feat(skills): add happyhorse-prompt-studio skill

Interactive prompt studio for HappyHorse 1.0 video generation. Guides users through 4-phase scenario discovery (manga drama, character PV, panel motion, virtual idol MV) and assembles production-ready prompts in JP/CN/EN with proper lip-sync, R2V character consistency, and cost estimates. Includes SKILL.md, README.md, and README.zh.md.
2026-09-19 04:50:42 +08:00 · 2026-06-04 15:33:13 +08:00
parent 59d6e5916c
commit 10e44a45f4
3 changed files with 607 additions and 0 deletions
@@ -0,0 +1,79 @@
+# HappyHorse Prompt Studio
+
+> [中文版 / Chinese →](README.zh.md)
+
+Interactive prompt studio for **HappyHorse 1.0** video generation — a guided conversation that turns "I want to make a video" into a production-ready prompt.
+
+Covers 4 scenario flavors + free-form mode:
+
+| Flavor | Description | Duration |
+|--------|-------------|----------|
+| **A · Voiced Manga Drama** | Multi-character dialogue with voice & lip-sync | 15–30s |
+| **B · Character Voice PV** | Single character self-introduction with voice | 8–10s |
+| **C · Manga Panel Motion** | Animate a static manga panel (hair, eyes, wind) | 5–10s |
+| **D · Virtual Idol MV** | Stage performance with singing & choreography | 30s |
+
+Supports JP / CN / EN prompts with native phrasing and proper lip-sync brackets.
+
+## How it works
+
+```
+Phase 1 · Inspiration Menu    →  "Here's what HappyHorse can do"
+Phase 2 · Discovery           →  "What do YOU want to make?"
+Phase 3 · Prompt Assembly     →  "Let me build it for you"
+Phase 4 · Quality Check       →  "Here's your prompt — want to tweak?"
+```
+
+The skill follows the **HappyHorse Formula**:
+
+```
+Scene + Subject + Motion + Audio + Quality
+```
+
+And outputs a prompt ready to paste into `bl video generate` or the HappyHorse web UI, with:
+- Character consistency constraints (`キャラの顔・髪・衣装が変わらない`)
+- R2V reference syntax (`@「Image n」`) for multi-image inputs
+- Native-language voice type descriptors
+- Cost estimates (720P: ¥0.9/sec, 1080P: ¥1.6/sec)
+
+## Quick start
+
+Install the skill, then tell your agent:
+
+```
+I want to make a HappyHorse video of my manga character introducing herself.
+She's a cheerful schoolgirl with pink hair, on a rooftop at lunch.
+```
+
+The skill will guide you through discovery and assemble a prompt like:
+
+```
+昼休みの学校の屋上、青空と白い雲、風が心地よい。
+桜色のロングヘアの少女がフェンスに寄りかかり、こちらを見て笑っている。
+
+少女が手を振り、カメラがゆっくり寄る。
+[少女、ネイティブな日本語、明るく元気な若い女性声、嬉しそう] 言う：
+「ねえ！来てくれたんだ！」
+
+背景に風の音、遠くで校庭のざわめき、明るいピアノの BGM。
+映画級質感、キャラの顔・髪・制服が変わらない。
+```
+
+## Prerequisites
+
+- [`bailian-cli`](https://www.npmjs.com/package/bailian-cli) (`bl`) installed — [install guide](https://bailian.aliyun.com/cli/install.md)
+- A DashScope API key for video generation
+
+## Run the prompt directly
+
+```bash
+# Image-to-video (one character reference)
+bl video generate --image ./char.png --prompt "[PROMPT]" --download output.mp4
+
+# Reference-to-video (multiple references for consistency)
+bl video generate --image ./char1.png --image ./char2.png --prompt "[PROMPT]" --download output.mp4
+```
+
+## License
+
+Apache-2.0
@@ -0,0 +1,79 @@
+# HappyHorse Prompt Studio（HappyHorse 提示词工作室）
+
