FCP multicam workflow?

[terrylb]

We create long-form, interview-driven training programs using FCP. The typical workflow is to import 1 or 2 angles of FX-6 footage for each interviewee and create multicams. The interview audio is exported and transcribed with timestamps using Sonix.ai. I wrote an app that takes an FCPXML of each interviewee, the questions they were asked, and their transcription, which is analyzed with a Claude API call and exports an FCPXML with favorited keyword ranges of the interviewee's multicam. The ranges are named after the questions answered.

Phew, after all that out of the way, I'm always looking for new AI-based workflows to speed up the editing process and wondering how doza-assist handles FCP multicams? Or what a proposed workflow would be if it doesn't? Also, we use macports not homebrew. Will doza-assist have problems with that?

[DozaVisuals]

Thanks for the detailed breakdown, Terry. That's a solid workflow you've built out.

Multicam isn't supported yet, but it's on the roadmap. Here's a workaround that should fit your current setup:

Process each camera angle as a separate project in Doza Assist. Import each FX-6 angle individually, let it transcribe and generate the FCPXML with markers/ranges. Then in FCP, create your multicam clip as you normally would, and use the exported FCPXMLs as reference to drop keyword ranges onto your multicam timeline. Since your angles share the same audio timecode from the interview, the timestamps will line up.

It's an extra step compared to native multicam support, but it gets you the AI-generated keyword ranges per angle that you can then apply to your multicam edit. The transcription and analysis results will be identical since they're working from the same interview audio.

Multicam support is planned, and when it lands the idea is exactly what you'd expect: import the multicam clip directly and have it handle all angles in one pass. I'll flag this thread when it ships.

[DozaVisuals]

Quick follow-up, Terry — v2.6.1 ships native FCP multicam and sync-clip support, so you no longer need the per-angle workaround. Your FCP-first workflow now round-trips end-to-end:

Release: https://github.com/DozaVisuals/doza-assist/releases/tag/v2.6.1

Workflow for your setup

1. In FCP: select the multicam on the timeline → File > Export XML… → save the .fcpxmld bundle locally.
2. In Doza Assist: drop the .fcpxmld bundle onto the dashboard. It parses the <mc-clip>, finds the <mc-source srcEnable="audio"> angle, and transcribes the underlying audio file directly from your edit drive. No audio export, no proxies.
3. Verify the detected audio path at the top of the project view — it should match the interview audio track inside the multicam.
4. Make selects: highlight soundbites in the transcript, run AI Analysis for story beats / social clips, or let it auto-label with color ranges. Works the same as any non-multicam project.
5. Export: on the project's Export tab, scroll to the "FCPXML Round-Trip" section. ⚠️ This is the important bit — use this section, not the plain "Export FCPXML" button at the top of the Export tab. The plain button emits a flat asset-clip timeline (audio only); the Round-Trip section is what preserves your multicam container and re-attaches video. Pick one:
   • Selects as new project — each select becomes an <mc-clip> on a fresh timeline ("{project} - Doza Selects"), reusing the same ref, same angle enablement, and same <resources> block as your source FCPXML.
   • Markers on existing timeline — duplicates your original timeline with color-coded markers at each select (completion / standard / to-do).
6. Back in FCP: File > Import > XML… → pick the file. The new project or marker-annotated copy lands in your event, pointing at the same multicam that's already in your library. Video + synced audio intact, bookmarks preserved byte-for-byte.

Source: <resources> is spliced in verbatim from the original, so FCP's bookmark validation passes and no re-linking is needed.

On MacPorts

Doza Assist ships as a self-contained .app — it doesn't touch Homebrew. First launch runs a bundled setup_runner.sh that provisions a Python venv inside the app bundle. If MacPorts gives you a working python3 on PATH, it'll use that; otherwise the setup script will prompt. Give v2.6.1 a try and if anything snags on the MacPorts side, drop the setup.log into this thread and I'll take a look.

