Script Generation
StoryFab uses AI to generate narration scripts for your video clips. In Commentary Mode (v3.0), this becomes a full narrative generation pipeline with style presets, coherence tracking, and quality control.
Two Modes of Script Generation
Clip Mode — Narration for Highlights（快速解说）
For Clip Mode, StoryFab generates short narration scripts for individual highlight segments.
Workflow:
- AI analyzes the selected clip's transcript
- Rewrites it as a concise, engaging narration
- Edge TTS converts the script to natural speech
- Voice-over is merged into the exported clip
Commentary Mode — Full Narrative Generation（🆕 v3.0）
For Commentary Mode, StoryFab generates a complete narrative structure with multiple acts.
Workflow:
- Semantic Understanding — LLM understands plot, characters, emotional tone
- Structure Planning — AI Director plans intro / acts / outro
- Script Generation — Generates commentary for each segment with coherence
- Quality Check — Validates reading time, relevance, repetition
- TTS Synthesis — Edge TTS generates voice-over
- Timeline Alignment — Audio is aligned to video segments
AI Providers
StoryFab supports multiple AI providers through a unified provider interface:
| Provider | Recommended Model | API Required |
|---|---|---|
| DeepSeek | V4-Pro（性价比最高） | Yes |
| OpenAI | GPT-5.5 | Yes |
| Anthropic | Claude Opus 4.7 | Yes |
| Gemini 3.1 Pro | Yes | |
| 阿里云 | Qwen3.6-Max | Yes |
| 月之暗面 | Kimi K2.6 | Yes |
Set your API key and default provider in Settings → AI → Provider.
Commentary Style Presets（🆕）
Commentary Mode supports multiple style presets to match your content:
| Style | Description | Tone | Best For |
|---|---|---|---|
| 幽默版 | 诙谐有趣，添加网络梗 | fast | 喜剧/搞笑视频 |
| 接地气版 | 口语化，像和朋友聊天 | normal | 情感/生活类视频 |
| 震惊版 | 夸张震惊，制造悬念 | fast | 悬疑/复仇类短剧 |
| 感动版 | 温情脉脉，情感共鸣 | slow | 爱情/亲情类视频 |
| 专业版 | 客观冷静，纪录片风格 | normal | 纪录片/科教类 |
Style Customization
You can customize:
- Tone — 语气（幽默/严肃/接地气/震惊/感动）
- Pacing — 语速（fast/normal/slow）
- Opening Pattern — 开场句式（如"没想到..."、"当...的时候..."）
- Emphasis Keywords — 强调词（重点突出）
How Script Generation Works（🆕）
1. Context Building
For each segment, StoryFab builds a context that includes:
interface SegmentContext {
video: { title, duration, genre };
segment: { start_ms, end_ms, plot_summary, characters, emotional_tone };
narrative: { previous_script, characters_so_far, story_arc };
constraints: { target_duration_ms, style };
}2. Prompt Engineering
StoryFab uses structured prompts for different parts of the narrative:
Intro Template — Opening hook (first 3 seconds must grab attention) Act Template — Main story segments with coherence Outro Template — Cliffhanger and call-to-action
3. LLM Generation
The LLM generates JSON-structured output:
{
"script": "解说词正文...",
"reading_time_ms": 15000,
"emphasis_positions": [5, 23, 45],
"keywords_to_highlight": ["关键", "精彩"]
}4. Quality Check
Before finalizing, StoryFab validates:
| Check | Criteria | Action if Failed |
|---|---|---|
| Duration | ±15% of target | Warning |
| Repetition | No repeated patterns | Warning |
| Relevance | Script matches plot | Error → Fallback |
5. Coherence Tracking
Across multiple segments, StoryFab maintains:
- Previous Script — Passed to LLM for context
- Character Track — Avoids re-introducing characters
- Story Arc — Maintains narrative flow
Script Customization
After generation, you can:
- Edit manually — Tweak the script text directly
- Regenerate — Ask AI for a new version
- Change style — Switch between style presets
- Adjust length — Shorten or expand the narration
- Add emphasis — Mark keywords to be highlighted
Voice Settings（🆕）
| Setting | Options | Default |
|---|---|---|
| Voice | Any Edge TTS voice | zh-CN-XiaoxiaoNeural |
| Speed | 0.5x – 2.0x | 1.0x |
| Pitch | -50% – +50% | 0 |
| Volume | 0% – 100% | 100% |
Voice Selection by Style
| Style | Recommended Voice |
|---|---|
| 幽默版 | zh-CN-XiaoxiaoNeural（活泼女声） |
| 接地气版 | zh-CN-XiaoyouNeural（年轻女声） |
| 震惊版 | zh-CN-YunyangNeural（新闻男声） |
| 感动版 | zh-CN-XiaobaiNeural（温柔女声） |
| 专业版 | zh-CN-YunxiNeural（稳重男声） |
Disabling Script Generation
Script generation is optional. You can export clips with:
- Original audio only — No narration
- Subtitles only — Burn-in text without voice
- Full Commentary — Narration + original + subtitles
In Commentary Mode, if TTS fails, StoryFab will:
- Retry up to 2 times
- If still failing, skip audio and export video-only with subtitles