+> [English →](README.md)
+
+**HappyHorse 1.0** 视频生成的交互式提示词工坊 —— 通过引导式对话，将"我想做一个视频"转化为可直接使用的生产级 Prompt。
+
+覆盖 4 种场景风味 + 自由模式：
+
+| 风味 | 描述 | 时长 |
+|------|------|------|
+| **A · 漫画配音剧** | 多角色对话，配音 + 口型同步 | 15–30s |
+| **B · 角色语音 PV** | 单角色自我介绍，配音 + 口型 | 8–10s |
+| **C · 漫画分格动态化** | 让静态漫画动起来（头发、眨眼、风） | 5–10s |
+| **D · 虚拟偶像 MV** | 舞台表演，唱歌 + 编舞 + 多机位 | 30s |
+
+支持 JP / CN / EN 三种语言输出，原生语感，正确口型标记。
+
+## 工作流程
+
+```
+Phase 1 · 灵感菜单    →  "HappyHorse 能做什么"
+Phase 2 · 需求发现    →  "你想做什么？"
+Phase 3 · Prompt 组装 →  "我来帮你拼好"
+Phase 4 · 质量检查    →  "Prompt 好了——要调整吗？"
+```
+
+遵循 **HappyHorse 公式**：
+
+```
+场景 + 主体 + 動き + 音声 + 品質修飾
+```
+
+输出内容包括：
+- 角色一致性约束（`キャラの顔・髪・衣装が変わらない`）
+- R2V 多图引用语法（`@「Image n」`）
+- 原生语言声色描述
+- 费用估算（720P: ¥0.9/秒, 1080P: ¥1.6/秒）
+
+## 快速开始
+
+安装 Skill 后，告诉 Agent：
+
+```
+我想用 HappyHorse 做一个视频，我的漫画角色是个粉色头发的活泼女高中生，
+在午休的屋顶上自我介绍。
+```
+
+Skill 会引导你完成需求发现，并组装出类似这样的 Prompt：
+
+```
+昼休みの学校の屋上、青空と白い雲、風が心地よい。
+桜色のロングヘアの少女がフェンスに寄りかかり、こちらを見て笑っている。
+
+少女が手を振り、カメラがゆっくり寄る。
+[少女、ネイティブな日本語、明るく元気な若い女性声、嬉しそう] 言う：
+「ねえ！来てくれたんだ！」
+
+背景に風の音、遠くで校庭のざわめき、明るいピアノの BGM。
+映画級質感、キャラの顔・髪・制服が変わらない。
+```
+
+## 前置要求
+
+- 已安装 [`bailian-cli`](https://www.npmjs.com/package/bailian-cli)（`bl`）—— [安装指南](https://bailian.aliyun.com/cli/install.md)
+- DashScope API Key
+
+## 直接运行 Prompt
+
+```bash
+# 图生视频（单张角色参考图）
+bl video generate --image ./char.png --prompt "[PROMPT]" --download output.mp4
+
+# 参考图生视频（多张参考图，角色一致性更强）
+bl video generate --image ./char1.png --image ./char2.png --prompt "[PROMPT]" --download output.mp4
+```
+
+## License
+
+Apache-2.0
@@ -0,0 +1,449 @@
+---
+name: happyhorse-prompt-studio
+description: Interactive prompt studio for HappyHorse 1.0 video generation. Guides users through scenario discovery with vivid examples, then assembles production-ready prompts in JP/CN/EN. Use when someone wants to create AI video content with HappyHorse but doesn't know where to start, or when they have a specific scenario and need a polished prompt. Covers manga drama, character PV, manga motion, virtual idol MV, and free-form scenarios.
+---
+
+# HappyHorse Prompt Studio
+
+A 4-phase guided skill that turns "I want to make a video" into a production-ready HappyHorse prompt — starting from inspiration, not from a blank page.
+
+## Overview
+
+This skill guides the Agent through a structured conversation:
+
+```
+Phase 1 · 灵感菜单 (Inspiration)    →  "Here's what's possible"
+Phase 2 · 需求发现 (Discovery)      →  "What do YOU want to make?"
+Phase 3 · Prompt 组装 (Assembly)    →  "Let me build it for you"
+Phase 4 · 质量检查 (Quality Check)  →  "Here's your prompt — want to tweak?"
+```
+
+## Phase 1 · Inspiration Menu (灵感菜单)
+
+**Start every conversation here.** Before asking any questions, show the user what HappyHorse can do. Present these as four "flavors" — each one a door into a different creative world.
+
+Use the language the user is using (JP/CN/EN). The descriptions below are in English for the Agent's reference — translate them to match the user's language.
+
+---
+
+### Flavor A · "让你的角色开口说话"
+**Voiced Manga Drama (漫画配音剧 / ボイスコミック)**
+
+> You have a manga, a webtoon, or an original story. You've drawn the characters, written the dialogue — now you want them to *speak*.
