編集とスタイル変換

スケッチで構図を組み直す

1回の編集で、アップロードした場面の物を別の配置スケッチに合わせます。

更新日: 2026/09/22

プロンプト全文

英語のプロンプトに入力例を添えています。生成前に被写体や詳細を調整してください。元の作例の入力内容と生成設定は完全には確認できていません。

推奨モデル: GPT Image 2

This generation is one edit and needs two uploads: the original scene, and a separate rough sketch of the new arrangement. Keep those roles apart. Move the objects that already exist in the original scene into the positions, sizes, and overlaps drawn in the sketch. Preserve each object's identity, colors, materials, and visible details from the original scene. Do not invent replacements for marks that the sketch only suggests as blobs. Ignore the sketch's paper color, line texture, and any handwriting. Hold the original camera height and lighting unless the sketch clearly changes the viewpoint. Add no new objects from the sketch; its marks guide only the layout of existing scene objects. Stop after this recomposition. A later cleanup is a new generation: upload this output as the current image, and keep the original scene on hand so you can compare what changed. User request: Recompose my uploaded cafe scene so the two chairs face the window, the plant moves to the left edge, and the table stays centered, matching my sketch.

プロンプトのヒント

この編集では、元の場面とスケッチを別々の参照画像として2枚アップロードしてください。
あとで整えるときは新しい生成を始め、この生成結果を現在の画像としてアップロードしてください。

出典とクレジット

OmniArt が構成を調整し、書き直しました。出典、変更内容、利用条件は以下をご確認ください。

Recompose a scene from a sketch · Synthetic_Copy
wangrunlin/awesome-gpt-image-2-5-prompts
Repository MIT covers tooling and original documentation only; community prompts and third-party media retain their original rights and terms (CONTENT_POLICY.md).
Rewritten as an original OmniArt single-edit workflow. It requires two uploads, the original scene and a separate layout sketch, and a later cleanup is a new generation that reuses the output manually. Localized into 19 locales. No source wording or media copied.

出典とクレジット