🎬 AI Video Editor

Edit your videos by just talking to it. Auto-cut filler, auto-place B-roll, auto-caption, and turn one long video into ready-to-post Shorts.

The open-source AI video editor for founders, creators, and marketers. Feed it a raw talking-head recording → get back a polished, captioned video plus vertical Shorts plus ready-to-post captions, hooks, and hashtags. No timeline. No menus. You talk, it edits.

Quick Start · What it does · Make it yours · How it works · FAQ

---

Why this exists

Editing video is the #1 bottleneck between a founder and consistent content. Hiring an editor is slow and expensive; learning Premiere is a rabbit hole; the "one-click" apps give you generic slop you can't customize.

This is different. It's an AI video editor you run yourself (or inside FounderOS) that does the mechanical 90% of editing — the part that's identical every time — to a consistent, professional standard, and lets you keep the taste. You describe what you want in plain English, it shows you the plan, you approve, it renders. Then it hands you Shorts and ready-to-paste captions for every platform.

Honest promise: this reliably produces solid, B-level automated edits — clean cuts, captions that pop, B-roll landed on the right word, color graded, loudness-normalized, with Shorts and post copy. It is A-level consistent: it nails the repeatable craft every time. The creative taste calls (which B-roll meme, the exact joke timing) stay yours — it shows you options and learns your style. It does not pretend to replace a senior editor's judgment; it replaces the 3 hours of tedium before judgment matters.

---

✨ What it does

🎙️ Transcribes	Word-level transcript so every cut, caption, and B-roll lands on the exact word.
✂️ Cuts filler + dead air	Removes "um, uh, like, you know," false starts, and long pauses — snapped to word boundaries so nothing clips.
🎞️ Finds + places B-roll	Classifies each moment (receipt / entity / concept), sources it from sources you trust, and lays it on the word being spoken.
🎨 Color grades	Auto, subtle, punch, or cinematic — or your own look.
💬 Burns captions	Bold, readable captions with pain words in orange, money words in green — the style that stops scroll. Auto-fits so text never clips on vertical.
📱 Finds the viral moments	Scores every moment by how it's delivered — audio energy spikes, laughter, dramatic pauses — not just the words, then an LLM picks the clips that would actually perform (rejecting stumbles and mid-thoughts) and cuts them into Shorts.
📝 Writes the captions to post	Per-platform caption + hook + hashtags for TikTok, Reels, YouTube Shorts, X, and LinkedIn.
🌐 Builds a review page	A beautiful HTML page: every clip playable, copy-paste captions per platform, posting checklist. Double-click to open.
🤖 Runs on autopilot	Drop videos in a folder → edited videos + Shorts + post copy come out.
🎯 Edits in your style	One taste.yaml file. Change a value → every future edit matches you.

---

🚀 Quick Start (2 minutes)

Requirements: macOS or Linux, Python 3.10+, ffmpeg. For best transcription, an ElevenLabs API key (free tier works). For sourcing B-roll from the web, yt-dlp.

# 1. Clone it
git clone https://github.com/EverythingAI-Pro/ai-video-editor.git
cd ai-video-editor

# 2. One-time setup (creates a local Python env with Pillow, numpy, requests)
bash scripts/setup.sh

# 3. Add your transcription key (optional but recommended)
cp .env.example .env        # then paste your ELEVENLABS_API_KEY

# 4. Edit a video — cut, caption, B-roll, Shorts, AND a post-ready review page
.venv/bin/python scripts/edit.py /path/to/your-video.mp4

When it finishes you'll have, next to your video in an edit/ folder:

edit/
├── your-video.edited.mp4      ← the polished main edit (cut + captions + color)
├── your-video.short0.mp4      ← vertical Shorts pulled from it
├── review.html                ← OPEN THIS: play everything, copy captions, post
├── your-video.postpack.json   ← captions + hooks + hashtags per platform
└── manifest.json              ← what happened

Open review.html and you've got a clip-by-clip dashboard with copy-paste captions for every platform and a posting checklist. That's the whole loop: raw recording in, content calendar out.

Using this inside FounderOS? You don't touch the terminal at all — just say "Edit my video" in chat and attach a clip. See SKILL.md.