+>
+> Upload 2-3 character reference images + a short script. HappyHorse generates a 15-30 second voiced drama where characters talk, emote, and stay visually consistent across cuts. Lip-sync included.
+>
+> **Vibe:** Movie dub meets manga animation. Your characters, their voices.
+
+### Flavor B · "一张立绘，开口自我介绍"
+**Character Voice PV (角色语音 PV / キャラボイス PV)**
+
+> You have a game character, a VTuber, or an original OC. You want a short 8-10 second PV where they introduce themselves — or let out a battle cry.
+>
+> Upload 1-3 character art images + a line or two. HappyHorse generates a voiced, lip-synced character PV.
+>
+> **Vibe:** Character reveal trailer. One illustration, one voice, one PV.
+
+### Flavor C · "让一格漫画活过来"
+**Manga Panel Motion (漫画分格动态化 / コマ動画化)**
+
+> You have manga panels, comic pages, or illustrated scenes. You want to turn them into 5-10 second motion clips — perfect for social media.
+>
+> Upload one panel as the first frame. HappyHorse animates it while preserving your art style.
+>
+> **Vibe:** Your drawing, but it breathes. Hair moves, eyes blink, wind blows.
+
+### Flavor D · "你的虚拟偶像，30 秒成 MV"
+**Virtual Idol MV (虚拟偶像 MV / バーチャルアイドル MV)**
+
+> You have a virtual idol, a VTuber group, or an original idol project. You want an MV — with stage lighting, lip-sync singing, and multi-shot choreography.
+>
+> Upload 3-5 multi-angle character images + a licensed song segment. HappyHorse generates a 30-second MV clip.
+>
+> **Vibe:** Your idol, center stage. No Live2D. No MMD. Just one prompt.
+>
+> ⚠️ **Note:** This scenario requires the strongest compliance guardrails. We'll check together.
+
+---
+
+### How to present the menu
+
+Present the four flavors conversationally, not as a dry list. Something like:
+
+> "HappyHorse can bring your characters to life in a few different ways. Think of it as four flavors:
+>
+> A · **Voiced Drama** — your manga characters talk to each other, with voice and lip-sync
+> B · **Character PV** — your game character or OC introduces itself out loud
+> C · **Panel Motion** — a single manga panel starts moving, hair blowing, eyes blinking
+> D · **Idol MV** — your virtual idol performs a 30-second MV on stage
+>
+> Which one sounds closest to what you're imagining? Or tell me about your project and I'll suggest."
+
+If the user already knows what they want, skip to Phase 2.
+
+---
+
+## Phase 2 · Discovery (需求发现)
+
+Once a flavor is chosen (or the user describes their own scenario), ask these questions. **Ask them conversationally, not as a form.** Group related questions together.
+
+### 2.1 Character & World (角色与世界)
+
+- What's your character's name and role? (protagonist / antagonist / side character)
+- What do they look like? (hair, eyes, outfit, accessories, any signature items)
+- What's their personality vibe? (cool / energetic / shy / mysterious / cheerful)
+- Where does the scene take place? (school rooftop / fantasy castle / neon city / café / etc.)
+
+### 2.2 Scene Intent (场景意图)
+
+- What's happening in this scene? (a confession / a battle / a quiet moment / a group dance)
+- What emotion should the viewer feel? (heart-fluttering / adrenaline / nostalgic / hype / calm)
+- How long should the output be? (5s / 10s / 15s / 30s)
+
+### 2.3 Voice & Sound (声音与音频)
+
+- Does your character speak? If yes:
+  - What language? (Japanese / Chinese / English)
+  - Voice type? (young woman / young man / child / mature / elderly)
+  - Voice color? (bright / low / soft / powerful / cool)
+  - What do they say? (provide the exact line, or ask me to suggest)
+- Background audio? (silence / ambient sounds / BGM style)
+
+### 2.4 Visual Style (视觉风格)
+
+- Art style reference? (anime / photorealistic / Pixar / watercolor / pixel art / etc.)
+- Color palette? (warm / cool / neon / pastel / high-contrast)
+- Camera preference? (close-up / medium / wide / rotating / slow push / static)
+
+### 2.5 Compliance Quick-Check (合规快检)
+
+Before proceeding, verify:
+
+- ☐ Is the character **your own original creation** or properly licensed?
+- ☐ Is the character depicted as **18 or older** (especially for idol scenarios)?
