depth, pose or lines; the prompt decides the rest.
About twice as slow. The upscaled size counts against your size limit.
Finds each face / hand and redraws it at full size. To fix an existing image: put it in Start image and set Repaint (start) to 0.
Variants: write {a|b|c} in the prompt — every combination becomes its own job, × Runs. E.g. a {cat|dog} in {snow|rain} = 4 prompts.
{a|b|c}
a {cat|dog} in {snow|rain}
Compare: the same settings on other models (LoRAs only carry over to models of the same kind).
Compare schedulers: the same job on each ticked scheduler too (Karras carries over where it applies).
This site makes AI images, edits and audio, and some models can make adult content. To use it you must be 18 or older and agree to the Terms of Service (and our Privacy Policy). In short: