ASMR 動畫結合卡通角色新 AI 實驗

本次實驗由 DotAI 團隊發起，嘗試結合 AI 影像生成技術與網路熱門創作題材「番梘切割 ASMR」，透過 Veo3 工具，製作出一段以「史迪仔」為主角的沉浸式動畫影片。我們不僅進行畫面生成測試，更針對三項創作難點進行技術驗證：

• ✅ 角色還原精準度

• ✅ 鏡頭運動與「一鏡到底」連貫性

• ✅ 畫面與 ASMR 音效的同步協調

本文將全面剖析創作流程、錯誤復盤與提示詞調校技巧，協助內容創作者、行銷團隊與 AIGC 使用者真正掌握 AI 動畫的生成原理與控制方法。

最終成果:

Veo3 操作教學｜3 步啟動 AI ASMR 動畫生成流程

若你亦想親身試玩 Veo3，製作如「切番梘動畫」這類沉浸式 AI 影像，以下是基本操作流程，無需任何動畫基礎，三步即可開始創作：

✅ Step 1｜開啟 VPN 並前往 Gemini 官方平台

由於 Gemini 為海外平台，香港用戶需先開啟 VPN（建議使用美國或新加坡節點）
👉 https://gemini.google.com/app

✅ Step 2｜輸入提示詞（Prompt）或參考圖像

在平台中輸入提示詞，然後開啟 Video 例如：
continuous single take, close-up macro shot of a furry blue claw cutting a navy-blue soap bar, Pixar style
如欲提升還原度，可同步上傳參考圖像作為風格輔助。

✅ Step 3｜點擊生成，等待影片輸出

確認指令後，點擊「Run」開始生成。數十秒後，即可下載生成 AI 影片

💡補充小貼士：如果你想免費試玩 Veo3，其實可以透過我們的教學，免費申請到 15 個月的 Google Gemini Pro 帳號。只要啟動 Gemini 帳號，即可無成本使用 Veo3 等生成式影像工具，搭配我們的提示詞模版，立即開始製作專屬的 AI ASMR 動畫。📦

15 個月的 Google Gemini Pro 免費領取教學: https://dotai.hk/blog/gemini-ai-pro-free-20250617

AI 錯誤不等於不可控｜關鍵在於提示詞設計與錯誤診斷力

在 Veo3 的實測過程中，我們發現 AI 生成錯誤（如角色變形、鏡頭斷裂、風格跑題）並非隨機現象，而是與提示詞的語意結構與引導策略密切相關。只要掌握適當的提示詞撰寫方法與錯誤判斷流程，便能有效提升畫面品質與生成穩定性。

五項實測重點修正建議：

1. 角色還原失真？
   　→ 加入角色專屬細節（如「毛茸茸藍色爪子」、「Pixar 風格 3D 爪形」）並排除如「human hand」、「realistic」等負面詞語。

2. 鏡頭不連貫？
   　→ 將「continuous single take」、「unbroken cinematic shot」置於提示詞開首，強化導演語法。

3. 動作不一致？
   　→ 簡化描述，集中於「soap cutting」等單一連續行為，減少過度複雜動作指令。

4. 風格錯位？
   　→ 加入參考風格如「Pixar」、「cinematic lighting」，保持動畫一致性。

5. 反覆失敗？
   　→ 每輪僅微調一項提示詞，進行逐步優化，避免重複無效生成。

📌 總結觀點：AI 創作不是「一試即中」，而是一種可進化的流程管理。正確運用提示詞結構與修正邏輯，方能掌控 AI 影像創作的可預期性與品質穩定性。

Prompt:

An ASMR-style video focusing on precise, satisfying soap cutting — with equally immersive ASMR audio. A single, unbroken, static macro close-up shot in a bright, minimalist white studio. The camera is tightly framed on the hands of Stitch — a cinematic CG version exactly matching his appearance in Disney’s *Lilo & Stitch* movie. His hand features four expressive fingers with curved claws, proportioned anatomically like a small, stylized creature. The fingers show defined knuckles, gentle skin folds, and realistic motion creases. The claws are semi-matte and dark, slightly translucent, with no plastic gloss. His fur is short, dense, and layered in texture, with slightly uneven strand lengths and clear directional flow. Fur thins naturally near the finger joints and claws, exposing skin texture beneath. Each strand subtly catches light without overexposure, producing natural shading and depth. Lighting is soft and cinematic: a diffused three-point setup with no blown-out highlights. Shadows gently wrap around the hands, highlighting fur depth, surface texture, and object interaction. In his left hand, Stitch holds a navy blue rectangular bar of soap. The surface is embossed with shallow, evenly spaced square grid grooves, dividing it into perfect cube-shaped segments. Centered across the top is the raised logo text “DOTAI,” with each letter spanning multiple cubes. Using a small carving knife in his right hand, Stitch slices down precisely along each groove — each motion clean and continuous. The blade systematically removes the “DOTAI” logo letter by letter. Every cut breaks off a cube of soap with geometric perfection, forming a neat pile of satisfying square shavings. The logo is fully erased, leaving behind a flawless soap surface. **ASMR Audio**: Close-mic’d and isolated. Crisp slicing sounds, dry crackles as the knife enters the grooves, and satisfying snapping crunches as the cube pieces break free and fall softly. No music. No voice. (no subtitles) Camera: locked macro eye-level shot, with a slight angle revealing both fur texture and soap relief. Style: 4K cinematic, photo-realistic CG, film-grade creature rendering, ASMR-focused, satisfying and immersive

An ASMR-style video focusing on precise, satisfying soap cutting — with equally immersive ASMR audio. A single, unbroken, static macro close-up shot in a bright, minimalist white studio. The camera is tightly framed on the hands of Stitch — a cinematic CG version exactly matching his appearance in Disney’s *Lilo & Stitch* movie. His hand features four expressive fingers with curved claws, proportioned anatomically like a small, stylized creature. The fingers show defined knuckles, gentle skin folds, and realistic motion creases. The claws are semi-matte and dark, slightly translucent, with no plastic gloss. His fur is short, dense, and layered in texture, with slightly uneven strand lengths and clear directional flow. Fur thins naturally near the finger joints and claws, exposing skin texture beneath. Each strand subtly catches light without overexposure, producing natural shading and depth. Lighting is soft and cinematic: a diffused three-point setup with no blown-out highlights. Shadows gently wrap around the hands, highlighting fur depth, surface texture, and object interaction. In his left hand, Stitch holds a navy blue rectangular bar of soap. The surface is embossed with shallow, evenly spaced square grid grooves, dividing it into perfect cube-shaped segments. Centered across the top is the raised logo text “DOTAI,” with each letter spanning multiple cubes. Using a small carving knife in his right hand, Stitch slices down precisely along each groove — each motion clean and continuous. The blade systematically removes the “DOTAI” logo letter by letter. Every cut breaks off a cube of soap with geometric perfection, forming a neat pile of satisfying square shavings. The logo is fully erased, leaving behind a flawless soap surface. **ASMR Audio**: Close-mic’d and isolated. Crisp slicing sounds, dry crackles as the knife enters the grooves, and satisfying snapping crunches as the cube pieces break free and fall softly. No music. No voice. (no subtitles) Camera: locked macro eye-level shot, with a slight angle revealing both fur texture and soap relief. Style: 4K cinematic, photo-realistic CG, film-grade creature rendering, ASMR-focused, satisfying and immersive

An ASMR-style video focusing on precise, satisfying soap cutting — with equally immersive ASMR audio. A single, unbroken, static macro close-up shot in a bright, minimalist white studio. The camera is tightly framed on the hands of Stitch — a cinematic CG version exactly matching his appearance in Disney’s *Lilo & Stitch* movie. His hand features four expressive fingers with curved claws, proportioned anatomically like a small, stylized creature. The fingers show defined knuckles, gentle skin folds, and realistic motion creases. The claws are semi-matte and dark, slightly translucent, with no plastic gloss. His fur is short, dense, and layered in texture, with slightly uneven strand lengths and clear directional flow. Fur thins naturally near the finger joints and claws, exposing skin texture beneath. Each strand subtly catches light without overexposure, producing natural shading and depth. Lighting is soft and cinematic: a diffused three-point setup with no blown-out highlights. Shadows gently wrap around the hands, highlighting fur depth, surface texture, and object interaction. In his left hand, Stitch holds a navy blue rectangular bar of soap. The surface is embossed with shallow, evenly spaced square grid grooves, dividing it into perfect cube-shaped segments. Centered across the top is the raised logo text “DOTAI,” with each letter spanning multiple cubes. Using a small carving knife in his right hand, Stitch slices down precisely along each groove — each motion clean and continuous. The blade systematically removes the “DOTAI” logo letter by letter. Every cut breaks off a cube of soap with geometric perfection, forming a neat pile of satisfying square shavings. The logo is fully erased, leaving behind a flawless soap surface. **ASMR Audio**: Close-mic’d and isolated. Crisp slicing sounds, dry crackles as the knife enters the grooves, and satisfying snapping crunches as the cube pieces break free and fall softly. No music. No voice. (no subtitles) Camera: locked macro eye-level shot, with a slight angle revealing both fur texture and soap relief. Style: 4K cinematic, photo-realistic CG, film-grade creature rendering, ASMR-focused, satisfying and immersive

