Reelsmith
Overview
Use this skill to turn a topic into a Facebook Reels-ready content package.
The output should feel like a practical creator workflow, not just a block of text.
Treat the core unit as:
- - hook
- short script
- scene breakdown
- on-screen text
- caption
- visual plan
- approval before posting or production
Core Workflow
- 1. Identify the topic and goal.
- Accept sports updates, news summaries, explainers, product demos, quick opinions, event recaps, or promotional ideas.
- Figure out whether the reel should:
- inform
- react
- explain
- promote
- recap
- 2. Identify the target style.
- Common modes:
- informative
- energetic
- professional
- punchy
- promotional
- If the user does not specify, default to
informative + engaging.
- 3. Decide the reel length.
- Default to
30-45 seconds.
- Shorter is usually better unless the topic clearly needs more room.
- 4. Write a strong hook.
- The first line should make the viewer care immediately.
- Prefer clarity over empty hype.
- Hooks should sound natural when spoken aloud.
- 5. Build the short-form script.
- Keep sentences short and speakable.
- Avoid overloading the reel with too many facts.
- Focus on the strongest 2-4 points only.
- 6. Break the script into scenes.
- Each scene should have:
- what is being said
- what is shown
- optional on-screen text
- Prefer 4-7 scenes for a normal reel.
- 7. Add packaging elements.
- Include:
- caption
- light hashtag suggestions
- cover/thumbnail concept
- optional visual prompt ideas
- 8. Keep it approval-first.
- Do not assume the user wants to publish immediately.
- Present the reel package clearly so it can be reviewed, revised, or later turned into a produced video.
Output Format
Unless the user asks for something else, use this structure:
Reel concept: [one-line concept]
Tone: [tone]
Length: [estimated seconds]
Hook
[hook]
Script
[full short-form script]
Scene breakdown
- 1. [scene 1 visual + spoken line + on-screen text]
- [scene 2 visual + spoken line + on-screen text]
- [scene 3 visual + spoken line + on-screen text]
Caption
[caption]
Hashtags
[list]
Cover concept
[headline + visual direction]
Approval status
Awaiting user approval before production or publishing
Writing Rules
- - Optimize for spoken clarity.
- Write in a way that sounds natural aloud.
- Keep hooks sharp.
- Avoid bloated intros.
- Avoid stuffing too many facts into a short reel.
- Prefer memorable structure over exhaustive detail.
Reel-Specific Guidance
- - Strong reel openings usually happen in the first 1-2 lines.
- Good reels often work best when they answer one clear question.
- If the topic is dense, simplify instead of cramming.
- If the user wants more depth, create a series or a longer script rather than overpacking one reel.
Visual Guidance
If the user wants visuals, offer:
- - scene image concepts
- thumbnail/cover concept
- on-screen headline text
- optional B-roll ideas
Do not assume full video automation exists yet. In v1, focus on producing a strong reel package that could later be recorded, edited, or automated.
Preview Generation
There are now two preview paths:
1. Text-card preview
A simple preview path can use
scripts/make_reel_preview.py to turn scene text into a rough vertical MP4 preview. This is for pacing and structure review, not final polished production.
Example flow:
- 1. save scene text into a file with one scene per paragraph
- run:
CODEBLOCK0
2. Visual scene preview
For a less slideshow-like reel, use
scripts/make_visual_reel_preview.py with scene JSON that points to per-scene background images. This keeps the old workflow intact while allowing a more video-like look with motion backgrounds and lighter text overlay.
Scene JSON shape:
CODEBLOCK1
Run:
CODEBLOCK2
Voiceover-Ready Workflow
The next production layer can add narration audio, then mux it with the preview video.
Current helpers:
Example flow:
- 1. generate narration audio with OpenAI TTS
CODEBLOCK3
- 2. mux it with the preview video
CODEBLOCK4
Expected env var:
Current voice style presets:
- - INLINECODE5
- INLINECODE6
- INLINECODE7
- INLINECODE8
- INLINECODE9
- INLINECODE10
For sports and fast-moving updates, prefer:
- - INLINECODE11
- INLINECODE12
- INLINECODE13
This keeps voiceover modular: script generation, preview assembly, narration generation, then final mux.
LTX Video Backend (Optional)
For a true AI-generated video option, Reelsmith can also use LTX via API instead of the local ffmpeg preview path.
Current helper:
Example:
CODEBLOCK5
Expected env var:
Recommended starting defaults:
- - model: INLINECODE16
- resolution: INLINECODE17
- duration:
8 or INLINECODE19
Use LTX mode when the user wants a reel to feel like a generated video instead of a slide-based preview.
Suggested Next Steps
After generating a reel package, offer one useful next step such as:
- - "Want a punchier hook?"
- "Want this turned into a 20-second version?"
- "Want a matching cover image?"
- "Want a rough preview video?"
- "Want this adapted for YouTube Shorts too?"
