In the lab · Built and working, but only in the direct download of mChatAI+, behind the Show Experimental switch on the home screen.
AI Director
Describe the video you want, and it picks the clips, sets the cut points and renders the finished file.
In the lab · Experimental, in the direct download only — switch on Show Experimental on the home screen to see it
What it does
- Builds a media library from files you import or from your Apple Photos library, with search, 'Analyzed' and 'Transcribed' filter chips, and a running count of how many assets have been processed
- 'Analyze All' walks each clip in turn: pulls keyframes, transcribes the audio (OpenAI Whisper API or Apple's on-device speech recognition), then sends the frames and transcript to a vision model (Google Gemini 2.5 Flash, OpenAI GPT-4o, Anthropic Claude 4.5, or Apple's on-device Vision framework) for a description, sentiment, quality score and per-scene notes
- In the Direct step you type a prompt, pick one of seven styles (Energetic, Cinematic, Documentary, Highlights, Story, Slow & Artistic, Fast-Paced) and drag a duration slider from 15 to 180 seconds; a 'Manual Direct' mode skips the model entirely and spreads the selected assets evenly across that duration
- The model is handed every analyzed clip's description, sentiment, quality score, scene list and timestamped transcript excerpts, and returns an ordered edit list — which asset, which in/out seconds, which transition, and a one-line reason for each cut
- The Edit step is a working timeline: trim in and out points, split, merge, reorder, choose transitions with their own durations, add audio tracks with their own trim and start time, and add text overlays with font, weight, size, colour, position, shadow and outline controls
- 'Scan Library & Suggest' asks a model to group your whole library into named clusters — trips, best shots — that you can pick from instead of hand-selecting clips
- Renders to MP4, MOV or M4V from 480p to 4K, including 9:16 TikTok/Reels, 1:1 square and portrait presets, then plays the result inline with 'Save As...' and 'Show in Finder'
- A 'Debug Logs' sheet records every analysis, transcription and render step, filterable by category and level and searchable, so a failed run can be inspected rather than guessed at
How to try it
- On the home screen turn on the 'Show Experimental' toggle, then open the AI Director tile.
- In the left Media Library pane press 'Import Files' (or 'From Photos'), then press 'Analyze All' — the button shows the pending count — and wait for the 'analyzed' and 'transcribed' totals to climb.
- In the right pane press 'New Project', add clips in the Select step, then press 'Continue to AI Director'.
- In Direct, type a prompt or press the 'Highlight reel' example button, pick a Style and Duration, and press 'Generate Edit'.
- Step on to Export, choose a Resolution and Format, press 'Render Video', then 'Save As...'.
Works with
- Apple Photos (import through the system photo picker, with an authorization request on first use)
- AVFoundation for keyframe extraction, audio extraction and the final composition
AI Director ships with mChatAI+. Native macOS app. Your own API keys, or fully on-device. Get mChatAI+