Mech Girl机甲少女

The first step is creating clean anchor images for both the protagonist and the mech. These define identity, materials, silhouette, and tone for everything that follows.

第一步是为女主角和机甲创建干净的锚点图像，定义后续所有内容的身份、材质、轮廓与色调。

The first pass was done in GPT Image 2 to establish the girl's post-apocalyptic look and the mech's friendly rounded design.

首轮在 GPT Image 2 中完成，确立女主角的末世造型与机甲的圆润友好设计。

These images serve as anchor references — not final production assets, but the visual DNA of the project.

这些图像作为锚点参考，不是最终生产资产，而是项目的视觉基因。

Rough concept images are converted into cleaner, production-ready reference sheets using a reusable PDF prompt template.

用可复用的 PDF 提示词模板，将粗糙概念图转化为更干净、可用于生产的参考表。

The same character_reference_sheet_prompt.pdf is reused for both the girl and the mech to maintain identity consistency.

女孩与机甲复用同一份 character_reference_sheet_prompt.pdf，保持身份一致性。

Before generating video, combine both reference sheets into a single scale image so the size relationship between girl and mech is locked.

生成视频前，将两份参考表合成一张比例图，锁定角色与机甲的尺寸关系。

This prevents scaling issues in later shots and makes it easy to verify proportions at a glance.

避免后续镜头出现比例问题，一眼即可验证尺寸。

Using the two reference sheets and scale image as a three-image input set, generate two different 15-second variations.

以两份参考表和比例图作为三图输入，生成两段不同的 15 秒变体。

Prompts are designed to preserve character identity, mech design, and scale relationship across all shots.

提示词设计确保所有镜头中角色身份、机甲设计与比例关系保持一致。

The second pass explores stronger actions and ending options without restarting the asset pipeline.

第二轮探索更强的动作与结尾方案，无需重做资产流程。