f2b08fecb3e274b79e9f2a0fac26b252843e2acc
QueuePanel reused one pixel TransformState for all non-fine-tuned items, mis-placing garments/prints of other sizes. Store the ratio-based template instead and recompute each item from its own garment/print size. - add core.composer.image_size() and resolve_transform() (pure, tested) - queue: set_template() replaces set_transform(); per-item resolve + size cache - template_panel: expose current_template(); main_window pushes it to the queue - tests: +4 covering fine-tune precedence, per-size recompute, empty inputs Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
cmbot
自动化图生图
Languages
Python
98.5%
PowerShell
1.5%