How to Use the Editor Agent 🎬

Let me show you how the editor agent works.

So it lives over here in the sidebar.

And when you open the sidebar up you probably see your transcript first, which you can edit and you should be pretty familiar with by now.

But down the bottom you can see there's actually an input to talk to an agent.

You can see I've got Sonnet 5.5 enabled here.

and you can chat to the agent with your transcript as the sort of primary context.

But if you click on the chat icon then you can switch to a full on chat mode where you can have long running conversations with your agent about editing your video.

So let's try it out.

This gives you a brief summary of the types of things that you can do with the agent.

But the general thing is that everything that you can do in the UI the agent can do.

And if you've taken a look at our MCP or API docs, everything that you can see in there the agent can do.

And we are expanding the scope of this beyond just things that you can do in the UI every day.

what it's best at are things that would take you a long time to do.

So there are a lot of changes in video editing that are still just kind of a click of a ui, but anything that requires a bunch of work, then that's where the agent is really great.

So I have something that I need the agent to fix for me which is this is a long video, it's 20 minutes long and there's loads of zooms in it and there's a bunch of these zooms that are too zoomed in.

Not all of them, some of them, or quite a few of them.

And if I want to go through and fix all the zooms, that would take me a long time to update all of the zooms I have an issue with.

So I'm going to get the agent to do this for me.

So change all zooms that currently 2x to 1.5x And then these zooms like this which are too zoomed in, I'm going to get a bunch of them across the whole video.

they should be less zoomed in at 1.5x.

So we can just let this run for a moment.

You'll see it's thinking and the things that it's doing, it can see that there's a lot of zooms that it's got to fix and so it's going to do a batch update there.

and then you would keep an eye on the individual tasks that it's doing here.

So the agent has a bunch of tools like it can look at the timeline of your video, it can make screenshots and visually check what's going on in the video.

and then it has access obviously to all of the features that you would use normally yourself.

The other cool thing is that the changes happen in real time.

So this is now 1.5x and you can see it's already slightly less zoomed in.

So every change that the editor makes, it's like working with someone in a real time collaborative tool.

It'll just make those changes as you ask it to and then you can go and work on something else.

So you could say, go and clean up clip one and then you could go and work on clip six or do something else.

You can also just let it go and work on something and then just leave.

because all of this happens in the cloud, it doesn't happen on your computer.

so you can set it to edit a big long video.

That's going to take a bit of time and you just come back later and see how it's done.

So you can see that this found 44 zooms that it needed to update, and switched them from 2x to 1.5x.

And it told us that there are a bunch that were already at 1.5x.

So it hasn't changed those.

And the neat thing is you get a lot of timestamps and links to clips, to show you the work that it's done.

So it's given me this link to clip five and I can just jump there and I can see.

yeah, there's already well all of these are 1.5x, but there would have been some in here before that were at 2.5x.

so that's really cool.

So another area where the agent really shines is with broader high level tasks.

sure, updating a bunch of zooms saves us a lot of work, but it's still quite a specific task, you know, changing from one zoom level to another here.

When I started editing this video, I just asked it to clean up the video and kind of get it ready for YouTube and make it more engaging.

not super specific.

I just wanted to give it this sort of job of doing a first rough cut, and you can see what it's done.

it has cut the video down from 30 minutes to 20 minutes.

So it's cut out a bunch of time for us.

it went and added, it removed a bunch of mistakes throughout that it went and added a bunch of cutaways to just shots of my camera like this, which is an edit that I really like.

And you can sort of see visually at a glance where it made these changes and we can seek to these places and see what's gone on there.

You can also see that it made some significant cuts in these areas and you get this sort of rich summary of all the work that it's done.

It also added chapters for YouTube, it enabled Studio sound, and then you get another snapshot here at the end of it.

this job would have taken me, I don't know how long, but it took 15 minutes to do a really solid rough cut cleanup of a 30 minute video.

And that gave me a great platform to move on with more specific edits.

So let me jump to another part of this transcript to show you some of those other things.

So now I'm going to play you the intro of the video.

to see what the agent did for me in making this intro.

It does this staggered text effect of all the things that are happening, all the topics that I cover in this video, they are all laid out at these different points.

And you can see there is a lot of text layers, and there's a bunch of sound effects.

This is a super complicated edit.

and I didn't do it, the agent did.

So you can see over here I just said add a staggered text effect to the intro clip, listing all the features we cover in this video.

I don't even need to tell it what we cover in the video.

I just said you go and add all the relevant info there.

I told it which text preset to use.

and I mean it would pick one if I didn't.

and then it just went and did all that.

First I just told it to do the staggered text effect, and then worked with it to tweak some stuff.

and then later, I said go and add a pop, sound effect to all of them.

And you can see it did that as a follow up then this would have been super annoying to do manually.

Like have to zoom way in and then just line all of these up with all of the text layers.

so again this is a combination of both broader creative task and kind of an annoying time consuming one.

And it can just do it really well and really fast.

Here's another cool example.

