Training-video editing: which AI tool to use
Florist training videos (Akram's OKR). Tested on the approved video in Floral Photography/11. Nour Zoghly, Sunday September 27, 2026.
Recommendation. Three AI editing tools, each strong at a different part of the job. Option 1 (Claude + a Floward skill) already made a first cut from the raw clips in about 5 minutes that closely matches the approved edit. Option 2 (Eddie AI) is a ready-made AI editor that builds the cut from a reference video, for about $3 to $10 per video. Option 3 (Wideframe) is the only AI tool that claims to line up silent cameras on its own, then hands a real editor a ready project. Suggested path: pilot Option 1 on video 2, and test Option 2 on its free credits in parallel. All three cost far less than the $200 per video we pay freelancers.
1. What the job actually is
I measured the approved edit against its raw clips. It is not a talking video: the three raw clips have no sound at all. That one fact rules out most "AI video editors", because they decide where to cut by listening to speech.
Around that sit things that are identical in every video: the front and side shots zoomed in, the overhead turned sideways, a slightly flat colour look, a 4-second blurred opening with the FLOWARD logo, a 6-second blurred ending with the logo, and a music track underneath.
The hard part is lining the three cameras up. With no sound, you can't line them up by audio the way editing software normally does. They have to be lined up by matching the picture. Of the AI editors checked, only one (Wideframe) even claims to do this automatically, and nobody has tested it independently. Option 1 does it, tested on our footage.
2. Proof: Claude's first cut next to the approved edit
Left: the approved edit. Right: what the Claude option produced on its own from the three raw clips (the logo and music are placeholders). Same order, same camera logic, same zoom and look, and the length is 6:06 against 6:05.
v2: re-cut using the editing handbook
v2 was made only from the raw clips and the editor's handbook, so each cut follows the hand action and each camera is chosen by the type of work, where v1 copied the look of the approved edit.
Honest differences: the exact moment of each switch is not identical (an editor would nudge a few), the logo is a stand-in font, and I borrowed the approved video's music. It has been tested on one video only. Full-quality file: analysis/demo/claude-first-cut-v3.mp4.
Blind test. A fresh Claude session with no notes from me, only the skill, correctly worked out which clip was which camera, lined them up, and rendered a 6:05 cut in under 2 minutes. It found 11 rough edges (a wrong time label on the review sheets, music shorter than the video, unclear rules for the opening and ending shots, side framing that could only move up and down). The fixes for those are already in the skill.
3. The three options
Claude + a Floward editing skill
A skill is a saved set of instructions and small programs that Claude follows. The operator drops the three clips in a folder and asks Claude to "make the training cut". Claude lines up the cameras by picture, writes a shot list (which camera, when, and why), shows it for approval, then renders the finished video in the house style.
- The only option tested on our footage, including a blind run
- The only one that picks angles by what the hands are doing, following rules measured from the approved edit
- House style locked in one file, so every video matches
- A prototype, proven on one video so far
- Needs a one-time setup on the operator's Mac
- No timeline to drag clips on; changes are made by asking Claude
Eddie AI (version 4)
An AI "assistant editor" on the web. It watches and logs every clip, groups the three cameras, and builds a cut that switches between them. New in version 4: you can hand it an existing video and say "cut it like this", so our approved edit becomes the brief. It delivers an MP4, or a project that opens in Premiere, Resolve or Final Cut.
- The most mature AI editor on the market (4th version)
- Uses the approved edit as a reference, not just a written prompt
- Cheapest per video; free to test right now
- Can also be run from inside Claude
- Can't line up silent cameras by itself; it asks you for the offset
- Its documented camera switching follows the speaker and shot tightness, which don't exist in our footage. Whether it can switch by "where the hands are" is untested.
- 1.35x speed, exact zooms and the blurred-logo opening aren't documented as things it does
Tip: Option 1's line-up step produces exactly the offsets Eddie asks for, so the two can be combined.
Wideframe
A Mac app built on Claude that works like an AI coworker for editors. It indexes every frame of the footage by what's in it, lines up the cameras (it says it uses visual cues such as movement when there's no sound), and builds a ready-to-cut Premiere project from a plain-English brief.