+- ☐ Is the outfit **SFW** (no suggestive or revealing clothing)?
+- ☐ Is the scene **SFW** (no sensitive locations like bedrooms/pools)?
+- ☐ If there's music, is it **licensed or original** (not a commercial song)?
+
+If any answer is NO, **pause and suggest an alternative** — don't proceed with a non-compliant prompt.
+
+---
+
+## Phase 3 · Prompt Assembly (Prompt 组装)
+
+Now build the prompt using the **HappyHorse Formula**:
+
+```
+场景 (Scene) + 主体 (Subject) + 動き (Motion) + 音声 (Audio) + 品質修飾 (Quality)
+```
+
+### 3.1 The Formula (公式)
+
+| Component | What it does | Example (JP) |
+|-----------|-------------|--------------|
+| 场景 Scene | Where + when + atmosphere | 夕暮れの古城の庭、金色の逆光、落ち葉が舞い散る |
+| 主体 Subject | Who + appearance + state | 甲冑とマントをまとった若い剣士が、剣の柄に手を添えて立っている |
+| 動き Motion | What happens + camera movement | 剣士がゆっくり顔を上げ、カメラが全身から胸像へとゆっくり寄る |
+| 音声 Audio | Voice lines + voice type + ambient sound + BGM | [剣士、ネイティブな日本語、低めの落ち着いた青年声] 言う：「俺の刃は、守るためにある。」背景に風、鐘の音 |
+| 品質 Quality | Style + consistency constraints | 映画級質感、リアルな光影、キャラの表情と装備が維持 |
+
+### 3.2 R2V Character Consistency Syntax
+
+When the user provides multiple reference images, use this syntax:
+
+```
+参考「Image 1」=[正面立绘]、「Image 2」=[侧面]、「Image 3」=[表情差分]
+```
+
+Or when referencing a specific character in a multi-character scene:
+
+```
+@「Image 1」のキャラが [action]、@「Image 2」のキャラが [reaction]
+```
+
+Key rules:
+- Always use `@「Image n」` to lock character identity across shots
+- Describe what each reference image shows (正面 / 側面 / 表情差分)
+- End with: `キャラの顔・髪・衣装が変わらない` (character's face/hair/outfit stays unchanged)
+
+### 3.3 Video-Edit Style Unification
+
+When the user wants to unify style across multiple shots:
+
+```
+参考「Image 1」の画風・筆致・色調に、動画全体のスタイルを変換。
+人物の動作、表情、運鏡、セリフのタイミングは 100% 保持。
+```
+
+Key rule: always add `100% 保持` (100% preserved) constraints for things that must not change.
+
+### 3.4 Language Rules
+
+| Language | When to use | Key markers |
+|----------|-------------|-------------|
+| Japanese (JP) | User speaks JP, or targeting JP market | `ネイティブな日本語`, use `「」` for dialogue, avoid translation-style phrasing |
+| Chinese (CN) | User speaks CN | `中文母语`, use `「」` or `""` for dialogue |
+| English (EN) | User speaks EN, or targeting global market | `native English`, use `""` for dialogue |
+
+**Japanese-specific tips:**
+- Use `ネイティブな日本語` to ensure natural Japanese (not translation-style)
+- Specify voice color with JP adjectives: `明るく元気な少女声`, `低めの落ち着いた青年声`, `柔らかい囁くような声`
+- Keep dialogue in `「」` brackets
+- Avoid mixing languages in dialogue unless intentionally bilingual
+
+### 3.5 Prompt Templates by Flavor
+
+#### Flavor A · Voiced Manga Drama
+
+```
+[场景描述：时间/地点/光线/氛围]
+
+[角色A描述] と [角色B描述] が [位置关系]。
+
+[角色A动作描述]。
+[角色A, 语言, 声色描述, 情绪] 言う：「[台词]」
+
+[角色B反应动作]。
+[角色B, 语言, 声色描述, 情绪] 答える：「[台词]」
+
+[镜头运动描述]。
+[环境音/BGM描述]。
+
+映画級質感、リアルな光影、キャラの顔・髪・衣装が変わらない。
+```
+
+#### Flavor B · Character Voice PV
+
+```
+[场景描述：地点/光线/氛围]
+
+[角色外观描述] が [姿态]。
+
+[角色动作 + 镜头运动]。
+[角色, 语言, 声色描述, 情绪] 言う：「[台词]」
+