Reelsmith
概述
使用此技能将一个主题转化为适合Facebook Reels的完整内容包。输出应体现创作者的实际工作流程,而非单纯的文本块。
将核心单元视为:
- - 钩子
- 简短脚本
- 场景分解
- 屏幕文字
- 文案
- 视觉方案
- 发布或制作前的审批
核心工作流程
- 1. 确定主题和目标。
- 可处理体育资讯、新闻摘要、解说、产品演示、快速评论、活动回顾或推广创意。
- 明确Reels应实现以下哪种目的:
- 提供信息
- 做出反应
- 进行解释
- 进行推广
- 进行回顾
- 2. 确定目标风格。
- 常见模式:
- 信息型
- 活力型
- 专业型
- 精炼型
- 推广型
- 若用户未指定,默认采用
信息型+互动型。
- 3. 确定Reels时长。
- 默认时长为
30-45秒。
- 通常越短越好,除非主题明显需要更多篇幅。
- 4. 撰写有力的钩子。
- 第一行应让观众立即产生兴趣。
- 优先保证清晰度,避免空洞的炒作。
- 钩子朗读时应自然流畅。
- 5. 构建短视频脚本。
- 句子保持简短,便于口播。
- 避免在Reels中塞入过多信息。
- 仅聚焦于最有力的2-4个要点。
- 6. 将脚本分解为场景。
- 每个场景应包含:
- 口播内容
- 画面展示
- 可选屏幕文字
- 常规Reels建议设置4-7个场景。
- 7. 添加包装元素。
- 包括:
- 文案
- 轻量标签建议
- 封面/缩略图概念
- 可选视觉提示方案
- 8. 以审批为先。
- 不要假设用户希望立即发布。
- 清晰呈现Reels包,以便审核、修改或后续制作成视频。
输出格式
除非用户另有要求,请使用以下结构:
Reels概念: [一句话概念]
基调: [基调]
时长: [预估秒数]
钩子
[钩子]
脚本
[完整短视频脚本]
场景分解
- 1. [场景1 画面 + 口播 + 屏幕文字]
- [场景2 画面 + 口播 + 屏幕文字]
- [场景3 画面 + 口播 + 屏幕文字]
文案
[文案]
标签
[列表]
封面概念
[标题 + 视觉方向]
审批状态
等待用户审批后制作或发布
写作规则
- - 优化口播清晰度。
- 写作方式应自然流畅,便于朗读。
- 保持钩子尖锐有力。
- 避免冗长的开场。
- 避免在短Reels中塞入过多信息。
- 优先采用易记的结构,而非详尽的细节。
Reels专属指导
- - 强有力的Reels开场通常出现在前1-2行。
- 优秀的Reels通常在回答一个明确问题时效果最佳。
- 若主题内容密集,应简化而非堆砌。
- 若用户希望更深入,应创建系列或更长的脚本,而非在一个Reels中过度填充。
视觉指导
若用户需要视觉效果,可提供:
- - 场景图像概念
- 缩略图/封面概念
- 屏幕标题文字
- 可选B-roll素材方案
不要假设已存在完整的视频自动化功能。在v1版本中,重点在于制作一个强有力的Reels包,以便后续录制、编辑或自动化处理。
预览生成
目前有两种预览路径:
1. 文字卡片预览
一种简单的预览路径可使用scripts/make
reelpreview.py将场景文字转换为粗略的竖屏MP4预览。此方式用于检查节奏和结构,而非最终的精美制作。
示例流程:
- 1. 将场景文字保存到文件中,每段一个场景
- 运行:
bash
python3 scripts/makereelpreview.py --title 我的Reels --scenes-file scenes.txt --output preview.mp4
2. 视觉场景预览
如需更少幻灯片效果的Reels,可使用scripts/make
visualreel_preview.py配合场景JSON文件,该文件指向每个场景的背景图像。此方式保留原有工作流程,同时通过动态背景和更轻量的文字叠加实现更接近视频的效果。
场景JSON格式:
json
[
{bg: /path/to/image1.png, title: 主标题, subtitle: 简短辅助文字, duration: 5.0},
{bg: /path/to/image2.png, title: 下一个要点, subtitle: 另一句简短文字, duration: 5.0}
]
运行:
bash
python3 scripts/makevisualreel_preview.py --scenes-json scenes.json --output preview.mp4
配音就绪工作流程
下一制作阶段可添加旁白音频,然后与预览视频合并。
当前辅助工具:
- - scripts/openaitts.py
- scripts/muxreel_audio.py
示例流程:
- 1. 使用OpenAI TTS生成旁白音频
bash
python3 scripts/openai_tts.py --text 你的Reels旁白 --style energetic --output narration.mp3
- 2. 与预览视频合并
bash
python3 scripts/muxreelaudio.py --video preview.mp4 --audio narration.mp3 --output preview-with-audio.mp4
所需环境变量:
当前语音风格预设:
- - default(默认)
- calm(平静)
- professional(专业)
- energetic(活力)
- confident(自信)
- broadcast(广播)
对于体育和快节奏更新,推荐使用:
- - energetic(活力)
- broadcast(广播)
- confident(自信)
此方式保持配音的模块化:脚本生成、预览组装、旁白生成,最后合并。
LTX视频后端(可选)
如需真正的AI生成视频选项,Reelsmith也可通过API使用LTX,替代本地ffmpeg预览路径。
当前辅助工具:
- - scripts/ltxtextto_video.py
示例:
bash
python3 scripts/ltxtextto_video.py \
--prompt 一部电影风格的竖屏体育视频,展示一场高风险的大学篮球对决,戏剧性的场馆灯光、观众热情和动感的镜头运动 \
--duration 8 \
--resolution 1080x1920 \
--output ltx-video.mp4
所需环境变量:
推荐起始默认设置:
- - 模型:ltx-2-3-fast
- 分辨率:1080x1920
- 时长:8或10
当用户希望Reels呈现为生成的视频而非基于幻灯片的预览时,可使用LTX模式。
建议后续步骤
生成Reels包后,可提供一个有用的后续步骤,例如:
- - 想要更有冲击力的钩子?
- 想要将其改为20秒版本?
- 想要匹配的封面图片?
- 想要粗略的预览视频?
- 想要也适配YouTube Shorts?