HOW TO — 2 / 6
Starting with one video you have already shot, we will go from creating a project to a finished MP4. You do not need to memorize commands: the example prompts are ordinary conversational Japanese, translated into English here.
1 — Create a project
Once you have installed it, launch akari and tell the AI agent:
One video = one folder. The template automatically creates the following:
See the File map for details on each folder. For now, just remember: footage goes into assets/, and finished videos come out of exports/.
2 — Choose your workflow
Each project has a Workflow form (進め方フォーム). Fill it in so the AI can start working. It is a form, but you only need to answer in chat:
| Question | Meaning | Example answer |
|---|---|---|
| Task (やること) | What you want to make | “One short video from my footage” |
| Length / output (尺・出力先) | Duration and format | “60 seconds, vertical” |
| Autonomy (おまかせ度) | How independently the AI may work | Choose one of the three levels below |
| Autonomy (おまかせ度) | Behavior |
|---|---|
| checkpoint(default, recommended) | Asks for approval only at key stages (editing plan and export), and handles the rest independently |
| collaborative | Discusses each step with you. For people who want detailed involvement |
| full-auto | Works through to the end without asking. Best once you are familiar with the workflow |
3 — Add footage and finish the video
Put your footage in assets/ (dragging in Finder is fine), then say:
The AI automatically creates a lightweight review video (a 720p proxy), a transcript, and keyframes. This runs locally, so it is free.
The AI explains its cut choices in a report with images and edits only after you approve. If the plan does not suit you, request changes here.
Following your approved plan, the AI assembles a first cut with cuts, telops, subtitles, and background music. All edits are recorded in one save-data file, edit.json.
Play the preview, drag telops to move them, and double-click to fix text. See the next page, Open and fix, for details.
Once you approve, the final render uses the original footage and saves an MP4 in exports/. The AI then checks the output with automated verification and visual inspection.
4 — Additional notes
Start with “I want to make a video.” The AI asks questions, develops a plan, and suggests how to source assets (concept → outline → storyboard → shot list).
Work handled locally—proxy generation, transcription, editing, and export—is free. Charges apply only when using external APIs, such as cloud transcription or narration generation. Setup and approval are always required beforehand.