Ruby Labs

Creative is the biggest lever in paid social — and the fastest to burn out

In performance marketing, targeting is increasingly automated; creative is where campaigns are won. But creative fatigues fast. The business problem is industrial: produce enough distinct, on-strategy concepts to keep testing pipelines full, learn from every result, and do it at a cost and speed that makes the economics work.

Seven verticals, one production discipline

Work spanned fiction, relationships, casino, puzzle, career, AI education, and privacy products — each with its own audience psychology and platform behavior. Goals were constant across all of them:

• Volume with intent — every creative is a hypothesis about a hook, not a decoration.
• Speed — typical concept-to-delivery in 48 hours, so UA never waits on design.
• Format discipline — every ad delivered in 9:16, 4:5, 1:1, 16:9 with placement-specific safe zones, so captions and CTAs never get clipped.

A system produces range, not one lucky ad

Shipped verticals — exactly as the feed sees them

Live campaign assets for Sagabox (fiction) and JobAssist (career), Meta/FB verticals — AI-generated, hand-finished, sound-off first:

Photoreal & cinematic range

The same pipeline, stretched across registers — documentary-real UGC scenes to gothic cinema, all generated:

Hook first, asset second, numbers last — then repeat

• Concept & hook writing — what stops the scroll and for whom, defined before any asset exists.
• AI-assisted generation — characters, scenes, and key art produced with AI image/video tooling; UGC-style testimonial ads with AI actors directed to feel native to the feed.
• Finishing — cut, paced, and polished in After Effects and Premiere: captions timed for sound-off viewing, sound design, CTA matched to the funnel.
• Testing with UA — variation sets (hooks, first frames, formats) built deliberately for structured tests, not produced at random.
• Iteration on data — winners get variation families; losers get autopsies. CTR tells us about the hook, CPA and downstream conversion tell us about the promise.

1. Generation is a starting point — composition and casting proven cheaply with AI
2. Craft happens in the edit — color, pacing, captions, sound design are manual
3. Judgment stays human — what ships is a creative decision, not a model output

Creative testing board — one hypothesis, three iterations (puzzle vertical)

V1 — Baseline

Hypothesis: a solvable-looking puzzle stops the scroll. Clean composition, single challenge.

V2 — Raised stakes

Same hook, new composition and difficulty cue — testing whether perceived challenge drives taps.

V3 — Recomposed

Winning elements kept, weak ones replaced — the variant family exists to learn, not to decorate.

Static

Key art generated and composited into ready-to-run templates. Cheapest to test — first into rotation, fastest read on the hook.

Motion

AI-generated video finished in After Effects: pacing, sound design, captions timed for sound-off. Deployed once a static proves the angle.

Decisions a media buyer would recognize

• Captions built for sound-off — most feed viewing is muted; the hook must land in text.
• Story teasers for fiction apps, talking avatars for product apps — curiosity converts stories; trust converts products. Format follows funnel.
• Reframed, never stretched — each aspect ratio is a recomposition, keeping the subject and CTA inside guaranteed-visible zones.
• AI where it compounds, humans where it counts — generation accelerates production; hooks, pacing, and judgment stay manual.

A creative pipeline UA can rely on

A production system, verified by the numbers on my CV: ≈6 finished creatives per day inside a team shipping 150–200+ ads per week for Sagabox across Meta, TikTok, Instagram/Reels and YouTube Shorts. Assets I created reached internal Winner Ad / Legacy Winner Ad status and scaled into five-figure ad spend while active; rated 115% on the designer productivity scorecard (May 2026).

Specific CTR/CPA values are internal; the volume, Winner-status, and spend-scale figures above are from my verified CV.

Data is a brief, not a verdict

Performance numbers don't replace design judgment — they sharpen it. A failed variant is the cheapest research available; the discipline is reading why it failed and encoding that into the next concept. This loop is the most transferable skill I bring to any product team.

Related projects: Creative Direction (visual storytelling behind the ads), Marketing + Motion in the Design Archive.

Why it was built this way