# 2026-09 qwen-meshing — multi-reference composition demo

Five zoo characters (pig/frog/koala/duck/axolotl white-bg cutouts as `images.image_1..5`
references to TextEncodeQwenImage21) merged into one group photo, 1 seed, 25 steps cfg 1.
Identity, style (facets vs gloss), and colors all preserved per character. RGBA out again.
Graph: /var/lib/pi-kit/rerun-graphs/qwen21_multi_reference_group.json

- `forest_scene.png`: panda/otter/hedgehog/deer/cat (v4-era figurine refs) → colored storybook forest diorama, one seed. Cat: pupil-less eyes (single-seed flaw class).
- `group_shot.png`: zoo five (pig/frog/koala/duck/axolotl) group photo merge.