AI Engineer World's Fair 2026
Building an Agentic Video Editor for Mass Consumer
Read the talk
Building an Agentic Video Editor for the Camera Roll
Reelful turns personal footage into shareable clips through media understanding, an approved creative plan, code-based composition, and a verification loop.
From a talk by Ekaterina Deyneka
Before you start: Basic familiarity with React helps with the composition example; no video-editing experience is required.
From media and directions to a finished clip
The starting interface is simple: drop in photos and videos, then provide context about what happened or directions for the edit. A request might ask for captions, music, or a voiceover. The agent must understand the media, find the right moments, assemble them, and produce the requested supporting material—including B-roll—before returning a ready-to-share clip.
A speaking-to-camera recording makes the editing task especially concrete. The source may contain long pauses and unsuccessful takes. The desired output keeps the useful speech and removes those interruptions, without requiring the user to locate every cut. Deyneka follows this example with an initial edited-video preview: the product’s output is a finished clip, rather than instructions for making one.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
An app-builder architecture with a different artifact
The infrastructure resembles an agentic app builder. Both begin with a user request and start a remote machine—a sandbox—where an agent works with tools and skills. In the video editor, the request also includes media. Deyneka corrects a duplicated heading on the comparison slide: its second column should read “agentic video editor.”
| Component | Agentic app builder | Agentic video editor |
|---|---|---|
| Input | Prompt | Media plus prompt |
| Execution | Agent, tools, and skills in a sandbox | Agent, tools, and skills in a sandbox |
| Working artifact | Codebase | Video composition |
| User-facing result | App preview | Rendered video |
The useful architectural connection is the working artifact: an agent modifies a structured project inside a remote environment, and the system turns that project into something the user can inspect.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
Real footage constrains the creative task
The shared infrastructure does not make the creative problems identical. Reelful focuses on editing personal, real-life footage and generates relatively little of the content. Deyneka considers this harder than working from a blank canvas: the agent has to choose the best available moments, decide what to omit, and organize what remains into a coherent sequence.
The footage may also be messy or incomplete. The quality target remains a polished, professionally edited result—ideally one whose viewers cannot tell whether a person or an AI edited it. That is the aspiration guiding selection and assembly, not a measured result presented in the talk.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
Understand the footage, then approve the plan
The workflow separates deciding what to make from executing the edit:
- Understand the media. Identify what happens in the supplied clips and photos. Transcribe speech when the input includes speaking-to-camera footage.
- Propose a creative plan. Let the user approve it, request changes, or regenerate it before editing begins.
- Start the execution environment. Once the user approves the plan, spin up the sandbox in which the agent will work.
The approval step gives the user a chance to correct the direction before the agent begins assembling the composition.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
Where editing judgment enters the workflow
Inside the sandbox, the agent brings skills that encode editing judgment. Deyneka’s examples locate taste and craft in specific decisions:
- Cut rules: how to select the best moments.
- Font pairs: which combinations suit a particular use case, and which do not.
- B-roll instructions: how to generate supplementary footage for the edit.
These skills guide how the available material becomes a composition.
The agent can also initiate subprocesses to generate music that fits the exact composition, create a voiceover, add sounds, or animate images. Supplied photos can therefore become moving elements rather than remaining static throughout the video. These assets feed into the next stage: the Remotion composition.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
Represent the edit as code, then verify it
Remotion makes video composition programmable in React. Deyneka describes the representation in terms of assets, tracks, and their order. More precisely, a composition combines a React component with video metadata such as frame rate, dimensions, and duration. Although she calls the framework open source, its published source currently uses a special license with eligibility and commercial-use conditions.
