▸ txt2img

POST /image
Pure SDXL from the prompt. No source image. Baseline for "what does the model think this looks like".
WAITING · CLICK GENERATE
IDLE seed — · —

▸ img2img

POST /img2img
Re-imagines the source through the prompt. Strength adaptive 0.45–0.80 based on intent.
strength:
WAITING · NEEDS SOURCE
IDLE seed — · —

▸ InstructPix2Pix

POST /instruct-pix2pix
Edit-by-instruction. Uses the instruction string above, not the prompt. SEEKDEEP_FEATURE_INSTRUCT_PIX2PIX must be on.
img_guidance: adaptive 1.2–2.0 · steps fixed at 30 · 512→1024 upscale chain
WAITING · NEEDS SOURCE + INSTRUCTION
IDLE seed — · —

▸ Inpaint

POST /inpaint
CLIPSeg-masked SDXL fill. Pipes the mask target phrase to CLIPSeg for the mask. SEEKDEEP_FEATURE_INPAINT must be on.
mask threshold 0.3 · MaxFilter(21) dilation · GaussianBlur(8) feathering · strength 0.95
WAITING · NEEDS SOURCE + MASK TARGET
IDLE seed — · —