Lumen.
Direct, edit, and finish faster. One studio — a cinematic AI platform for solo filmmakers and small studios.
onboarding to delivery
- What
- A cinematic AI platform for solo filmmakers: generation, editing, colour, sound and delivery in one studio.
- Why
- Going from raw idea to finished film means juggling tools; Lumen keeps one timeline, prompt and project.
- My role
- UI/UX and branding: design system and an interactive prototype.
- Result
- 28 screens in one prototype · 6 weeks from blank canvas · 12 LUT looks
A cinematic AI platform that takes creators from raw idea to finished film — for solo creators and small studios that ship every week.
“Lumen is an all-in-one cinematic AI platform that streamlines content creation. Direct, edit, and finish faster with a seamless workflow.”
Read the full story
Generation, color, sound and delivery in one calm canvas.
A cinematic AI platform that helps creators move from raw idea to finished film. Built around fast generation, intentional editing and a filmic visual language — for solo creators and small studios that ship every week.
A concept exploring the future of cinematic video creation, designed and built at Stepikin Lab. Full design system and interactive prototype included.
28 screens. One product.
Onboarding → generation → post-production → delivery — every surface designed for the cinematic workflow.
Six tools, six bills, six exports. One studio instead.
A film crew, not a chatbot
Six tools, six bills, six exports
One timeline, one prompt, one project
“I lose 40 minutes a week just moving files between apps.”
Read the full story
Creators juggle multiple tools and slow workflows. Lumen solves this by combining generation, editing, and finishing into one fast, AI-powered platform.
Task — make AI feel like a film crew, not a chatbot. Design a workspace where prompt-based generation, frame-accurate editing, and finishing share one mental model — so creators stay in the work, not in the tool-switching.
Problem — six tools, six bills, six exports. Solo creators stitch generation, NLE, color, audio, captions, and publishing across half a dozen apps. Every handoff loses momentum and chips away at intent.
Solution — one timeline, one prompt, one project. A unified studio: generate clips inline, AI-cut to the beat, grade with curated LUTs, and ship — addressable from one canvas. Native, not glued together.
Pain points we heard: “I lose 40 minutes a week just moving files between apps.” · “AI generations look generic — I can’t shape them.” · “Captions look like an afterthought.” · “Color grading scares me out of trying.”
Design principles in response: Filmic by default · The prompt is the entry point everywhere · Density is intimacy, not clutter · Generative is a verb, not a feature tab · Earned ornament — every accent has to mean something.
From scribbles to scaffolding.
Low-fi wireframes locked the information architecture before any pixel got coloured.
Read the full story
Before any pixel got coloured, we built the bones. Low-fi wireframes lock the information architecture and prove the workflow stands up under load — five tools deep, four AI panels, one timeline, no clutter.
Project timeline. Designed through research, wireframing, and final UI execution, focused on a fast, seamless creator experience: week 1 research, week 2 information architecture, week 3 wireframe, week 4 UI design, week 5 prototyping.
Welcomed, not interrogated. Then straight to work.
Four steps to the first generationPick today’s workflow, import footage — the studio configures itself.
Where the work livesLibrary, preview, AI panel, multi-track timeline.
Read the full story
First run. Four steps from sign-in to first generation. Pick the workflow that matches today’s job, import the footage you have, and the studio configures itself accordingly. No empty canvas paralysis.
The editor. The full studio: media library left, preview centred, AI generation panel right, multi-track timeline below. Every region collapses; every action has a keyboard shortcut.
From prompt to picked clip.
Type the idea, watch four directions land, choose the one that matches intent. The work never leaves the canvas.
Color & sound, inline.
Two tools that traditionally take days — now part of the same canvas.
Curated LUTs, scope-drivenTwelve looks named for the films they reference.
Three stems, one earDialogue, score, SFX — auto-routed on import.
Read the full story
Curated LUTs honour cinema-grade looks; an audio mixer that auto-ducks dialogue, isolates stems, and sweetens room tone with one toggle.
Color grading. Twelve looks named for the films they reference — “Golden Hour”, “Neo Tokyo”, “Cold Pastoral”. A live waveform scope keeps you honest. Per-clip strength, never per-edit.
Audio mixing. Dialogue, score, SFX — auto-routed by AI on import. Auto-ducking under speech, room-tone matching across cuts, broadcast-safe loudness — all defaults you can override but rarely need to.
