Imagild original image-model comparison worksheet / Imagild 原创图片模型小样评分表
Prepared: 2026-10-16

This is an empty reader worksheet. It contains no generated outputs, measured
model scores, provider benchmark, or claim that one model is better.
这是供读者填写的空表。没有生成结果、实测模型评分或供应商排行榜。

1. Keep one approved source and one saved brief for each scenario.
   为每个场景固定一份已确认的源图和一份带版本号的任务说明。
2. Have a collaborator replace model names with aliases A and B. Keep the
   alias-to-model mapping separately until the review is complete.
   由协作者用 A、B 代号代替模型名，评审完再打开单独保存的对应关系。
3. Record actual output dimensions. A 2K menu label is not a measured width.
   记录文件实际宽高。菜单中的 2K 标签不能代替实际像素测量。
4. Use 0 = unusable, 1 = needs repair, 2 = meets this brief. Score identity or
   object fidelity, the requested change, and artifact control independently.
   0=不可用，1=需修复，2=符合本次要求；分开评人物/物体保真、指定改动和瑕疵。
5. A serious identity, wording, or geometry error can make delivery unacceptable
   even when the sum is high. Write the visible reason in reviewer_note.
   即使总分高，严重身份、文字或几何错误仍可判交付不合格；写明可见原因。
6. Compare only the scenarios you actually ran. One A/B pair cannot establish
   a general success rate, model ranking, or expected cost for other inputs.
   仅比较实际执行的小样；一对结果不能证明普遍成功率、模型排名或其他素材的成本。

All score fields are intentionally empty. The example source IDs are fictional.
评分栏有意留空，示例源图编号是虚构的。模型调用和收费生成未由本文件执行。