Raw Footage to Rough Cut with FCPXML Export
Turn raw talking-head takes into an approved paper edit, a frame-accurate FCPXML/EDL timeline, and captions synced to the final cut.
Before & after
Same prompt, same agent — the only difference is the skill.
Video editing is time-consuming because you have to watch raw footage, decide what to keep, fix audio, add captions, potentially add music or graphics. Normally you'd use Adobe Premiere or Final Cut Pro, but those require manual work for every step. Some people use tools to help with specific parts like transcription or auto-captioning, but coordinating it all is tricky. If you had raw footage and wanted an autonomous system to edit it, you'd need to tie together transcription, scene selection, effects, and caption generation. That's complex. How long is your raw footage?
BUTTERCUT VIDEO EDIT PIPELINE INPUT Raw footage: 45 minutes of unedited talking-head interview (MP4, H.264) Requirement: autonomous editing, no manual touching PIPELINE (6 steps, orchestrated by Claude Code) STEP 1: TRANSCRIPTION (5 min) ButterCut extracts audio from video Feedic.ai transcription service converts speech to text with timestamps Output: Full transcript with timing (0:00 → 45:30) STEP 2: ANALYSIS (3 min) Claude Code skill analyzes transcript: • Identifies natural story arcs (problem → solution → result) • Flags key moments (when speaker becomes animated, pauses for emphasis) • Detects technical issues (background noise, audio dropout) • Scores engagement (high → keep, low → trim) STEP 3: ROUGH CUT (8 min) ButterCut removes: • Long pauses (>3 seconds) • Filler words (um, uh, like) • Tangents (off-topic sections < 10 seconds) • Technical errors (audio dropout, phone rings in background) Result: 45 min → 28 min STEP 4: CREATIVE DIRECTION Creative director persona (Claude) decides: • B-roll insertion points (when speaker gets abstract, show examples) • Graphics (when speaker mentions metrics, display chart) • Music (emotional underscore during key moments) • Color grading (consistent tone throughout) STEP 5: EFFECTS AND CAPTIONS Motion graphics added: • Speaker name and title (0:00 → 5:00) • Key quotes highlighted (auto-positioned on screen) • Charts for data points (animated in when mentioned) Captions generated from transcript: • Burned into video (subtitle track embedded) • Punctuated correctly (manual QA check) • Speaker labels (if multi-person interview) STEP 6: FINAL EXPORT Format: 1080p MP4, YouTube-optimized (h.264, AAC audio) Duration: 28 minutes (90% less than raw footage) File size: 2.1 GB Timestamp: ready for upload END-TO-END TIME Total processing: 20 minutes (automated) Vs manual editing in Premiere: 8-12 hours (for same quality) OUTPUT CHECKLIST ✓ Rough cut removes dead time ✓ B-roll inserted for visual interest ✓ Captions match video timing (within 100ms) ✓ Audio levels consistent (no volume jumps) ✓ Color grading uniform (no sudden shifts) ✓ Title cards added (viewer context) ✓ Outro links to CTA (calls to action) ✓ File ready for YouTube upload (resolution, aspect ratio) CUSTOMIZATION OPTIONS • Target duration (edit to 10 min, 15 min, 30 min) • B-roll source (royalty-free library, custom clips, none) • Music style (cinematic, upbeat, minimal) • Caption style (burned-in, .srt file, none) • Creative persona (documentary, informal, corporate) COST CALCULATION Manual editing (freelancer): 400-600 per hour × 10 hours = 4000-6000 ButterCut automated edit: 45 (processing) + 30 (prompting) = 75 ROI: 50x faster, 98% cost reduction
About this skill
You have 40 minutes of takes, restarts, and "sorry, let me try that again," and scrubbing through it eats your whole afternoon. AI attempts at editing usually go wrong in three ways: words get clipped at the cut points, the XML won't import or lands on the wrong frames, and the captions drift out of sync after the first cut. This skill works like an assistant editor. It picks your best takes, lays out a paper edit for you to approve, and hands back a frame-accurate timeline you can open in your editor.
What it does
- Probes every clip (frame rate, timecode start, audio channels) and flags variable-frame-rate phone footage and mixed frame rates before they break the timeline
- Transcribes verbatim with word-level timestamps, so restarts and stumbles stay visible for take selection
- Ranks repeated takes by completeness, stumbles, filler rate, pace, and audio clipping, and gives a reason for each pick
- Builds a paper edit with source in/out points, a hook-first structure, and a running time against your target
- Writes FCPXML (or FCP7 XML / EDL for Premiere and Resolve) on the exact frame grid, with source timecode handled and markers for B-roll and graphics
- Generates SRT/VTT captions that follow the edited timeline, using standard line-length and reading-speed limits
What you get
A paper edit table to approve, then a generator script that outputs an importable timeline file and a captions file synced to the cut, plus editor notes for jump cuts that need cover.
Best for
YouTubers, course creators, and small video teams who shoot talking-head or interview footage, finish in a real NLE, and want Claude Code to do the assembly edit.
SKILL.md preview
---
name: raw-footage-rough-cut-fcpxml
description: Turns raw talking-head or interview footage into an editable rough cut. It probes the media, transcribes with word timestamps, ranks repeated takes, and builds a paper edit for approval. It then writes frame-accurate FCPXML (or FCP7 XML/EDL) and captions remapped to the edited timeline. Use when the user says "cut my raw footage into a rough edit", "pick the best take", "turn this transcript into a Final Cut timeline", "make 40 minutes of rambling into a 6 minute video", "add captions to my edit", "my XML won't import into Final Cut", "my clips land on the wrong frames", "my captions are out of sync after the cut".
---
… (sign up to view the full skill)More creator skills
View all Creator skills →Remotion Explainer Scenes Synced to Voiceover
Turns a timestamped voiceover into Remotion scenes: frame-exact timeline, one style module, exact charts and counters, and a sync check.
AI Film Look Bible and Intention Lock
Turns 'make it cinematic' into a locked film bible: ending, camera stance, one light state, measured look, cast, and a contradiction check.
Remotion Product Teaser From Your Website
Turns your site's real logo, colors, copy, and screenshots into a frame-exact storyboard and Remotion code for a 10-60s teaser.
Long-Form to Shorts Clip Finder
Find the moments in a long video, podcast, or post that stand alone, and package them as verbatim clips or scripts with exact cuts.