Edits that cut themselves.
Drop B-roll over a track and Auto-edit lays cuts on the beat.
Cut on the beat — every decision exposed, every one overridable.
Type the speaker, not the time
Edit the transcript; the captions follow.
Read the full story
Captions transcribe, segment, and styles ride a single timeline alongside the picture — corrections are inline, not modal.
Auto-edit — cut on the beat. Pick a track, set the energy curve, hit run. Lumen aligns cuts to the rhythm, lifts emphasis on the chorus, and exposes every decision so you can override what the AI got wrong.
Captions & transcript. Word-level timing, speaker diarisation, four caption presets that pass broadcast review. Edit the transcript and the captions follow — never the other way around.
Find it, finish it — together.
Search footage by content, deliver to every channel, and pin notes to a frame.
Read the full story
Media library — visual search across hours of footage. Footage searchable by content, not filename. Auto-tagged on import: faces, locations, camera moves, even mood. Type a phrase, get a shortlist with timecodes — no more scrubbing through bins at 2am.
Deliver — one project, every aspect ratio. Delivery presets for every channel, with adaptive transcoding and per-platform safe-area previews before render. Render queue with presets — TikTok 9:16, YouTube 16:9, Instagram square, broadcast 23.976. Safe areas overlay on the final preview so subtitles never end up cropped.
Live collaboration — cursors, comments, presence. Reviewers leave notes that pin to a timecode and a region — not “second 47”, but here, on this frame, with context. See who’s looking at what; resolve threads inline — the timeline keeps moving while critique happens. Multi-cursor presence and version history make async feel synchronous.
Version history — every cut, recoverable. Auto-snapshots on every render and at smart breakpoints — first cut, after color, after audio. Compare any two versions side-by-side or roll back without confirmation theatre.
Quiet pages, loud product.
Settings and billing get the same type ramp and restraint — trust is built where users least expect it.
Decisions you make onceStudio-level AI defaults every project inherits.
One bill, transparent metersCredits always visible in the topbar.
Read the full story
Settings and billing don’t get to be afterthoughts. They use the same type ramp, the same ratios, the same cinematic restraint as the main canvas — because trust is built where users least expect it.
Settings · AI defaults. Model, ratio, render preset, captioning, color science — set them once at the studio level and every new project inherits, every junior collaborator gets the same baseline.
Credits & billing. Generation credits visible in the topbar at all times. No surprise overages — when you hit 90% the studio dims the spark icon and asks before continuing.
Deep black. A generative gradient.
Black grounds the canvas; violet, magenta and amber form the generative signature.
- 01GeistProduct surfaces — 300 / 500 / 700.
- 02Instrument SerifItalic editorial accent, used sparingly.
- 03Geist MonoTimecodes, metadata, render specs.
Read the full story
Palette. A warm cinematic system. Deep black grounds the canvas; violet, magenta and amber form a generative gradient signature; a single editorial gold marks intent. Generative Violet #8A6FD1 — AI-touched surfaces: generation, suggestions, magic. Studio Magenta #D168A3 — active state, brand pulse, the tone that says “this is alive”. Filmic Amber #FFB877 — cinematic warmth: golden hour, render success, ready-to-ship. Deep Black #0A0A0C — the room lights down; every cinematic frame begins here.
Typography. Geist for product surfaces — a clean, modern sans for readability and density; tight tracking, neutral skeleton, works at every size from caption to display. Instrument Serif italic as the editorial accent — hero phrases, pull quotes and slow, deliberate moments, never utility text. Geist Mono 400 for technical metadata — timecodes, file metadata, render specs, model versions: anywhere the system speaks.
Six weeks. Solo, with AI.
From blank canvas to a 28-screen interactive prototype — AI-assisted across discovery, IA and visual design.
Read the full story
- Figma — UI design, components, flows.
- Framer — interactive prototypes.
- Claude (Sonnet 4.5) — discovery, copy, code review.
- Claude Code — front-end, screens, case build.
- Midjourney v7 — reference imagery, project art.
- Runway / Sora — motion references for AI flows.
- Notion — spec, decisions log.
- Geist · Instrument Serif — type system.
Thank you.
From brief to build — in one tool. A concept exploring the future of cinematic video creation, designed & built at Stepikin Lab.
