1. Generate dubbing via TTS → Get narration.mp3 + narration.json (timestamps)
2. Read narration.json and analyze the semantic meaning of each sentence
3. Determine which sentences each image corresponds to (match by content semantics, not average distribution!)
4. Duration of each image = end timestamp of the last corresponding sentence - start timestamp of the first corresponding sentence
5. Verification: Sum of all durations ≈ total audio duration (error < 1s)
6. ⛔ Run verify_alignment.py for validation (must pass before synthesis!)
7. Only write video_config.yaml after passing validation; otherwise, stop!

Timestamps in narration.json:
  Sentence 0 [0.1-2.6]  "Quiet Night Thoughts, Tang Dynasty, Li Bai."
  Sentence 1 [2.6-5.7]  "Before my bed a pool of light, I wonder if it's frost on the ground."
  Sentence 2 [5.7-8.6]  "I lift my eyes and look at the bright moon, I lower my head and think of my hometown."
  Sentence 3 [8.6-16.4] "This poem was written by Li Bai, a great poet of the Tang Dynasty..."
  Sentence 4 [16.4-20.9] "In just twenty characters..."
  Sentence 5 [20.9-22.6] "Before my bed a pool of light."
  ...

Image Allocation:
  01_title.png    → Sentences 0-2 (Full poem recitation) → duration = 8.6 - 0.1 = 8.5s
  02_poet.png     → Sentences 3-4 (Poet introduction) → duration = 20.9 - 8.6 = 12.3s
  03_moonlight.png → Sentences 5-8 (Interpretation of the first line) → duration = 38.6 - 20.9 = 17.7s
  ...

Validation: 8.5 + 12.3 + 17.7 + ... = 114.4s ≈ Audio duration 114.5s ✅

undefined

undefined

undefined

**Rules**:
- Prompts must be Chinese-only
- Any text in generated images must be Chinese
- No English is allowed

undefined

**Abundant Visual Elements**:

**Information Density Principles**:
- Each image must have a clear text title and key points
- Text content must correspond to the dubbing content
- Visualization is more important: icons, charts, illustrations, decorations
- Pure text images or pure decorative images are forbidden

undefined

**6 Essential Elements for Prompts** (None can be missing):

| Element | Description | Example |
|---------|-------------|---------|
| Subject | Who/what, specific | "Blue rounded metallic AI robot" instead of "robot" |
| Action | What it's doing, posture | "Hands on keyboard typing, slightly tilted head" |
| Environment | Where it is, background | "Modern minimalist office, city night view outside floor-to-ceiling windows" |
| Details | Surrounding items/elements | "Coffee cup, succulent plant, sticky notes on desk" |
| Style | Art style/lighting effects | "Cyberpunk style, neon lighting, volumetric lighting effect" |
| Color | Main color tone | "Blue-purple gradient main tone, orange accents" |

undefined

undefined

┌─────────────────────────────────────────────────────────────┐
│ Layer 1: Story → Split into scenes → Parallel generation of scene main images (text-to-image) │
│                                                              │
│   Havoc in Heaven →  Scene 1: Sun Wukong humiliated as Horse Keeper │
│              Scene 2: Returns to Flower Fruit Mountain on Cloud Somersault │
│              Scene 3: Jade Emperor sends troops │
│              ...                                              │
│              → Parallel call text_to_image.py to generate main image for each scene │
└─────────────────────────────────────────────────────────────┘
                              ↓
┌─────────────────────────────────────────────────────────────┐
│ Layer 2: Each scene main image → Generate close-up shots via image-to-image (maintain character consistency) │
│                                                              │
│   Scene 1 main image → Close-up 1: Sun Wukong doubts the official seal │
│              Close-up 2: Sun Wukong kicks over the horse trough │
│   Scene 2 main image → Close-up 1: Soars on Cloud Somersault │
│              Close-up 2: Proclaims himself Great Sage Equal to Heaven at Flower Fruit Mountain │
│              → Parallel call image_to_image.py with main image as reference │
└─────────────────────────────────────────────────────────────┘
                              ↓
┌─────────────────────────────────────────────────────────────┐
│ Layer 3: Generate dubbing + subtitles + validation + video synthesis │
│                                                              │
│   1. tts_generator.py generates dubbing + timestamps │
│   2. [Rule] Precisely calculate duration of each image based on timestamps │
│   3. Generate SRT subtitles │
│   4. Generate video_config.yaml │
│   5. ⛔ Run verify_alignment.py for validation (must pass!) │
│   6. video_maker.py synthesis: │
│      → Image synthesis (with transitions) │
│      → Audio merging │
│      → Burn subtitles (ASS format, fixed at bottom center) │
│      → Auto-append outro (QR code + "Follow for more content") │
│      → Add BGM │
└─────────────────────────────────────────────────────────────┘

**Rule: All videos must auto-append outro!**

assets/generated/{project_name}/
├── scene1/
│   ├── main.png         # Scene 1 main image (text-to-image)
│   ├── shot_01.png      # Close-up 1 (image-to-image)
│   └── shot_02.png      # Close-up 2 (image-to-image)
├── scene2/
│   ├── main.png
│   ├── shot_01.png
│   └── shot_02.png
├── ...
├── narration.mp3        # Dubbing
├── narration.json       # Timestamps
├── subtitles.srt        # Subtitles
├── video_config.yaml    # Video configuration
└── {project_name}.mp4   # Final video

undefined

---

undefined

**Note**: `ratio` must be enclosed in quotes, e.g., `"16:9"`, otherwise YAML will parse it as a time format.

---

import json
with open("narration.json", "r") as f:
    timestamps = json.load(f)
audio_duration = timestamps[-1]["end"]
print(f"Total audio duration: {audio_duration:.1f}s")

undefined

undefined

for file, start, end in scenes:
    duration = end - start
    print(f"{file}: {duration:.1f}s")

total_duration = sum(end - start for _, start, end in scenes)
assert abs(total_duration - audio_duration) < 1.0, \
    f"Duration mismatch! Total image duration {total_duration}s vs Audio duration {audio_duration}s"

undefined

**Note**: video_maker.py also has built-in hard validation — duration must be a precise positive float. Missing or non-numeric values will trigger immediate exit(1)! duration: auto has been completely removed!

| Scene Image | Corresponding Content | Start | End | Duration |
|-------------|-----------------------|-------|-----|----------|
| cover.png | Opening Introduction | 0s | 12.5s | 12.5s |
| scene01.png | AI Agent Era | 12.5s | 26s | 13.5s |
| ... | ... | ... | ... | ... |
| **Total** | | | | **{total}s** |

Total Audio Duration: {audio_duration}s
Difference: {diff}s ✅/❌

python video_maker.py config.yaml [options]

python verify_alignment.py video_config.yaml

python tts_generator.py --text "Text" --output audio.mp3 [options]

undefined

---

video-creator/
├── SKILL.md
├── scripts/
│   ├── video_maker.py        # Main script: Images + Audio → Video (built-in duration hard check)
│   ├── verify_alignment.py   # Mandatory pre-synthesis validation (Duration + Semantic cross-check)
│   ├── tts_generator.py      # TTS Voice Generation
│   └── scene_splitter.py     # Scene Splitter (Optional)
├── assets/
│   ├── outro.mp4             # General outro (16:9)
│   ├── outro_9x16.mp4        # Portrait outro
│   ├── outro_3x4.mp4         # 3:4 outro
│   └── bgm_*.mp3             # 22 BGM tracks (see BGM Resource Table)
└── references/
    └── edge_tts_voices.md

undefined

undefined