MuziGeek/selective-train-motion

車流靜止主體 · Selective Transit Motion

將照片主體轉為車流靜止主體，背景橫向運動，主體邊緣和臉部保持清晰。保留可辨身份與主要空間關係。

A sharp stationary subject contrasted with background-only lateral transit motion. Keep the original person or main object perfectly sharp while stretching only the existing moving background into coherent horizontal trails.

本方法改編自 gptimage25.ai contributors 的 MIT 授權說明。原作者歸屬、來源與改動見 provenance.json，完整上游許可及其版權聲明見 LICENSE-THIRD-PARTY.txt。

Adapted from gptimage25.ai contributors's MIT-licensed instructions. provenance.json records the original author, pinned source and modifications; LICENSE-THIRD-PARTY.txt retains the complete upstream license and copyright notices.

原始歸屬 · Original attribution: Copyright (c) 2026 gptimage25.ai contributors Adapted by MuziGeek from gptimage25.ai contributors; original-post attribution is retained where the collection provides it.

輸入分析與實際保留

1. 讀取使用者提供的原圖；若沒有可用圖片，先請求圖片。不要用本倉庫範例代替使用者輸入。
2. 觀察原圖主體、身份、數量、連接、遮擋、光線和畫幅。依據本 Skill 的保留約束與幾何邊界，區分需要保留及明確允許轉化的部分。樣片說明不能當作使用者輸入的事實。
3. 使用支援參考圖編輯的可用工具，把原圖作為編輯輸入。先說明觀察到的主體，再結合下方機制及 recipe.yaml 按順序組合指令；工具名稱不是必須安裝的依賴。
4. 預設沿用原圖比例；僅按使用者要求及本 Skill 明確寫出的構圖或幾何轉化調整。配方的 4:5 是目錄樣片比例，不授權裁去必須保留的主體。
5. 對照輸入檢查結果；只修正發現的問題，不為了增強風格改變身份或添加無關景物。區分已完成的視覺檢查與尚未驗證的效果。

Input analysis and preservation

1. Inspect the supplied image; request an image only when none is available. Never substitute a repository example for the user's input.
2. Inspect the real subjects, identity, counts, attachments, overlaps, light and framing. Distinguish preserved attributes from explicitly permitted transformations using this Skill’s preservation and geometry rules; sample descriptions are not facts about user input.
3. Use an available reference-image editing tool. Combine the observed facts with the ordered mechanisms in recipe.yaml; its model-family label does not require an installation script.
4. Keep the input ratio by default; change composition or geometry only as requested by the user and expressly permitted below. The recipe’s 4:5 ratio is a catalog-sample default, not permission to crop protected subjects.
5. Compare output with input and correct observed failures. Report only visual checks actually completed; stronger styling must not override identity or scene integrity.

風格機制 · Style mechanism

• 將照片主體轉為車流靜止主體，背景橫向運動，主體邊緣和臉部保持清晰。保留可辨身份與主要空間關係。
• 識別特徵：清晰主體、背景橫向模糊、正確遮擋邊緣。
• Medium anchor: A sharp stationary subject contrasted with background-only lateral transit motion.

10. medium — A sharp stationary subject contrasted with background-only lateral transit motion.
11. composition — Keep the original person or main object perfectly sharp while stretching only the existing moving background into coherent horizontal trails.
12. texture — Respect foreground occlusions and support points; blur must not smear faces, hands or flowers.
13. palette — Use a controlled palette appropriate to the requested medium; retain identifying source colors whenever the mechanism permits.
14. lighting — Keep illumination internally consistent across the entire edited image, with contact shadows appropriate to the requested medium.
15. detail — Retain source subject count, essential openings and connected anatomy; remove no identifying detail unless the permitted transformation requires it.
16. finish — Deliver one finished image at the input aspect ratio, without comparison panels, new lettering, logos, watermarks or an external frame.

保留約束 · Preservation constraints

在實際輸入觀察結果之上應用以下約束；清單不授權憑空添加圖片中不存在的元素。

Apply these constraints to observed content; they never authorize inventing absent subjects.

• Source subject count, recognizable identity, main pose and connected anatomy, subject to the explicitly permitted transformation.
• Essential openings, attachments and foreground-to-background relationships.
• The complete main subject within the input frame; no accidental clipping.

適用範圍與幾何邊界 · Applicability and geometry

• 適用：夜景。
• Applies to: Night.
• 僅改變明確指定的媒介、表面或光照；主體數量、可辨身份、姿勢、必要開口與相對位置保持。
• Change only the explicitly requested medium, surface or illumination; keep subject count, recognizable identity, pose, essential openings and spatial relationships.
• 未明確允許的結構變化均不屬於此風格；不把這些方法解釋為復刻某位藝術家、品牌或現成作品的授權。
• Structural changes not explicitly allowed above are outside this method. This is not permission to reproduce a named artist, brand or existing artwork.

排除項 · Avoid

• Invented people, extra limbs, duplicate main subjects or disconnected anatomy.
• Unrequested logos, decorative writing, watermarks or fabricated technical claims.
• A generic filter that fails to implement the specified physical or optical mechanism.

驗收 · Acceptance

• 背景橫向運動，主體邊緣和臉部保持清晰。 Respect foreground occlusions and support points; blur must not smear faces, hands or flowers.
• 僅改變明確指定的媒介、表面或光照；主體數量、可辨身份、姿勢、必要開口與相對位置保持。 Change only the explicitly requested medium, surface or illumination; keep subject count, recognizable identity, pose, essential openings and spatial relationships.
• 主體清楚可辨，連接自然；效果必須體現具體機制，不能只是通用濾鏡。 The subject remains legible with coherent connections; the requested physical or optical mechanism is visibly implemented.
• 只交付一張完整成圖，無無關文字、品牌、邊框或虛構測量數據。 Return one finished image without unrelated lettering, brands, frames or fabricated measurements.

配方、來源與許可 · Recipe, provenance and license

• recipe.yaml: structured image-editing recipe / 結構化圖像編輯配方。
• provenance.json: original authorship, method context and example status / 原創歸屬、方法背景與範例狀態。
• LICENSE: MIT applies to MuziGeek's original contributions / MIT 適用於 MuziGeek 的原創貢獻。
• LICENSE-THIRD-PARTY.txt: gptimage25.ai contributors · MIT. Adapted material retains these terms and attribution / 改編材料保留原許可和歸屬。
• 範例中的前後圖均為 AI 合成，不是真人攝影或真實製作工藝的證據。結果會隨模型與輸入變化，不承諾相同輸出。
• Example before/after images are AI-generated, not photographs of real people or evidence of physical craft production. Results vary by model and input.
• MIT 不授予使用者輸入、第三方作品、人物、商標或生成平台條款之外的權利；範例許可單獨見 provenance。
• MIT grants no rights to user inputs, third-party works, likenesses or trademarks, and does not override generation-service terms; example terms are recorded separately in provenance.

範例狀態 · Example status

ready：原圖與效果圖的真實生成及視覺審核記錄齊全；COS 外鏈範例另保留上傳回讀憑證，不在倉庫儲存圖片。

ready: the source and effect have recorded real generation and approved visual reviews. COS-hosted examples also retain upload readback receipts; their image bytes are not bundled.

AI 合成原圖 · Source	通過審核的效果 · Approved effect

• AI source / AI 合成原圖
• Generated interpretation / 生成效果
• Generation and visual review / 生成與視覺審核記錄