The angle is time recovery. The Slack frame is a trigger moment, which the brief flags as retargeting only, and I agree with that read. Hooks 1 and 2 are the retargeting cuts. Hook 3 opens on the message rather than the deadline, which is the cold version of the same spine.
Opens mid-squint, reading the message. No setup, no establishing shot. The situation is the hook. The only one that ends on a face.
Testing whether a face can earn the open without costing the middle. The brief says talking head structures hold worse, so the face gets three seconds and hands off to the scene. If hold drops against the other two, the handoff came too late.
A team chat with the ask already in it. The situation is not an event, it is where you are. Only the reply animates, and it holds 2.3 seconds. Ends in the same channel, two minutes later.
Testing recognition. The viewer sees a moment they have lived before the product appears. Expect the most qualified clicks.
No product, no person. The line, big, on Deep Ocean. “Need a deck?” lands and holds a beat before it answers itself. The wordmark sits at the bottom. Closes back in the channel.
Testing clarity. States the need with no setup. Expect the lowest hold and the highest click to signup from the people who stay.
All three are judged on trial to paid, not on cheap clicks. A hook that wins CPC and loses activation is a loss.
The middle is where their viewers leave. The spine never cuts away from the screen, so the payoff and the product are the same shot.
Nothing in the image claims speed. The clock does. The boss asks at 9:46, the deck lands at 9:47, and the reaction at 9:52 asks the question the viewer is already forming. Delivered at the brief’s sizes, with the wordmark held at its full clearspace of one “M”.
Each one links to the full-resolution file. Supplied at 2× for retina placements; the trafficked sizes are the ones listed.
Download all assetsnate-fox-gamma-take-home.zip · 3 video, 3 staticI build the system first and the output second. The product UI in these spots is a build, not a screen recording, so the copy, timing, cursor path, captions and slides all live in config. The first one cost most of the time in the table below. The hundredth is a diff.
| Claude Code (Opus 5) | The build environment. Research, asset acquisition, writing the animation engine, driving every render. |
|---|---|
| Gamma’s own product footage | Reference. Their CDN serves first-party 1200×950 captures of the real Generate flow, frame-sampled for exact colours, radii and spacing. |
| Figma | Design layouts and pacing. Boards for the three ratios and a beat strip for the 20 seconds, so timing was settled before anything rendered. |
| Gamma brand book + kit | Canonical palette, the −2% headline tracking rule, wordmark files. |
| ES Build + PP Mori | Gamma’s actual typefaces inside the work itself. Not lookalikes. |
| Custom HTML/CSS/JS engine | The UI is rebuilt as vector, not filmed. One seek(t) function describes any frame of any variant in any ratio. |
| Puppeteer + ffmpeg | Frame-exact capture and encode. H.264 for delivery; ProRes 4444 and VP9 with alpha for the layered files. |
The first pass built the system, not the spot. Copy, timing, camera, captions and the slides live in one config, and the UI is built rather than filmed, so a variant is a config change and not a new edit.
That got tested mid-build. The concept changed completely: new scene, new hooks, new line, new spine. The rebuild took about 50 minutes, and the previous concept still renders untouched from its archived config. The config is the asset.
| Research, brand and asset acquisition | 30 min |
|---|---|
| Animation engine: chat, title card, spine | 50 min |
| Three hook variants, 9:16 | 10 to 15 min each |
| Layered alpha exports (ProRes 4444 + VP9) | 15 to 20 min |
| Statics, three ratios, layered | 30 min |
| Process note and this page | 30 min |
| Further iterations and concepts | 30 to 45 min |
| Total | ~3.5 hours |
These three took about three and a half hours. The next thirty take one to two, because most of that time went into things that outlive the asset: the layer template, the prompt set, and the render config. A new variant changes contents, not structure. Past that point the constraint is review, not production.
Nothing ships until the benefit reads with sound off and the caption hidden, the text clears the safe area in every ratio, and the product UI matches the current build. A failed check goes back rather than shipping as a lower tier variant.
The brief says hold is the weak spot and hook rate is fine. If that is right, three hooks against one spine tests the part that already works. I would swap it for a 20 second and a 5 second cut of the same concept. Canva is running 5 second spots well right now. A cut that short does not need hold, it needs the idea to land immediately, and it tells you fast whether the concept survives compression before you spend on the long version.
Directions explored, not built. The one I shipped clears the three second clarity bar for cold traffic. These lean the other way, toward brand. Every angle on the performing list is pain removal, which works, and none of them test whether anyone remembers Gamma the next day. In a category getting more crowded every quarter, that is the untested axis.
Dexter Lawrence is “Sexy Dexy” to every NFL fan right now. Sexy Decksy borrows the zeitgeist during season, with a talent partnership and a domain to match. Highest attention ceiling of anything here. Needs licensing, so it’s a real campaign, not a four-hour test.
“So easy your grandma can use it.” Character-led humor, tests the attention axis the main concept doesn’t touch. The risk is the pattern the brief flags: great CTR, no signups. Worth one variant to find out where the line is.
Magical, Expressive, Limitless is the only language in Gamma’s brand book with real heat, and it is sitting in the design system section instead of in market. Every performing angle in the brief is pain-removal. MEL is the aspiration side. Untested hypothesis: does “limitless” beat “effortless” for cold traffic?
The rallying cry for an all-nighter, inverted. Nobody has to stay late. The pun carries the product benefit rather than just being clever about the category.
The framing extends past one spot into episodic content about workplace time pressure, closer to Built’s Roomies than to a performance ad. Wrong tool for CPA, right tool for the brand layer that makes performance ads cheaper over time.