A code-based composition turns editing decisions into a task a coding agent can perform. For example, a small React composition can express two supplied photos appearing in sequence. Here, from and durationInFrames define the timeline, while the composition supplies the output metadata:
tsx
import {
AbsoluteFill,
Composition,
Img,
Sequence,
registerRoot,
staticFile,
} from 'remotion';
const PhotoEdit = () => (
<AbsoluteFill style={{backgroundColor: 'black'}}>
<Sequence from={0} durationInFrames={60}>
<Img
src={staticFile('arrival.jpg')}
style={{width: '100%', height: '100%', objectFit: 'cover'}}
/>
</Sequence>
<Sequence from={60} durationInFrames={90}>
<Img
src={staticFile('gathering.jpg')}
style={{width: '100%', height: '100%', objectFit: 'cover'}}
/>
</Sequence>
</AbsoluteFill>
);
const Root = () => (
<Composition
id="PhotoEdit"
component={PhotoEdit}
durationInFrames={150}
fps={30}
width={1080}
height={1920}
/>
);
registerRoot(Root);
The filenames and timing illustrate the representation; Reelful’s agent must derive the actual asset choices and arrangement from the media and approved plan. Deyneka’s reason for choosing this approach is that agents are already good at writing code.
Code generation is followed by a verification layer. It checks that the composition is clean, well-defined, and renderable. If it finds problems, the agent iterates on the composition before producing the polished clip. The described loop checks the generated artifact rather than assuming the first composition will work; the talk does not specify the individual checks or their implementation.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
Hide the workflow without removing control
Media understanding, planning, sandbox execution, asset generation, composition, and verification form a complex workflow. The consumer should not need to understand all of it. Reelful’s first interface choice is mobile-first creation, intended to fit into everyday activity. Deyneka mentions walking and lifting weights, and also names driving as a usage scenario; interaction with an editor should wait until safely parked.
The next two choices address different sources of friction:
- Directional templates reduce prompting work. Users can select a speaking-to-camera, B-roll, or voiceover direction and drop in their media. Deyneka says this can work without an explicit prompt.
- A built-in editor preserves familiar control. After agentic generation, users can make small manual changes, such as removing a second or correcting a word in the captions.
The agent produces the initial video, while the familiar editor handles precise adjustments. Deyneka then turns to videos she recently created with Reelful.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
An event recap, then a return to recording
The played example begins with an invitation to an exclusive Cursor Creators Dinner. Playback pauses briefly while Deyneka checks whether the audience can hear it. The narration then describes a custom event space transformed into a tropical sunset setting, establishing the venue before moving through the experience.
Next come the food and atmosphere: private chefs from Los Angeles, lobster, wagyu, crab, and strawberry ice cream. The recap moves on to conversations about people’s work and the ease of connecting with the community, then reaches gifts before playback cuts off. It is a concrete example of the desired output: an event recording assembled into a narrated social recap.
Deyneka says these videos were assembled using only the agent, without a regular video editor, and that she was already posting them on social media. That connects the demonstration to the opening problem: recordings that previously remained unshared had become material she was publishing.
At the conference, she offers access to a beta of Reelful’s second version. The product is still early and under active development, and she invites feedback, contact by email, and accounts of how people actually use it. She also announces recent funding from a16z Speedrun.
Then she asks to film a video with the audience. The talk ends where it began—with someone capturing an event, now with a workflow for turning that recording into something shareable.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
Resources
From the talk
iPhone video editor that turns supplied media and directions into a planned, editable video.
Source code for creating and rendering videos with React, with links to documentation and licensing.
Further reading
Explains frames, video metadata, and React compositions through small code examples.
Updates since the talk
Current project setup instructions, including command-line scaffolding and coding-agent workflows.
Read the complete timestamped transcript
- 0:00
[upbeat music] Hi, everyone. I think we can start.
- 0:15
Uh, but before we start, I want to ask you a couple of questions. So first of all, how many of you took a photo or video during this conference?
- 0:24
Please raise your hands. Okay. And how many of you actually posted any video content from it online?
- 0:37
Not that many. And, um, to be honest, that was me. I, I was recording a lot of content during conferences, events, trips, meetups, uh, and I never posted them online because video editing is hard.
- 0:53
Uh, it sounds and it is a lot of work, uh, it's tedious, and it's largely still manual. So... And, and it also feels like an art and not really automated.
