How AI edited SV Academy's Instagram videos
The brief, the footage, the storyboard and the feedback: a practical look inside two videos edited entirely with AI.
SV Academy and SV News share a publisher.Details +
Jump to a section
Sandee is sitting in an interview chair, explaining why people caught in a flood needed better information. A short video has to make that problem matter, show his response and leave the viewer with a complete thought. That means choosing what to say first, which pictures belong under it and where to stop.
The Sandee reel and participant compilation posted by SV Academy were 100% AI edited, the academy confirms. Prom supplied the creative direction, reviewed the cuts and requested changes. AI carried out the editing. This is an account of our own production process: SV Academy is part of SV Innovation, the publisher of SV News.
Start with the story you want people to understand
A useful brief says who the film is for and what they should understand when it ends. For Sandee, the thread is a problem people faced, the information they lacked and the map he built in response. His face and voice give the story a person to follow.
The source records connect each selected passage to an interview or demonstration and its timecodes. The storyboard then gives every beat four jobs: show the picture, specify the sound, write the subtitles and explain why that beat follows the previous one. You can inspect the reasoning before committing to a render.
This is where coherence becomes something you can check. A shot of an application belongs when the speaker explains that application. A smiling classroom shot can support a moment of relief. If a picture adds no information or emotion to that particular sentence, the editor needs a reason to keep it.
Two published examples, two different structures
Start with Sandee's posted video. One person carries the story. The supporting pictures help explain what he is talking about, while Thai and English subtitles let readers follow the same speech. The app is context for his account. For a quick look at the product itself, visit Sanroo.
The second posted video asks why the participants wanted to build applications. A compilation needs a shared question to hold its different speakers together. The useful lesson is to select contributions that develop that question, then finish on a thought that resolves it. Giving everyone a turn is not, by itself, a story.
Give AI an editing method it can follow
Prom Motion turns the method into reusable instructions: source selection, storyboards, complete speech, bilingual captions and export checks. Version 1.0.1 adds a companion handoff guide for readers who want to prepare assets and build a finished composition:
- Prom Motion plans the story, records source timecodes and reviews the final video.
- SV Edits prepares artwork, trims existing footage and extracts frames. Download the SV Edits ZIP.
- SV Motion combines those assets with titles, reviewed subtitle cues and animation in an editable HTML composition, then renders it. Download the SV Motion ZIP.
These are separate skills. The guide connects their work through source IDs, trim times and caption cues; installing Prom Motion does not install the other tools. Use the companions where needed, or keep an existing FFmpeg workflow. This is a setup readers can try, rather than a claim that both companions produced the two reels above.
Follow the Prom Motion Quickstart to install the Claude Code plugin and invoke /prom-motion:prom-motion. For Codex, install the skills/prom-motion folder and invoke $prom-motion. Anthropic's plugin documentation explains plugin loading.
For the companions, follow the SV Edits setup guide and SV Motion setup guide. A ZIP contains skill sources; editing-tool downloads and configuration are separate. Ask your assistant:
Check whether SV Edits and SV Motion are installed. Read their setup guides and recommend only the tools this project needs. Keep the originals unchanged. Pass prepared clips, original timecodes and reviewed subtitle cues between steps, then use Prom Motion to review the exported video.
Give it a first instruction such as:
Make a 45 to 60 second story from these interviews and demonstrations. Show a specific problem, the response and a visible result. Propose three hooks that actually exist in the footage, with exact source times. Keep complete sentences and Thai above English subtitles. Show the storyboard first. Do not render, upload footage or buy anything yet.
Review the moments that determine whether it works
Once the storyboard is approved, ask for a short playable proof of the opening and one speech-to-picture change. Watch it with sound. A good sentence on a transcript can still have a hesitant start or a cut that lands before the speaker finishes.
Our revisions included weak openings, interrupted sentences, unrelated pictures and unwanted blur. Those problems required specific feedback and another export. A useful note is: “At 00:45 the sentence is incomplete. Find the full thought in the source, update both subtitles and keep the approved ending.”
AI can now carry out a coherent edit from a clear structure and respond to detailed creative direction. These published videos are examples of that capability. They also show why the brief, source evidence and review belong in the workflow. Start with one story and a short proof, then build on the version that works.
FAQ
Does 100% AI edited mean nobody directed the videos?
No. AI performed the editing; Prom supplied the brief, judged the cuts and requested revisions. The percentage describes how the edits were executed, not the absence of human direction.
Do I need generated scenes or paid video models?
No. Start with your existing footage. The method does not require new generated scenes. Local rendering has no per-render service fee, though agent usage, optional services and licensed assets can cost money.
Can the plugin work from any folder of clips?
It needs readable media, suitable tools and enough context to identify the speakers and events. Supply transcripts when available. Ask it to flag missing evidence and uncertain words before it edits them.
