Prepended to your prompt before encoding. Recommended for most images.
Model
Turbo defaults applied: 10 steps, CFG 1. You can still change them.
Aspect ratio
Snapped to multiples of 16, 512–1536 px.
Sampling
AdvancedNegative prompt · sigma shift
0 uses the model default.
LoRAUp to 3 · HF / Civitai / URLOff
HF repo (user/repo or user/repo/file.safetensors), Civitai model link or a direct .safetensors URL. Weight 0–2; empty slot = off.
ControlPose · lineart · inpaintOff
Control adapters are trained on Base v1.0. With the guidance may drift or be ignored.
Using recommended CFG & steps
Drop a photo or artwork with the pose you want, or browseWe’ll find the pose right away. Whole figures work best.
Drop a skeleton image or browseA colored stick figure on black, in the usual pose-map style.
Drag the joints to adjust the pose.
No person found
Try a photo or artwork where the whole figure is visible and not too small in the frame.
Fit
Style
People
The pose holds best around 1024 px.
Drop a photo or artwork with the pose you want, or browseA 3D body with fingers is estimated from it (uses a few seconds of GPU).
Estimating 3D body…
Couldn’t build a 3D body
Drop lineart, a sketch or a grayscale image, or browseA color photo as-is gives poor results — use lines or grayscale.
Preview in the proportions of your image.
Drop the image to repaint, or browseThen paint over the part that should change.
48
White = repaint, everything else stays. The mask is sent as a separate black-and-white image.
Active on steps
Your image will appear here
Write a prompt, pick a model and a frame, then press Generate.
Waiting
—
Session