[terrylb]

Thanks Chris! I appreciate the follow-up. I just discovered SpliceKit https://splicekit.fcp.cafe/ <https://splicekit.fcp.cafe/> , which is built to actually run inside a copy of FCP and provides pseudo-native access to LLMs, so I want to explore this before trying a new app workflow. Thanks, -Terry On Apr 21, 2026, at 3:56 PM, Chris Cardoza ***@***.***> wrote: Quick follow-up, Terry — v2.6.1 ships native FCP multicam and sync-clip support, so you no longer need the per-angle workaround. Your FCP-first workflow now round-trips end-to-end: Release: https://github.com/DozaVisuals/doza-assist/releases/tag/v2.6.1 <https://github.com/DozaVisuals/doza-assist/releases/tag/v2.6.1> Workflow for your setup 1. In FCP: select the multicam on the timeline → File > Export XML… → save the .fcpxmld bundle locally. 2. In Doza Assist: drop the .fcpxmld bundle onto the dashboard. It parses the <mc-clip>, finds the <mc-source srcEnable="audio"> angle, and transcribes the underlying audio file directly from your edit drive. No audio export, no proxies. 3. Verify the detected audio path at the top of the project view — it should match the interview audio track inside the multicam. 4. Make selects: highlight soundbites in the transcript, run AI Analysis for story beats / social clips, or let it auto-label with color ranges. Works the same as any non-multicam project. 5. Export: on the project's Export tab, scroll to the "FCPXML Round-Trip" section. ⚠️ This is the important bit — use this section, not the plain "Export FCPXML" button at the top of the Export tab. The plain button emits a flat asset-clip timeline (audio only); the Round-Trip section is what preserves your multicam container and re-attaches video. Pick one: * Selects as new project — each select becomes an <mc-clip> on a fresh timeline ("{project} - Doza Selects"), reusing the same ref, same angle enablement, and same <resources> block as your source FCPXML. * Markers on existing timeline — duplicates your original timeline with color-coded markers at each select (completion / standard / to-do). 6. Back in FCP: File > Import > XML… → pick the file. The new project or marker-annotated copy lands in your event, pointing at the same multicam that's already in your library. Video + synced audio intact, bookmarks preserved byte-for-byte. Source: <resources> is spliced in verbatim from the original, so FCP's bookmark validation passes and no re-linking is needed. On MacPorts Doza Assist ships as a self-contained .app — it doesn't touch Homebrew. First launch runs a bundled setup_runner.sh that provisions a Python venv inside the app bundle. If MacPorts gives you a working python3 on PATH, it'll use that; otherwise the setup script will prompt. Give v2.6.1 a try and if anything snags on the MacPorts side, drop the setup.log into this thread and I'll take a look. — Reply to this email directly, view it on GitHub <#14?email_source=notifications&amp;email_token=ADDL7YZSEM7BKDGOPI5GZCD4W74BRA5CNFSNUABIM5UWIORPF5TWS5BNNB2WEL2ENFZWG5LTONUW63SDN5WW2ZLOOQXTCNRWGYZDGMZQUZZGKYLTN5XKMYLVORUG64VFMV3GK3TUVRTG633UMVZF6Y3MNFRWW#discussioncomment-16662330> , or unsubscribe <https://github.com/notifications/unsubscribe-auth/ADDL7Y7DRKFV7QMXPFNCBEL4W74BRAVCNFSM6AAAAACX5ROKUWVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTMNRWGIZTGMA> . You are receiving this because you authored the thread. <https://github.com/notifications/beacon/ADDL7Y76UYXG7UFZTLCVMYT4W74BRBFCNFSM6AAAAACX5ROKUWWGG33NNVSW45C7OR4XAZNRIRUXGY3VONZWS33OINXW23LFNZ2KUY3PNVWWK3TUL5UWJTQA7Y7TVJTSMVQXG33OUZQXK5DIN5ZA.gif> Message ID: ***@***.***>