Qwen-Image-2.1 Edit 被指总照搬参考图人脸,难以融合身份
Real-Bed-1815 · reddit · 2026-09-25
A user running the Qwen-Image-2.1 Edit workflow in ComfyUI reports that the model keeps copy-pasting faces from reference images instead of blending two identities into one merged character, and prompting alone can't fix it. They're asking the community whether this is a model limitation or a workflow error — a useful data point for identity-blending tasks in image editing.
「多模态」频道最新
- 用 GPT-6 Astra 加 Three.js 开源复刻概念 X-16 四排缸引擎 — techartist_ · 2026-09-25
- Asake AI 生成歌曲席卷 TikTok，UGC 视频超 3 万条 — saheedniyi_02 · 2026-09-25
- 「让 Claude 做一部西方文明史视频」引推圈惊叹 — MaxUnfried · 2026-09-25
- BuildingBench 榜单易主:Opus 5.5 得 86.8,成本是竞品 6 倍 — ZhitingHu · 2026-09-25
- 视频模型为何违反物理？去噪前 10% 已锁死轨迹 — linoy_tsaban · 2026-09-25
- Gemini 3.8 Flash 可按文字/视频/图片生成 MIDI 导入 GarageBand — DynamicWebPaige · 2026-09-25