+[环境音/BGM描述]。
+
+映画級質感、キャラの顔・髪・衣装・装備が変わらない。
+```
+
+#### Flavor C · Manga Panel Motion
+
+```
+[图片中已有内容，无需重复描述]
+
+[角色简短动作：眨眼/头发飘动/微笑/转头]。
+[环境动态：风吹/光斑/落叶/雨滴]。
+[镜头运动：缓慢推近/横移/拉远/固定]。
+
+[环境音/BGM描述]。
+
+原画風、キャラの顔・髪・衣装・構図・色調が変わらない。
+```
+
+#### Flavor D · Virtual Idol MV
+
+```
+参考「Image 1」=[正面]、「Image 2」=[侧面]、「Image 3」=[表情差分]
+中的虚拟偶像角色形象。
+
+[舞台场景描述]
+[灯光描述]
+
+分镜 1（N 秒）：[景别]，角色 [姿态/动作]。
+[角色, 语言, 声色, 情绪] 歌う：「[歌词]」
+
+分镜 2（N 秒）：[景别]，角色 [舞蹈动作]。
+[歌词]
+
+分镜 3（N 秒）：[景别]，角色 [表情]。
+[歌词]
+
+[J-POP MV 美学 / 电影级光影]，キャラの顔・髪・衣装が変わらない。
+口型与歌词重音精准匹配。
+```
+
+### 3.6 Assembling the Output
+
+Present the final prompt in a code block so the user can copy it directly. Include:
+
+1. **The prompt itself** (in the user's language)
+2. **A brief breakdown** of what each part does
+3. **Suggested model variant** (t2v / i2v / r2v / video-edit)
+4. **Estimated cost** (720P: ¥0.9/sec, 1080P: ¥1.6/sec)
+
+Example output format:
+
+```
+Here's your prompt — ready to paste into HappyHorse:
+
+```
+[PROMPT HERE]
+```
+
+**Breakdown:**
+- 场景: [what this part does]
+- 主体: [what this part does]
+- 動き: [what this part does]
+- 音声: [what this part does]
+
+**Model:** happyhorse-1.0-r2v (you have 2+ reference images)
+**Duration:** ~10 seconds
+**Cost estimate:** 720P ≈ ¥9, 1080P ≈ ¥16
+
+Want me to adjust anything? (voice tone, camera angle, add another character, change the line?)
+```
+
+---
+
+## Phase 4 · Quality Check (质量检查)
+
+Before finalizing, run through this checklist silently. If anything fails, fix before presenting.
+
+### 4.1 Prompt Quality
+
+- ☐ Does the prompt follow the Scene + Subject + Motion + Audio + Quality structure?
+- ☐ Is the camera movement explicitly stated? (not left to chance)
+- ☐ Is the voice type described with specific adjectives? (not vague)
+- ☐ Is the dialogue in the correct brackets for the language? (「」 for JP, "" for EN)
+- ☐ Is the "stays unchanged" constraint included at the end?
+- ☐ Is the prompt length between 150-300 characters? (too short = under-specified; too long = hard to control)
+
+### 4.2 Compliance Check
+
+- ☐ No existing anime/manga/game IP referenced?
+- ☐ No real person likeness?
+- ☐ Character depicted as adult?
+- ☐ Outfit is SFW?
+- ☐ Scene location is SFW?
+- ☐ If music is involved, it's licensed/original?
+
+### 4.3 Optimization Tips
+
+If the prompt looks good, offer these pro-tips:
+
+- **"Try 3 variants"** — HappyHorse results vary; generating 3-5 and picking the best is standard practice
+- **"Start 720P, finish 1080P"** — do test runs at 720P (cheaper), then re-generate the winner at 1080P
+- **"Shorter lines = better lip-sync"** — if the voice line is over 15 characters, consider splitting into two shots
+- **"Specific beats vague"** — "camera slowly pushes from full-body to chest close-up" beats "camera moves"
+
+---
+
+## Free-Form Mode (自由模式)
+
+If the user's scenario doesn't fit Flavors A-D, use the formula directly:
+
+1. Ask: "What's your scene? Describe it like you're telling a friend about a movie you just watched."