---

🎛️ Make it edit like you (taste.yaml)

This is the part that makes it yours. One commented file controls everything:

captions:
  font: "HelveticaNeue-Bold"   # or "Impact", "Menlo", or a path to your .ttf
  case: "upper"                # UPPER (ad style) · sentence · lower
  words_per_line: 3            # 2 = punchy · 4-5 = calm
  position: "lower"            # lower (safe for Shorts UI) · center · upper
  highlight: true              # pain words orange, money words green
cut:
  aggressiveness: "medium"     # gentle · medium · tight
  remove_silences_over_s: 1.2
color:
  grade: "auto"                # auto · subtle · neutral_punch · warm_cinematic · none
broll:
  density: "medium"            # light ~10% · medium ~25% · heavy ~50%
  trusted_sources:             # B-roll search stays INSIDE this list
    - "@anthropic-ai [tech]"
    - "Lenny's Podcast [business]"
shorts:
  count: 3
  min_hook_score: 55           # only clips with strong hooks ship

Change one value, save, re-run — your next edit reflects it. Make your captions yellow and centered, cut tighter, switch to a cinematic look — it's all one file, no code.

---

🧠 How it works

your video ─▶ transcribe ─▶ cut filler/silence ─▶ find + place B-roll ─▶ color grade
                                                                              │
   review.html ◀── post pack ◀── make Shorts ◀── burn captions ◀────────────┘
   (open & post)   (per platform)  (hook rubric)   (last, over the b-roll)

Each step is a small, readable Python script in scripts/ that shells out to ffmpeg. Nothing is a black box:

• ve_transcribe.py — word-level transcript (ElevenLabs Scribe, with caption + provided-JSON fallbacks)
• ve_cut.py — filler + silence removal → an edit plan you can read
• ve_broll.py — route classification + trusted-source search + place-on-the-word
• ve_captions.py — caption rendering (works even where ffmpeg lacks libass)
• ve_render.py — assembles it all in the correct order
• ve_shorts.py — the 8-pattern viral-hook rubric
• ve_postpack.py — per-platform captions / hooks / hashtags
• ve_publish.py — the HTML review page
• edit.py — runs the whole thing in one command
• autopilot.py — folder in → edited videos out

See references/ for the editing craft (B-roll routing, caption craft, Shorts rubric) and SKILL.md for the conversational flow.

---

🤖 Posting (optional)

The recommended way to post is native upload from the review page (better reach than API posts) — the page gives you copy-paste captions and a checklist. If you want true hands-off scheduling, ve_post.py can push to Buffer or your own Make/Zapier/n8n webhook using your keys. It's off by default and never posts without --confirm.

---

❓ FAQ

Is it actually good, or is it AI slop? It does the mechanical work (cuts, captions, color, B-roll timing, Shorts, post copy) to a consistent professional bar, and keeps you in the loop on taste calls. The output is genuinely postable. It is not a magic "make me Casey Neistat" button — nothing is.

Do I need to know how to code? No. Run one command, or use it inside FounderOS by saying "Edit my video." To customize, you edit one plain-English config file.

Do I need API keys? Only for best-quality transcription (ElevenLabs, free tier fine) and web B-roll sourcing (yt-dlp, free, no key). Everything else — cutting, captioning, color, Shorts, the review page — runs locally with no keys.

What platforms does it make content for? Vertical Shorts for TikTok, Instagram Reels, and YouTube Shorts, plus caption packs for those + X and LinkedIn.

Does it work on Windows? The engine is ffmpeg + Python, which run on Windows, but setup is tuned for macOS/Linux. Windows via WSL works today; native Windows support is on the roadmap.

---

🛠️ Built on great open source

• The editing pipeline and production-correctness rules build on the proven video-use methodology.
• B-roll sourcing/placement is adapted from b-roll-finder by Louise de Sadeleer (MIT — carried in vendor/b-roll-finder/).
• Captions, color, and concat run on ffmpeg. Transcription via ElevenLabs Scribe. B-roll downloads via yt-dlp.

📄 License

MIT — see LICENSE. Use it, fork it, ship it.

---

Built by EveryThingAi — the operating system for founders running an AI company. If this saved you an afternoon, ⭐ the repo so other founders find it.

ai video editor · auto captions · auto b-roll · make shorts from long video · ffmpeg video automation · content repurposing · faceless video editing