The big one first, and it is overdue: Eddie is on your iPhone, and has been for a little while without us saying so. After that, a week of audio you can actually work with — every microphone kept separate, any track silenced for good, and captions that look the same in the export as they do on screen.
Eddie is on your iPhone
Eddie is in the App Store. Sign in and your projects are already there — the same projects, not a mobile-shaped copy of them.
Edit on the move. The cloud project model that landed in 3.5 is what makes this work: the project lives in the cloud, so the phone is just another way into the same edit. Start something on the web, pick it up on the train, hand it back to the desktop for native playback and local rendering. Nothing to sync, nothing to export, nothing to keep track of.
Stop waiting at your desk. Imports, analysis and edit runs take real time, and until now the only way to know Eddie had finished was to be sitting in front of the machine. The app notifies you the moment an edit or a task completes, so you can go and do something else and come back when there is actually something to look at.
Review, ask, export, invite. You can watch an edit, ask for changes in your own words the same way you would anywhere else, export the result, and invite your team into the project — all from the phone. It is the whole review loop, not a viewer.
Get Eddie on the App Store → · Eddie for iPhone →
Eddie won a Cine Gear Expo Technical Award
Cine Gear Expo is where the film and television industry sizes up the tools it actually uses — cameras, lenses, lighting, and the software that the footage passes through afterwards. The people walking that floor are working cinematographers, editors and post supervisors, not a general tech audience. Eddie took the 2026 Technical Award for Post-Production Technology — the only winner in that category. The full winners list is on the Cine Gear site.
We are glad about it, and we are aware of what it is worth. An award is a snapshot of one moment in front of one panel; the work that matters is the interview that cuts cleanly, the multicam that comes back from Premiere with its angles intact, the import that does not fall over at 2am. Most of what is in this very post exists because someone told us one of those things was not good enough yet. Thank you — genuinely — to everyone who has been cutting real work in Eddie and saying so plainly when it fell short.
Every microphone lands on its own track
Plenty of shoots record more than one microphone: a lav on the subject, a shotgun overhead, a board feed, the camera's own mic as a safety. Cameras and recorders that handle this properly write each mic as its own discrete channel in the file. Until this week, that separation did not survive the trip into your NLE — everything arrived pre-flattened, and the first thing you had to do was rebuild a mix you had already recorded correctly.
A Premiere project now opens with one timeline audio track per microphone. On a multicam, it goes further: every angle's microphone gets its own track, so a three-camera interview with a lav on each subject arrives as separate, labelled, workable stems. Solo the lav. Drop the on-camera mic. Ride one source without touching the others. The mix is yours to make, from the material as recorded.
The interesting part is what happens in the formats that can't express this. There is now a single shared rule for what counts as "a microphone," and every export format asks the same question of it — because two different pieces of code each deciding that separately is exactly how you end up with a different split coming out of Premiere than out of Final Cut for the same source file.
Final Cut Pro is the good example of why that matters. FCPXML declares how many audio sources an asset has, and Eddie used to hardcode that number to one for every file. For a camera writing two discrete mono microphone streams, that declaration was simply false — the file was described as "one source of two channels" when it was really "two sources of one channel each." Final Cut validates that declaration, so the lie had consequences. It now tells the truth.
And where a format genuinely cannot carry a multi-microphone layout, Eddie says so on the way out instead of quietly folding your mics together and letting you discover it later. Carry it, or disclose it — never flatten it in silence.
Mute a track, and it stays muted everywhere
Some cuts are not supposed to have camera audio under them. A narrated piece with voice-over over B-roll wants the narration and the music, and nothing else — but the clips' original audio kept riding along underneath, and silencing it meant remembering to do so separately in the player, in the render, and again in whatever you exported.
You can now mute a whole lane of an edit: the clips' own camera audio, the voice-over lane, or the music lane. Use the speaker toggles beside the timeline, or simply ask Eddie — "mute the original audio" works, and there's a proper tool behind it, so it also works when Eddie is driving from a prompt or from Claude.
What makes it reliable is that there is only one stored answer. The toggle you flick, the tool Eddie calls, the player, the cloud MP4 render, the local desktop render and every NLE export all read the same value. There is no separate export-time setting to forget. Mute it once, and it is muted everywhere it could possibly matter.
