SolCut

Raw footage in. Finished cuts out—with an AI director.

Comment

SolCut — Direct the Cut with AI

Animated SolCut editor preview

Raw footage in. Finished cuts out—with an AI director.

SolCut removes the editing-software bottleneck between having footage and publishing a video. Creators bring their media, describe the story they want, and remain in control while specialised agents operate an auditable timeline.

Inspiration

I kept encountering the same problem while building projects: I could build the product, record the screen, and gather footage—but creating a clear demo video still became its own difficult project. Timelines, keyframes, captions, audio balance, reframing, and export presets are each sensible tools; together, they turn “I have the footage” into a specialised job.

I wanted the editor I needed myself: a tool for indie builders, students, hackers, founders, and creators who can direct a story but should not need a professional editing career just to ship it.

Why this matters now

Signal What it means
3M+ creators in the YouTube Partner Program The creator audience is already enormous. YouTube, March 2026
$37B projected U.S. creator-economy ad spend in 2025 The category was projected to grow 26% year over year—about 4× faster than the media industry overall. IAB
50M global creators; 10–20% CAGR Goldman Sachs Research’s creator-population outlook shows the workflow problem is expanding, not shrinking. Goldman Sachs
19% of non-video marketers cite lack of time The bottleneck is not only creativity; it is the work required to get from raw media to a publishable story. Wyzowl

More people can make footage than ever. Finishing it—quickly, well, and in the right format—remains the skill-heavy step.

What it does

SolCut is a local-first AI video director. A creator uploads clips or imports an authorised source link, keeps the original media available for inspection, then directs the edit in plain English:

  • “Make a 30-second highlight.”
  • “Turn this into a vertical reel.”
  • “Add animated captions.”
  • “Tighten the opening.”

The result is not a black-box template. It is a working editor with source insights, a responsive preview, linked video/audio/caption tracks, caption editing, timeline actions and undo, aspect/layout controls, and local FFmpeg delivery to MP4, SRT, ASS, and an edit manifest.

From footage to final cut

solcut-workflow

Agent / system Responsibility Guardrail
Ingest Agent Retains the source, creates an edit-safe proxy, records media facts The original is never overwritten by a highlight or reel.
Visual Agent Finds scene and visual-beat evidence Cuts are based on observable source moments.
Transcript Agent Produces timestamped, speaker-aware speech evidence; accepts hotwords/context Captions retain source timecodes and do not invent word timing.
Reframe Agent Uses verified subject-focus paths for Auto Reframe No silent fallback to an untracked centre crop.
Editor Agent Applies source-grounded timeline operations Video, original audio, and captions stay linked through edit actions.
FFmpeg delivery Renders the creator-approved timeline locally Outputs remain traceable to the original source ranges.

solcut-architecture

The key product decision is that agents operate the timeline; the creator directs the story. Every edit can be reviewed, undone, or reset back to the retained original.

How we built it

I started with a PRD and worked milestone by milestone. The first two Build Week days were spent on product ideation, deciding which features to include or defer, and crafting the landing-page prototype with GSAP and Three.js motion. I built that initial prototype on my laptop, transferred it into this workspace, then restructured it into a modular React/Vite editor backed by a local FastAPI media pipeline.

Layer Implementation
Editor React + Vite workspace for media, preview, captions, layouts, timeline, and AI Director controls
Orchestration FastAPI project state, validated operations, source evidence, and recovery-safe jobs
Media Retained originals, edit-safe proxies, timeline-linked tracks, and local FFmpeg rendering
Evidence Visual-beat analysis, timestamped transcription, recognition context/hotwords, and verified focus paths
Quality Python unit suite plus Playwright browser coverage for actual user workflows

I used Codex as an implementation collaborator: it accelerated the FastAPI backend, FFmpeg render pipeline, agent orchestration, React editor UI, source-grounded data model, and test harness. I made the product, architecture, and design decisions; Codex helped me iterate through each milestone rapidly and validate the difficult edges instead of stopping at a visual prototype.

GPT-5.6 Ultra was used to develop test cases and edge-case coverage for the editor workflow: source retention after a highlight cut, playback over source-cut boundaries, linked track actions, caption timing driven by the video clock, reframe evidence and smoothing, local caption rendering, export capability gates, responsive timeline behaviour, and recovery from interrupted ingest/export work.

Challenges we ran into

Challenge Why a shortcut fails How SolCut handles it
Original media Generating a highlight must not destroy the source needed for a later edit. Retains originals and proxies separately; Reset restores full sources and native framing.
Playback continuity A nice-looking timeline is not enough if pause, seek, or resume break at cuts. Tests normal playback and exact cut-boundary handoff across source ranges.
Honest captions Edited text cannot pretend it still has the original ASR word timings. Preserves source evidence; word-synchronised styles require verified alignment.
Reliable Auto Reframe A centre crop can look plausible while losing the speaker or subject. Requires verified focus-path evidence and bounds camera smoothing.
Creator control A black-box “AI edit” is difficult to trust or correct. Every operation is represented on an undoable, source-grounded timeline.

Accomplishments that we're proud of

  • Built a runnable local editor rather than stopping at a landing page or visual prototype.
  • Kept original media, proxies, source ranges, captions, and final delivery outputs traceable.
  • Made the AI director operate an editable, undoable timeline instead of returning an opaque template.
  • Connected responsive media, preview, layout, caption, timeline, and export workflows into one product experience.
  • Verified the core workflows with 60 Python unit tests and 20 Playwright end-to-end tests passing locally.

What we learned

The strongest AI creative tools do not replace the creator’s judgement; they remove the mechanical work between intent and output. I also learned that AI editing only earns trust when it can show its evidence: where a caption came from, which source range is used, why a frame follows a subject, and how the final export relates to the original footage.

SolCut is built around that principle: the creator directs; the agents operate the timeline.

What's next for SolCut

The next goal is to make SolCut more production-ready without giving up creator control:

  • Pluggable local and hosted transcription, visual-analysis, and subject-tracking workers.
  • Multilingual captions, dubbing, and consent-first voice workflows.
  • More creator-controlled caption, colour, motion, and layout presets.
  • Quality scoring for cuts, framing, caption readability, and audio ducking.
  • Background job queues, progress recovery, scalable rendering, and collaborative review links.

Built With

Share this project:

Updates