第三阶段改进

prompts/complete-v1.txt

@@ -1,5 +1,8 @@
-Complete only the hidden portion of the specified character part on the first
-input image (the local canvas). Any subsequent images are context references.
+Repair and complete the specified character part on the first input image.
+The task declares editable regions for segmentation gaps, occlusion, joint
+extensions, and removal of overlaid attachments from their underlying base.
+Only regions explicitly marked editable may be changed. Preserve all other
+visible pixels. Any subsequent images or labeled reference panels are context.
 Preserve the exact original pose, scale, screen direction, perspective, colors,
 line thickness and shading. Do not mirror, rotate, resize, recenter or redesign.
 Keep the canvas dimensions and the position of every already-visible feature.
@@ -8,5 +11,6 @@ body parts, props, labels, annotations or extra versions of the same part.
 If an edit mask is supplied, transparent mask pixels are editable. Otherwise,
 use the task's bounded extension instructions and keep the remaining background
 pure white or transparent. Return the same canvas size without recentering.
-Downstream checks will reject misalignment, restrict additions to allowed regions,
-and restore the source visible pixels. Do not rely on restoration to hide changes.
+Downstream checks reject misalignment and constrain modifications to edit regions,
+restore protected source pixels, and reverse the documented canvas transform.
+Do not rely on restoration to hide movement or redesign.