Acollaborativelearningenvironmentforvideoeditingagents.|
Launch multiple AI editors on the same raw video, compare their cuts as versioned branches, score them with a verifier, and turn team approvals into reward signals for future edits.
Watch agents create, compare, and merge video cuts.
One raw video becomes multiple agent-generated branches. Aetos tracks what changed, scores each cut, and lets the team merge only the best creative decisions.
Main Cut
Agent sandbox output for the YC Startup Launch Video.
From raw footage to learning loop.
Aetos gives autonomous video agents the environment they need: agent sandboxes, versioned branches, verifier scores, selective merge controls, and memory from approved edits.
UPLOAD TIMELINE
Import raw video files or XML timelines directly from Premiere/FCP.
ARENA BRANCHING
VideoAgents generate parallel sandbox branches (Viral, Story, Brand Cuts).
VERIFIER JUDGES
Automatic verifiers score each cut against brand style parameters.
HUMAN APPROVAL
Editors compare differences side-by-side and merge edits to Main.
MEMORY RETENTION
Approved choices update preference vectors for the next agent run.
Different agents. Different strategies. One stronger final cut.
Aetos lets multiple autonomous editing agents compete and collaborate. Each agent explores a different creative direction, then the verifier and human team decide which decisions deserve to survive.
Run Agent ArenaViral Agent
Optimizes for hook strength, fast pacing, retention, and short-form energy.
Premium Agent
Optimizes for polish, brand safety, clean transitions, and premium tone.
Story Agent
Optimizes for narrative order, emotional clarity, and viewer understanding.
Brand Agent
Optimizes for client style, previous approvals, and brand guidelines.
Branch video timelines like code. Compare cuts like diffs.
Aetos tracks video editing as timeline metadata, not duplicate render files. Every agent attempt creates a lightweight branch, so teams can see exactly what changed, compare creative decisions, and merge only the best parts.
Version control gives Aetos the history of what was tried, what changed, what worked, and what failed.
Semantic Diffing
Aetos translates timeline changes into plain language: hook shortened, captions changed, CTA moved, music updated, clips reordered.
Selective Merging
Keep the Viral Agent's hook, the Premium Agent's color style, and the Story Agent's structure while rejecting weak changes.
Conflict Resolution
When two agents change the same timeline section differently, Aetos presents creative options instead of technical merge conflicts.
A verifier that turns video edits into rewards.
For agents to improve, every attempt needs feedback. Aetos scores each branch using a verifier, then combines that score with human approvals to create reward signals for future editing agents.
Aetos is RL-ready, not fake-RL. The MVP builds the environment where agents can try actions, receive rewards, and improve over time.
Viral Agent Cut
"The pacing feels slightly slow here. Let's swap the product demo and customer proof clips to keep momentum high."
Humans and agents work together without overwriting each other.
Editors, founders, designers, clients, and AI agents can all create suggestion branches, leave timeline-pinned feedback, compare edits, and merge approved changes into the final cut.
Timeline-pinned Feedback
Comments attach directly to clips, timestamps, and branches, so feedback never gets lost in Slack threads.
Suggestion Branches
Any teammate or agent can fork a cut, propose changes, and send it back as a reviewable branch.
Human Approval Loop
Approvals, rejections, and merges become the strongest signal for what the team actually wants.
Memory emerges from approved edits.
Aetos learns from approved branches, rejected cuts, merged changes, and team comments. Over time, it understands which hooks, captions, pacing, transitions, and styles your team actually prefers.
Approval-Based Memory
Every accepted merge increases confidence in the creative pattern behind it.
Rejection Signals
Rejected branches teach Aetos what not to repeat, like loud music, slow intros, cluttered captions, or off-brand transitions.
Recency-Aware Style
Recent campaign decisions matter more than old habits, so style memory can evolve over time.
Branch metadata, not giant video files.
Instead of duplicating heavy export drafts like final_v2_edit.mp4, Aetos creates lightweight metadata branches. Editors experiment freely without altering trunk timelines.
Every approved video makes the next one cheaper.
Video teams lose hours in revision loops, unclear feedback, and final-file confusion. Aetos turns every agent attempt and human approval into reusable learning data, reducing repeated work across future projects.
Aetos turns creative feedback into compounding production intelligence.
Fewer Revisions
Teams compare branches instead of rewatching five final files manually.
Faster Approvals
Clients review clear creative differences and approve specific changes.
More Output
The same team can ship more videos because agents explore directions in parallel.
Compounding Learning
Every approved edit improves the next agent run.