+2. Extract: scene, subject, motion, audio, quality from their description
+3. Assemble using the formula
+4. Apply the quality check
+
+This mode is especially useful for:
+- Product advertisements
+- Educational explainers
+- Abstract / artistic videos
+- Non-character-driven content
+
+---
+
+## Common Pitfalls (常见问题)
+
+| Problem | Fix |
+|---------|-----|
+| Character face drifts across shots | Add more reference images + emphasize "顔・髪・衣装が変わらない" |
+| Voice sounds like machine translation | Add "ネイティブな日本語" marker; avoid translation-style phrasing |
+| Lip-sync is off | Shorten dialogue to ≤15 characters; split into two shots |
+| Motion is stiff | Replace generic "動いている" with specific actions: "髪が風に右へ流れる" |
+| Camera wanders | Always specify camera: "固定", "ゆっくり寄る", "横にパン" |
+| Style drifts from original art | Add "原画風・色調が変わらない" at the end |
+| Multiple characters swap positions | Use `@「Image n」` syntax to lock each character's position |
+| Q-version character becomes realistic | Add "デフォルメ / ちびキャラ / Q 版" style keyword |
+
+---
+
+## CLI Quick-Start (for users who want to run it immediately)
+
+If the user has `bailian-cli` installed, they can run the prompt directly:
+
+```bash
+# Text-to-video (no input image)
+bl video generate --prompt "[PROMPT]" --download output.mp4
+
+# Image-to-video (one input image)
+bl video generate --image ./char.png --prompt "[PROMPT]" --download output.mp4
+
+# Reference-to-video (multiple input images for character consistency)
+bl video generate --image ./char1.png --image ./char2.png --prompt "[PROMPT]" --download output.mp4
+
+# Video editing (style transfer on existing video)
+bl video edit --video ./input.mp4 --image ./style-ref.png --prompt "[PROMPT]" --download output.mp4
+```
+
+---
+
+## Example Interactions
+
+### Example 1 · First-time user (flavor A)
+
+```
+User: "I have a manga character I want to animate. She's a schoolgirl with pink hair, and I want her to say something cute."
+
+Agent: "That sounds like Flavor B — a Character Voice PV! Let me ask a few things:
+
+1. What's her name and personality? (shy? cheerful? cool?)
+2. What's she saying? Got a line in mind, or want me to suggest one?
+3. Where is she? (school rooftop? cherry blossom park? classroom?)
+4. Voice vibe — bright and cute, or soft and shy?
+
+And quick compliance check: is she your own original character?"
+
+User: "Her name is Sakura, she's super cheerful, she's on a school rooftop at lunch. She says 'Hey! You made it!' in Japanese. Bright voice!"
+
+Agent: [assembles prompt using Flavor B template]
+
+"Here's your prompt:
+
+```
+昼休みの学校の屋上、青空と白い雲、風が心地よい。
+桜色のロングヘアの少女がフェンスに寄りかかり、こちらを見て笑っている。
+
+少女が手を振り、カメラがゆっくり寄る。
+[少女、ネイティブな日本語、明るく元気な若い女性声、嬉しそう] 言う：
+「ねえ！来てくれたんだ！」
+
+背景に風の音、遠くで校庭のざわめき、明るいピアノの BGM。
+映画級質感、キャラの顔・髪・制服が変わらない。
+```
+
+Model: happyhorse-1.0-i2v (one image input)
+Duration: ~8 seconds
+Cost: 720P ≈ ¥7.2, 1080P ≈ ¥12.8
+
+Want to tweak anything?"
+```
+
+### Example 2 · Experienced user (free-form)
+
+```
+User: "I need a 10-second shot of a samurai drawing his sword at sunset, with a dramatic voice line."
+
+Agent: "Got it — free-form, single character. Quick questions:
+
+1. What does he say? (Japanese line?)
+2. Voice type — deep and gravelly?
+3. Camera — slow push to close-up? Or wide shot?
+4. Art style — realistic? Anime?
+
+Compliance: original character?"
+```
+
+---
+
+## Final Notes for the Agent
+
+- **Always start with Phase 1** unless the user is clearly experienced and already knows what they want
+- **Be creative with descriptions** — don't just ask "what's the scene?", say "paint the picture for me — where are we, what time of day, what's the vibe?"
+- **Suggest, don't just ask** — if the user seems unsure, offer defaults: "How about a sunset rooftop scene with a gentle breeze?"
+- **Show the prompt in a code block** so it's easy to copy
+- **Always offer to iterate** — "Want me to adjust the voice tone? Change the camera angle? Add a second character?"
+- **Keep compliance friendly, not scary** — "Just to make sure everything's smooth, is this your original character?" not "COMPLIANCE CHECK: CONFIRM IP STATUS"