One deliberate detail: a lane you have never touched stays untouched, while a lane you explicitly unmute records that as a real choice rather than reverting to a default. The distinction sounds pedantic until a future default changes underneath an edit you already signed off on.
Captions that match your preview, word for word
Burned-in captions had a quiet divergence problem. The preview grouped words into cards one way, and the MP4 renderer grouped them another way, using its own duplicate copy of the logic. When the preview's grouping was made sentence-aware, the renderer never got the change. So you would review captions on screen, approve the phrasing, export, and get subtly different line breaks — a card ending mid-clause where the preview had broken cleanly.
The duplicate is gone. The renderer now uses the same implementation the preview does, and a test pins them together so they cannot drift apart again.
On top of the fix, captions got real controls. You can set how many words ride on a card — anywhere from one to twelve, with seven as the default. One or two words a card gives you the fast, punchy, one-beat-at-a-time look that reads well on a phone; ten or twelve gives you something closer to broadcast subtitling. There are new punch and editorial looks, and a set of bundled display faces that actually render in the weights they claim, rather than silently falling back to a substitute font when a heavy weight is missing. Every existing caption style keeps rendering exactly as it did.
Switch a camera angle right on the timeline
If a clip in your edit came from a multicam group with two or more video cameras, it now has a switch-angle menu in the arrange view. Open it, pick a different camera, and that clip is on the new angle.
Previously this meant going back to a prompt. Now it is a menu on the clip, and the choice is written the same way a prompted angle change is written — so it reaches the player, the rendered MP4, and every NLE export identically. Audio-only members of a group are never offered as an angle, because they are not one.
It closes the loop on the multicam work from a few weeks back: bring a synced timeline in with its angles intact, then re-cut which camera is on screen without leaving the timeline.
Working with multicam angles →
A busy stretch on the desktop app
The desktop app shipped almost daily this week.
Premiere Pro and DaVinci Resolve project files are downloadable straight from the Export menu, and you can export ProRes with an alpha channel and burned-in captions — useful when the edit is a layer in someone else's finish rather than the finished piece. Send to DaVinci Resolve now always lands a timeline; there is no longer a path where the media arrives but the sequence does not.
RED shooters got the most attention. RED (.R3D) footage imports on Windows when REDCINE-X PRO is installed — the detection that gates local RED decoding was previously macOS-only, which meant Windows users with a perfectly good REDCINE-X install were told their footage could not be read. Beyond that, RED decoding is now judged by the artifact it actually produced rather than by the decoder's exit code, and a near-complete export is salvaged instead of failing the whole import.
Two smaller frictions went away: clips of any length now import, with no minimum duration, and MP4 export stopped refusing to start over a low-memory reading that turned out to mean nothing. It warns and tries.
The desktop app → · RAW and RED footage →
Imports that don't give up
A stalled connection no longer takes an entire import down with it. Uploads ride out a transport interruption and carry on rather than treating a stall as a failure.
If an upload does break off, you have seven days to pick that same upload back up where it stopped, reconciled against what actually landed in storage rather than against what the app believed it had sent. And an import where only some files made it now keeps what arrived instead of discarding the batch — the files that landed are yours, and the ones that didn't are named.
Also this week
- Eddie tells you how much of your footage a run actually looked at, so a long edit's coverage is never a guess.
- MP4 exports open on a real frame instead of a black one — better thumbnails wherever you post or share them.
- B-roll shot in a different orientation to your timeline is fitted rather than stretched in exports.
- Resolve-flavoured Final Cut exports now import into Final Cut Pro cleanly, and audio-only sources relink properly.
Missed last week?
The headlines from the week before, in case this is the first one you've read.
- Dub your edit into another language — Eddie translates your edit's own words and speaks them in time with the cut, in 90+ languages, ducking the original dialogue underneath.
- Voice-over you can audition — have Eddie draft the script, listen to the voices before committing, and record another take without losing the one you had.
- One-click Generate on the timeline lanes — titles, captions and music each got a Generate button on their own lane, no prompt required.
- Footage costs half as much to bring in — importing and transcribing dropped from 500 to 250 credits per hour of source.
As always, just tell Eddie what you want and it'll do it.
— The Eddie AI team