So when I did the first rough cut, I said go and add a bunch of camera, only cutaways to kind of break things up and make it more engaging.

But I wasn't really happy with the state of clip two.

This was a long, seven, nearly eight minute clip, with just me talking, doing a demo, and I wanted more of these cutaways to, to keep the viewer engaged.

And so I just said go back to clip two, and do a bunch more of those cutaways, and so it went and did another pass, added a bunch more.

And when it does these big sort of batch edits, it can give you really nice summaries like this, where it tells you where it made the edits, and you can click and jump to that point and it tells you how long that edit was and it gives you again a nice summary.

So the whole idea is that you can work with, with your agent, and it can give you feedback about the work that it's done.

That still saves you from having to go and watch the entire video, and review it all, which then sort of defeats the purpose of it.

What this should feel like is an editor that you really trust where you give it a brief and tell it what to do and then come back and say here's what I've done and you can sort of review what it's done and see it sort of highlights.

But you don't actually have to go and become an editor again yourself to watch the entire thing and, and proof everything.

You should still eventually review the full video, but you don't need to be doing that for every single thing that your editor or your agent editor, does.

Here's another fun example again in the intro.

So if I play the intro back after the sort of text effect, sequence, There's this cool effect with a bunch of media layouts.

So media layouts are where you can incorporate media, from somewhere else into your video.

And you can see that it's added four of them here.

And the media that's used in each of these is actually, my product demos happening throughout the rest of the video.

And so what I said was make Clipbond better.

Go and grab some short clips from the rest of the video, highlighting the most interesting stuff, and make that into a montage, for the intro.

So it figured out the right place to put this.

I didn't tell it to put it here, but this is the perfect sort of place, where I say one big video and it stitches together, the highlights of the one big video.

And it sort of went and found nice examples of all the features.

Initially it grabbed too many.

It grabbed all eight of the features that I'm demoing.

And that just took too long and felt too busy.

So I told it to shorten it, and tweak it a little bit.

But the core work of going and finding these clips and integrating them in here, which would still take ages, It did a really great job of.

And again, it did pretty, fast.

I'll bring this home in a minute, but I want to give you sort of a glimpse into the future of where this is going.

So if you look at these media, layouts, the clips in these look kind of funky.

There's a fun scan line effect and there's some chromatic abrasion, on here.

Now that's not how these clips look in Intella.

Wherever this one was from.

I think clip two.

That's not how clip two looks.

I just said make these look more interesting and add some scan lines and some vignette.

And a bit later.

yeah, here you go.

I added some chromatic abrasion there.

And it just went and updated the clips that it made and put these effects over the top.

and soon you'll be able to do much more of this, slightly more creative use, some more of these creative effects via the agent to just create much, more imaginative things.

But the agent can already do a bunch of stuff like this and you should play around with it and try it out.

So yeah, that gives you a small sample of things you can do with the agent.

It's still very early and we're trying to add more and more capabilities to the agent and more creative options that the agent can do and use.

you can switch models down here so you can change models, you can change thinking level at this early stage based on our tests, Sonnet 5.5 makes a good cost effective default.

It's very fast and it's very good at the quite specific tasks.

So when I did this one earlier where I said go and update all these zooms, it's good at stuff like that.

If you want to do a broader more high level creative edit then switching to something like Opus 5.5 You'll get a better result with that.

But that is a more expensive model and it thinks a bit longer so it can take a longer time.

Generally with video stuff we sort of find that at this stage and the type of capability that types of things you can do with Tella you don't need super high thinking levels.

medium on 5.5 and sonnet is pretty good and that means it's faster.

If you're really doing something ambitious use 5.5 and then turn it up to high.

I haven't really felt the need to crank it all the way up yet so you can use the models that we have made available in here.

We will also be rolling out the option to connect your own ChatGPT subscriptions so that you can use your own tokens basically.

we don't have this available just yet I don't think but it's coming soon.

And the other thing is that this is all just using our mcp.

So the things that the agent is doing is using all of the capabilities that the MCP has.

It's not something different.

So you can actually just connect whatever agent or harness that you use day to day and you can just can do all the things I'm doing here but just externally outside of Tella and then you can just preview the video in Tella.

The nice thing with Tella is you get a very fast iterative experience where if you're in the editor you can see these changes coming in and you can work you can get it to a bit of a flow state with it which is cool.

Some other stuff, I showed you the transcript view you will see the last thing that the agent said down the bottom here.

And then you can switch back between the transcript view and the agent view.

you can pull in context from the transcript.

So you can just select stuff in the transcript and you can say add to chat.

And then you could say like cut this or change the layout here or add the zoom here.

so you can use that as context.

we working on a lot of Still figuring out a lot of the right UI around this, how it integrates with other parts of the ui.

That's all still work in progress.

you can keep track of your tokens here And you can buy more credits, here.

And here you've got options to attach images and files, videos and playlists.

Skills.

We're still working on memories.

A lot of the stuff is still very, very early but it gives you a sense of some of the core functionality.

have fun.

let us know what you think and how we can make it better.

This really feels like the future of where video creation is going