- 1:07
And that's why we're building Reelful, and we're trying to tackle agentic video editing problem from, uh, video editing problem from the agentic standpoint. I'm Kate. I'm founder and CEO at Reelful.
- 1:23
But let's first talk about what agentic video editing is. So as a user, you just drop in your media photos and videos and provide some context. It can be, uh, the context what happened in these media files, or it can be some directions.
- 1:40
For example, like add captions, add music, add voiceover, and something like that. And then the agent will go understand your media, uh, find the right moments, assemble everything together, generate captions, music, voiceover, B-rolls, uh, and give you a ready-to-share clip.
- 2:02
Uh, and, um, yeah, so basically, um, that's, uh, video... agentic video editing. So agent does everything by itself. Or another example, you recorded, uh, a speak to camera video, and ha-- you have a lot of pauses, unsuccessful shots, and you expect an agent to figure it out, to remove unsuccessful shots, remove
- 2:27
pauses, and give you a ready-to-share clip. Um, and, uh, the, the interesting thing is that a lot of... actually, a lot of the things inside this pipeline can be automated, and this is exactly what we're doing at Reelful.
- 2:45
Oh, this is, um, the example of, uh, video edited.
- 2:52
And, um, since we are at AI engineering conference, I wanted to talk a little bit about infrastructure. And from the infrastructure standpoint, agentic, uh, video editor is very similar to agentic app builder.
- 3:08
Uh, sorry, there is a typo on the slide. So the, uh, second column is agentic video editor. So both of them have a prompt, uh, a UI for prompt, uh, and in the video editor case, it's a media pr-- plus prompt, and usually on back end, what's happening?
- 3:26
There is a remote machine which is called Sandbox, uh, which is spinning up, and inside this machine, there is an agent with tools and skills which is working on, uh, what you're, uh, you're, you're asking it to do.
- 3:40
Uh, in the case of the agentic, uh, app builder, it's a code base. Uh, in the case of the agentic video editor, it's a video, video composition. And as a result, in the a-agentic app builder user get an app preview, and for the video editor users, uh, user get rendered, a rendered video.
- 4:03
And but yes, the, the infrastru-- from infrastructural standpoint, it's pretty similar, but there are a couple of differences. Uh, and, um, this is actually the most interesting to me, generating versus editing.
- 4:17
At Reelful, we are focusing on editing real footage, so we do not generate a lot of content. We are expecting you to provide your real life, your, uh, personal content, and we will edit it for you.
- 4:33
And actually, this is, uh, a more complex problem because if the agent has a blank, blank canvas, it can do whatever they can. But in the editing case, the agent has to figure out, uh, which moments are the best, uh, what to omit, what to use, how to organize everything together.
- 4:55
And also, um, sometimes, uh, fo-footage can be, um, messy or incomplete, and agent still has to deliver a very polished result, professionally made, so that ideally, the viewers of this content don't get if it is, like, AI or, uh, human edited.
- 5:17
So let's actually have a look how we do it, uh, at Reelful. So we start, as I already mentioned, with your media plus a prompt, some directions like how you want it to be edited, and we need to get a polished clip.
- 5:35
So let's go through it step by step. So we are doing first me-media understanding. We need to understand what is, what's actually happening, uh, on those clips and photos, and we also need to transcri-transcribe speech, for example, in the case if you have speak to camera videos.
- 5:54
Then we are providing a creative plan for the user so that they can approve if they like it or not, what they want to change or maybe regenerate, uh, and we create this plan before actually starting editing.
- 6:10
Once the user approve this plan, uh, we spin up a sandbox, the remote, uh, remote machine that we already discussed, and this is an environment for the agent to, uh, execute everything.
- 6:24
So the agent comes with the skills, and in our case, in, in the case of, uh, video editing, our skills are, for example, cut rules, how, for example, how to select the best moments.
- 6:37
Uh, also font pairs, which fonts are, uh, more suitable for this use case, which are not. For example, how to generate B-rolls, and this is where taste and craft, uh, live, actually.
- 6:50
Um, and then also agent, uh, can, um, can initiate some other subprocesses. For example, generating music that will fit this exact composition, generating voiceover, adding sounds, animating images.
