Reelful turns raw photos and video plus a short direction into a ready-to-share edited clip.
2
Editing real footage is harder than generating from a blank canvas because the agent must choose what to keep, remove, and arrange from messy material.
3
Reelful uses media understanding, a user-approved creative plan, sandboxed agent work, Remotion compositions, and verification before hiding the workflow behind mobile templates.
Summary
Ekaterina Deyneka presents Reelful as an agentic video editor for people who record content but rarely publish it. A user adds media and a direction, then the agent identifies useful moments, removes pauses and failed takes, assembles the material, and adds elements such as captions, music, voiceover, and b-roll. Deyneka compares the infrastructure to an agentic app builder: a prompt starts work in a sandbox where an agent uses tools and skills, but the output is a rendered video. The difficult part is editing real footage. The agent must make creative choices from incomplete and untidy source material. Reelful first understands the media and transcribes speech, then gives the user a creative plan to approve. The agent works with editing skills and creates a Remotion composition as React code. A verification layer catches composition problems and sends the agent back to revise them. Mobile templates and a built-in editor hide this complexity from consumers while allowing small manual corrections.
Reelful targets the gap between recording video and publishing it
Ekaterina Deyneka opens by asking who recorded photos or video at the conference and who posted any of it online. Many hands go up for recording, while far fewer remain up for posting. She says she had the same problem at conferences, events, trips, and meetups because editing video is tedious, largely manual, and feels like an art rather than an automated task. Reelful is her attempt to let people turn the footage they already have into something publishable without doing all the editing themselves.
A user can give Reelful media and a direction, then receive a finished clip
The workflow starts with photos and videos plus context or instructions. The user might explain what happened in the files, or request captions, music, or voiceover. The agent understands the media, finds suitable moments, assembles them, and generates additions such as captions, music, voiceover, and b-roll. Deyneka also describes a speak-to-camera case, where the agent removes pauses and unsuccessful shots. The intended result is a ready-to-share clip produced without the user manually sorting every take.
The agentic video editor has the same infrastructure shape as an app builder
Deyneka compares Reelful's infrastructure with an agentic app builder. Both begin with a prompt interface, although the video editor takes media plus the prompt. A remote machine, or sandbox, spins up on the back end. Inside it, an agent works with tools and skills. The app builder changes a codebase and returns an app preview. The video editor changes a video composition and returns a rendered video. The execution environment is similar even though the artifact is different.
Editing real footage forces the agent to make choices that generation can avoid
Reelful focuses on editing personal footage instead of generating most of the content. Deyneka says a blank canvas gives an agent freedom to create whatever it wants. Real footage imposes decisions: which moments are best, what should be omitted, and how the remaining pieces should be organized. Source material can also be messy or incomplete. The system still has to produce a polished result that viewers ideally cannot distinguish as AI-edited or human-edited.
Reelful asks the user to approve a creative plan before editing starts
The pipeline begins with media understanding. Reelful works out what is happening in the clips and photos, and transcribes speech when the footage includes a speak-to-camera video. It then creates a creative plan for the user to review. The user can approve it, request changes, or regenerate it before the actual editing begins. After approval, Reelful starts a sandbox where the agent executes the plan rather than immediately making an opaque set of edits.
Skills encode editing taste while the agent handles the composition
Deyneka describes video-editing skills such as rules for selecting good moments, font pairings, and methods for generating b-roll. She says this is where Reelful's taste and craft live. The agent can also start other processes for music that fits the composition, voiceover, sound effects, and image animation. These skills give the agent guidance about what a finished edit should feel like while leaving the execution inside the same agent workflow.
Remotion makes the video an artifact agents can write as code
Reelful uses Remotion, an open-source framework for creating videos as React code. Deyneka describes the composition as a file that specifies the assets, tracks, and their order. This fits the system because agents are good at writing code. After the agent creates the composition, a verification layer checks whether it is clean, well defined, and capable of rendering. If it finds a problem, the agent revises the composition before Reelful produces the polished clip.
Mobile templates hide the agentic workflow while keeping small edits possible
Deyneka says consumers should not need to understand the complex workflow behind the product. Reelful takes a mobile-first approach and provides directional templates for tasks such as speak-to-camera videos, b-roll, and voiceover. A user can select a template, add media, and avoid writing a prompt. After the agent generates a video, a built-in editor lets the user make small changes, such as removing a second or correcting a caption word. The product combines automated generation with a familiar editing surface.
"If the agent has a blank canvas, it can do whatever it can. But in the editing case, the agent has to figure out which moments are the best, what to omit, what to use, how to organize everything together."04:38
Who should watch
You are building an agent that must transform messy user-owned media into a finished artifact rather than generate from scratch.
You want to understand how sandboxed agents, domain skills, code-based composition, and verification can fit into a creative workflow.
You are designing a consumer AI product and need to hide prompting and orchestration behind templates while still allowing lightweight manual corrections.