挑戰一：角色還原失控｜AI 難以準確辨識 IP 特徵

在初步生成階段，即使輸入「Stitch」等角色名稱，Veo3 仍無法準確生成「史迪仔」的外觀特徵，常出現比例失調、類似「人手」或「皮革手套」等不正確內容。

解決方案：

• 細化角色提示詞：補充毛色、爪型、動畫材質等具體描述。

• 負向提示詞排錯：明確排除「realistic」、「human」、「glove」等誤導性風格元素。

• 引入圖像提示（Image Prompt）：上傳官方角色圖片作為風格參考，有效提高還原度。

📌 分析重點：面對 IP 級角色生成，應將提示詞當作「角色資料集」進行管理，使 AI 能辨識出真正具代表性的視覺語言。

挑戰二：一鏡到底失敗｜AI 難以理解鏡頭語法與時間邏輯

雖然原始設計希望影片為無剪接的「一鏡到底」，但 Veo3 卻經常自動切換角度或中斷連貫性。

解決方案：

• 提升鏡頭語法指令優先度：將「continuous cinematic take」語句移至提示詞前方，明確宣示結構要求。

• 簡化動作敘述邏輯：刪除「旋轉手腕」等複合動作，只保留「切割」核心行為，避免觸發自動轉場。

• 策略補強：若平台允許生成更長時段片段，可先生成 15 秒素材，於後製中裁剪出流暢片段。

📌 觀察總結：AI 無法像導演一樣理解故事節奏，因此創作者必須將「時間性」與「連貫性」嵌入提示詞架構中，重建 AI 對鏡頭語法的理解機制。

提示詞工程的真正價值｜從「抽卡式生成」走向「可控創作系統」

許多創作者在使用 AI 影像工具時仍依賴反覆「抽卡」的嘗試，希望一次中靶。然而我們認為，高品質創作應建構於系統性提示詞設計與邏輯錯誤修正之上。

提示詞迭代的三大關鍵：

1. 錯誤診斷： 每次生成後進行分類分析（角色錯位／鏡頭錯誤／光影偏差）。

2. 精準修正： 局部優化關鍵詞、順序與風格參數，非全面重寫。

3. 生成效率管理： 減少資源浪費，依照成效資料進行逐步優化。

📌 創作心法：「AI 不會自動理解你要什麼，但你可以一步步教會它靠近你的創作意圖。」

想生成 AI 動畫唔再靠運氣？你可以一步一步學識 AI 圖像與影片創作

今次 Veo3 實測，只係 AIGC 創作世界的一個切入點。由提示詞設計、角色還原，到風格控制與鏡頭流暢度，每一步都可以學習與掌握，關鍵不在 AI 本身，而係你有冇一套清晰嘅操作邏輯與實戰策略。

💡 如果你都想由「玩 AI」變成「用 AI 做出作品」，DotAI 最新【AIGC 內容創作變現大師班】已全面開放報名！

🎓 課程重點包括：

1. Veo3 / Midjourney / CapCut 等主流工具實操

2. 商業級 AI 動畫、短影片、品牌內容的生成技法

3. 實戰作業＋提示詞設計模板，一站式學會圖像＋影片創作流程
   
   

   

📦 無需基礎，課後即用，幫你建立自己的 AI 圖像／影片創作技能樹
📣 點此了解詳情｜立即報名 AIGC 變現大師班 →