- The only AI tool that claims to line up silent cameras automatically
- Runs on the Mac: footage never leaves the machine
- Fits editors who already work in Premiere
- Stops once the cameras are lined up: the creative cut is still done by hand
- Its silent line-up claim comes only from the vendor's blog (not independently tested)
- Needs an Apple Silicon Mac and a Premiere licence; the most expensive option
Side by side
| 1. Claude skill | 2. Eddie AI | 3. Wideframe | |
|---|---|---|---|
| Lines up the silent cameras | Automatic (tested) | You type one offset per camera | Automatic (vendor claim) |
| Picks the camera angles | Automatic, by what the hands do | Automatic, but speaker-based (untested on silent) | No, by hand in Premiere |
| Learns from the approved edit | Rules measured from it | Takes it as a "cut it like this" reference | No |
| House style (speed, zoom, logo, music) | Locked, automatic | Partly (colour, music, titles) | By hand |
| Output | Finished MP4 | MP4, or a Premiere/Resolve/Final Cut project | Premiere project |
| Cost | Claude seat only | ~$3 to $10 per video | $100/month + Premiere |
| Person's time per video | ~15 to 20 min | ~30 to 60 min (est.) | ~1 hour+ (est.) |
| Free trial | n/a | 1,000 credits, no card | 7 days, card required |
| Tested on our footage | Yes | No | No |
4. Other AI editors checked
I checked 20 AI editing tools. Almost all of them choose the camera by who is speaking, so silent footage gives them nothing to go on.
| Tool | Verdict |
|---|---|
| Adobe Firefly Quick Cut | Worth one free test. Assembles a first cut from the picture with style prompts, but treats the three cameras as separate takes, so it will likely scramble the order. Beta, stays inside Firefly. |
| Mosaic | Worth a look later. Best at saving a repeatable recipe (intro, outro, colour, music) and running it on every video automatically. It has no multi-camera step, and no free trial ($50/month to test). |
| Invideo Editor (new, Sep 2026) | Needs one camera with sound to line up the rest, and switches by speaker. Revisit if we record sound. |
| Cardboard, Kapwing, ChatCut | Good AI editors for single-camera clips; no way to combine lined-up cameras. |
| Descript (Underlord), Cutback Selects, Loopdesk, Premiere AI Assistant | Camera switching follows the speaker. |
| Gling, AutoCut, FireCut, Opus Clip | Built around speech: cutting pauses and filler words, or making short clips. |
| Runway Aleph, Google Vids / Flow | Restyle short clips (2 to 30 seconds). They don't assemble an edit. |
| video-use, VEED OpenEdit, Twelve Labs Jockey | Toolkits you run with your own AI agent. That is what Option 1 already does, and it borrows video-use's method. |
| Kino, Vizard | Too little public detail to judge. |
Non-AI fallback, not one of the three: if the team would rather cut by hand, a normal editor with a saved template works too (DaVinci Resolve is free).
5. One change at the shoot that helps every option
Start each session with one clap, visible to all three cameras, and leave the phone microphones on. With sound, Eddie AI and most of the other AI editors line the cameras up automatically, which removes Option 2's manual step. It also makes Option 1's automatic line-up foolproof. It costs nothing. It does not fix angle choice: with no one speaking, only Option 1 picks angles by what the hands are doing.
6. Next steps
- Test Option 2 for free: upload the three clips of video 11 to Eddie with the approved edit as the reference. Score it on: cameras lined up correctly, order kept, how many of the ~27 switches match, minutes of fixing. This needs someone to create the account.
- Pilot Option 1 on video 2 to prove it works beyond one video; one-time setup on the operator's Mac.
- All options: the official FLOWARD logo file and font, and a licensed music track.
- Decision needed: who operates it (in-house junior, photographer or freelancer).
Sources
Everything below was checked on September 27, 2026. Prices marked as reported come from reviews, not official pages.
- Eddie AI: multicam guide (silent clips need a typed offset; switching by speaker and shot tightness), pricing, CineD on version 4 (reference edits, visual-only edits), No Film School (runs inside Claude)
- Wideframe: home, pricing, multicam sync blog (vendor claim of visual sync; switching left to the editor)
- Adobe Firefly Quick Cut: Adobe help; Mosaic: docs, launch thread
- video-use: github.com/browser-use/video-use ("audio is primary"; open issues #153, #157, #188)
- Descript automatic multicam: help.descript.com; Gling: gling.ai; AutoCut: autocut.com
- Full research notes:
research/ai-native-editors.md(20 AI editors, trials, test protocol),research/agentic-tools.md,research/commercial-tools.md. Measurements:skill-prototype/floward-training-edit/reference/edit-rules.md.