- 7:05
Yes, this is actually what we do. Uh, if you provide photos, we can animate your photos to make them more, uh, dynamic and engaging. And then comes Remotion composition.
- 7:18
So here, a little bit of background. What's Remotion? Remotion is a framework, open source, open source framework, uh, to create videos as code, as React code. Uh, so basically it's just like a, a file with, uh, the order with all your assets, uh, and tracks and how they're following each other.
- 7:39
And, um, why it is important, because, uh, agents are really good at writing code, and therefore we can use them to create videos with this Remotion framework. And then, uh, the last thing is the verification layer.
- 7:55
Of course, agent can make mistakes, and that's why we developed this verification layer to make sure that all the, um, the composition is clean, is well-defined, everything will be rendered, and if there, there are some problems, then the agent will, uh, reiterate on the composition.
- 8:15
And this is how we got to a polished clip. [sighs]
- 8:19
So it's a lot, right? Uh, it's like a very complex workflow. Uh, and ideally, we don't want our users to even know anything about it. And this is even maybe a bigger problem, how to deliver this complex agentic workflow to mass consumer.
- 8:40
And this is how we're tackling that at Reelful. So we decided to go mobile first so that users can edit videos, videos while driving, walking, or maybe lifting weights.
- 8:53
Uh, also, I know that prompting videos can sometimes be also challenging. That's why we create directional templates. For example, like speak to camera videos, or maybe you want to add B-rolls or voiceover so that users can just select these directional templates, drop their media, and that's it.
- 9:15
Even without any prompt, it will, it will work. And the third thing is a build-in editor. Uh, why? Because we want to make this experience convenient and familiar for users.
- 9:28
So a lot of people are already sort of, uh, using, uh, regular video editors, and that's why we want to provide this experience as well. So how it works, user first generates a video agentically, but if they want to tweak it, for example, remove a second or maybe, uh, correct some word in the captions, they can go
- 9:51
into build-in editor and edit it a little bit. Um, yeah, and, uh, actually, I have a couple of examples here that I recently created with Reelful. Uh, I will play them, just maybe one of them.
- 10:06
Oh, sorry.
- 10:08
Last week, I was invited-
- 10:09
Do, do you hear? Barely. A little bit. Okay. Y-you just can [laughs] enjoy the video.
- 10:15
Exclusive Cursor Creators Dinner, and honestly, it was one of the best event experiences I've had. First of all, the venue was stunning. It was a custom event space transformed into this tropical sunset feast.
- 10:26
The whole atmosphere felt so warm and cinematic. Second, the food was way beyond my expectations. They brought in private chefs from LA, and we had lobster, wagyu, crab, and this incredible strawberry ice cream that I was still thinking about.
- 10:39
But most importantly, the conversations were so much fun. It was kind of atmosphere it felt really easy to connect with people, talk about what they're working on, and just enjoy the community.
- 10:49
And lastly, we got gifts. One of the highlights-
- 10:52
So, uh, yeah. Basically, all these videos, they were assembled only using agent, no, uh, regular video editor, and I already posting them on social media. And yeah, it-- I, I have a lot of fun with that.
- 11:08
And-
- 11:09
Last week, I was invited-
- 11:09
Oh, sorry. Uh, and exclusively for this conference, we are, uh, giving our beta, uh, which is our new f- new second version. Uh, please give it a try, uh, and, um, let me know if you have any feedback.
- 11:28
Uh, here is my, uh, email. Please, uh, feel free to reach out. Uh, we're still early. We're actively working on it, uh, so we will have-- uh, we will be happy to hear any feedback, and also curious how you use it.
- 11:44
And also, um, some exciting news. We recently got funded by a16z Speedrun, uh, so I'm very excited to, uh, continue working. [audience applauding]
- 11:57
Um, yeah, that's it. Thank you so much. And because it's a presentation about content and how to edit videos, I have to, uh, film a video with you all. [laughs]
- 12:12
Yeah. It's so good. Yay. [upbeat music]