Eddie AI v4 Explained – Visual B-Roll Edits, Reference Edits, an Agent That Joins Your NLE Project

by Nino LeitnerToday

Subscribe to News - Software

Eddie AI v4 moves the prompt-based assistant editor past transcript-driven rough cuts and into visual work: B-roll sequences, montages, and edits cut to the beat of a music track. At IBC 2026, we spoke to co-founder and CEO Shamir Allibhai about what changed under the hood, why credits have replaced subscriptions, and an agent that joins your Blackmagic Cloud or Premiere Productions project as another editor.

We have followed Eddie AI closely this year, from the hands-on demo at NAB 2026 to the v4 feature list published ahead of the show. Shamir admitted he had not planned another full version number so soon after v3, which arrived right before NAB, but said that calling this release 3.5 would have undersold it. The basics are unchanged: you import up to 100 hours of material, and Eddie AI logs it, syncs clips, builds multicam groups, and separates interviews from visual shots. Hand it a teleprompter read or a fictional script and it collates the takes and cuts between camera angles. Everything after that happens in natural language, with a handful of manual editing controls for the moments when typing is slower than dragging.

From soundbites to visual edits

When version 1 launched in October 2024, Eddie AI did one thing, which was soundbite-based talking head videos. Version 4 handles visual and B-roll based edits, and Shamir listed event sizzles, montages, and wedding films as examples of what it now cuts. It can edit to a music track you supply or generate one, cut to the beats, apply a color grade pass, and send the result to your NLE or render an MP4. He described the output as something you can now take to a fine cut rather than leaving it at a rough cut, and put the overall jump at “10x more power and capabilities”, adding in the same breath that this might be an exaggeration.

Placing B-roll in the right spot has been the hard part for every tool in this category, so we asked what changed. Shamir was open about the fact that Eddie AI benefits from the pace of frontier model development; the app connects to ChatGPT and Claude, and the company adds its own core technology and fine-tuning on top. He pointed to the speaker identification system as an example of in-house work, which combines audio and visual cues to determine who is speaking when. Better visual understanding is what allows higher fidelity B-roll sequences.

Transcript-based editing, by contrast, is old news to him. His previous company, Simon Says, was founded in 2017 and did exactly that, and it happens to be the first automatic transcription tool I ever used.

Prompt on the right, reasoning in the middle

Shamir demonstrated the Mac app. The left side holds the navigation, split into edits and media, with the media already sorted into interviews and B-roll. The prompt box sits on the right, and the center pane shows the result. Typical instructions would be to build a sequence of every shot in which a named person appears, or to deliver a five-minute rough cut on a given topic.

While it works, Eddie AI displays its thinking, meaning the logic behind the choices it is making. Shamir’s argument is that an explanation builds trust and makes iteration practical: if you can see why the assistant went in a certain direction, you can tell it to change the narrative arc instead of starting over. In the demo, the finished edit showed the soundbite-based cut on V1 and the B-roll placed on V2, ready to skim through before exporting to an NLE, which now includes Avid Media Composer.

Backgrounder now accepts reference edits

The Backgrounder is where you give the assistant context. It already took faces of the people in your footage and a dictionary of terms specific to your company or industry. New in v4 is the option to bring in a reference edit, either as a video file or as an XML. Eddie AI reads the structure (open on a wide shot, move to a close-up, return to the wide) and applies that pattern to the new cut.

Where the footage goes

Eddie AI is a standalone app, but the analysis does not run locally. The app transcodes low-resolution proxies and uploads those for server-side analysis, while the full-resolution media stays where it is. Because projects are cloud synced, the same edit is available in the Mac app, the PC app, the web app, and the iPhone app, and you can drive Eddie AI from inside ChatGPT or Claude. Shamir mentioned story producers working on the web while editors use the desktop app, with footage coming from MAM drives and cloud drives.

On the security side, he said Eddie AI is SOC 2 Type II compliant and does not train on customer data. The user base has shifted accordingly. It started with independent freelancers and small production companies and now extends to large video teams inside enterprises.

Agents you can tag, and an editor that joins your project

Two of the newest features go beyond the prompt box. The first is expert agents that you tag in an edit. Shamir’s example was a short-form video: tag the MrBeastMethod agent, get feedback on the timeline, then have Eddie AI implement the notes. According to the company’s agent page, these agents are trained on a filmmaker’s public talks and on writing about their work, the filmmakers have not necessarily endorsed them, and an agent that cuts always creates a new edit instead of overwriting yours.

The second reverses the direction of travel. Until now you brought footage to Eddie AI; now it can join a Blackmagic Cloud project, a Premiere Pro Productions project, or another collaborative team project as an additional editor, working in real time next to the humans. Shamir described it as agentic AI that operates the graphical interface of the NLE itself, moving the cursor and clicking.

AI as a tool, and the jobs nobody wanted

We also asked about the resistance to AI that is still loud in many parts of the industry. Shamir’s position is that speech recognition counted as AI not long ago, and that these are tools to use when they help and ignore when they do not. He compared refusing them on principle to refusing VFX, while acknowledging that software touching an editor’s professional identity is a sensitive subject. The audience, in his view, only asks whether the result was entertaining and informative.

His case rests on the work nobody wants. No child dreams of becoming a logging expert, as he put it. I can relate: for my diploma film 20 years ago, I spent an entire summer transcribing 50 hours of interviews by hand just to build a paper cut. Shamir describes post-production as wearing ten hats, of which you hate three, and sees Eddie AI’s job as taking those three.

Price and availability

Eddie AI v4 is available now, and subscriptions are gone. The app runs on credits that you buy when you need them, which Shamir said suits people who have a project one month and nothing for the next two. Credits are deducted at two points. Import costs a maximum of $2.50 per source hour, and less if you buy larger credit packs. Prompts are then charged by the work involved, so logging interview footage is cheap and generating B-roll costs more.

According to Eddie AI’s pricing page, packs start at $10 for 1,000 credits, new accounts receive 1,000 free credits, and credits do not expire. Existing subscribers have been moved over with a considerably better credit conversion as a thank-you. For more information, please visit the Eddie AI website.

Would you let an AI agent join your shared Resolve or Premiere project as another editor? Don’t hesitate to let us know in the comments below!