Case 07 · Graduation Day
Pixel Rotator
The pitch buries its one fundable idea — Cursor-grade edit traces training a personal style model per artist — under three minutes of feature-consolidation talk, plays its only demo clip in silence, and ends with no name, no numbers, no team, and no ask.
01 Scorecard · 12 dimensions
02 Klaff 5-lens audit · 1/5
THE ONE REORDER — Open on the Cursor insight — nobody owns creative edit traces, we're building the instrument that collects them — let consolidation be the how, and end on the ask.
03 Delivery
64 fillers ("uh" ×26, "like" ×22, "um" ×6, "kind of" ×6) — nearly 1 word in 10, and the first sentence restarts itself. The measured pace is padded pace: an unrehearsed talk track, not an unlucky night.
04 Beat-by-beat
| Time | Beat | What works | The fix |
|---|---|---|---|
| 0:00 | Open | The product gets named immediately | The sentence restarts itself — memorize the first two sentences cold, and open on the insight, not the category |
| 0:25 | Problem: app-hopping | Concrete texture — the color-grading app, the effect "only this one app does," VHS scan lines | Attach a cost and a person: name one creator and what fragmentation costs them per project |
| 1:15 | One product | The scope is stated: video + photo + live in one studio | Three surfaces in one breath dilutes it — claim one beachhead surface, mention the rest as expansion |
| 1:50 | Unified vocabulary | "How the actual knobs work… across both of these spaces for consistency" — a real product principle | A feature paragraph doing a positioning job; compress to one sentence |
| 2:25 | Agents | "Anything you can do with your hands in the app… the agents will be able to do" — the strongest capability claim | All future tense. Say what the agent does today, or label it honestly as roadmap |
| 3:05 | Cursor traces | The secret: traces "kind of like Cursor does," before/middle/after steps, train image and video LoRAs | This is the opening beat, not minute three. Untangle it into three short sentences |
| 4:05 | Personal style model | "Every artist has their own style model" vs the common design language of shared engines — the most fundable line spoken | Land it and stop; as spoken it trails off and gets cut short by the demo |
| 4:30 | Demo clip | An artifact exists | "I'll just play one. Hmm. Cool." — thirty seconds of silence where narration should be. Script what the room is looking at |
| 4:45 | Close | — | "And yeah, that's Pixel Rotator. … Cool." is the anti-close. Replace with name + build status + the ask |
05 The gaps investors block on
- No ask — the room is given nothing to do, which alone caps the pitch.
- No name, no team: nobody on stage says who they are; "we" never resolves to people.
- Shipped vs roadmap is indistinguishable — every capability is "we're making" / "going to enable" / "will be able to."
- Zero numbers anywhere: no users, no market math, no price. Not one number in 4:49.
- The demo proves nothing as shown — a silent clip on a style-capture product could be raw footage.
- The moat is asserted by analogy ("kind of like Cursor does"), never argued — why can't incumbent editors collect the same traces?
The style-model play, with the app as the instrument. The pitch is two companies told in sequence: a consolidation app fighting every incumbent editor at once, and a trace-dataset play where every artist gets their own model. Pick the second — the pitch itself says consolidation is what makes the traces collectible. Lead with that causality instead of arriving at it in minute three.
“Pixel Rotator is the creative studio that turns your own edits into your own model — every knob agent-drivable, every edit a training trace, so your style ships as a LoRA that belongs to you, not to a common engine everybody uses.”
06 The rewritten pitch
Cursor got great for one reason: it owned the traces — every edit a developer made inside it became training data nobody else had. Nobody owns those traces for creative work. I'm [name], and Pixel Rotator is the instrument that collects them.
Creators today hop between apps — one for color grading, another because it's the only one that does VHS scan lines right. Pixel Rotator is one studio for video, photo, and live, where every knob speaks the same language — grain means grain, whether you're grading a film or a photo — and every knob can be driven by an agent: tell it the story you're trying to tell, and it does the masking and the typography, on iOS, iPad, and laptop.
But here's the point of consolidating: every edit inside it is a trace — before, middle, after — and those traces train a style model per artist. Not a shared engine with a common design language baked in. Your LoRA, learned from your edits, that only you own — to speed up your own workflow, or to license.
Today [build status: exactly what works, on which platform]. [N] creators [use it / are on the waitlist]. We're raising [$X] to [milestone]. This thirty-second clip was [what the clip actually is — made in the product, by whom, in how long]. Come make one.
Brackets are the founder's homework — the copy is done, the proof is theirs to supply. Every claim must match what's actually built.
07 Do these before the next take
- Pick the wedge and reorder the pitch around it — the Cursor-traces insight opens; consolidation becomes the how.
- Write the ask — amount, use of funds, 3 milestones. It is a paragraph of work and the difference between a talk and a pitch.
- Say your name, and add one team sentence: who "we" is and why these people can build this.
- State build status truthfully — one sentence separating what works today from what ships next; kills the tense ambiguity undermining every claim.
- Script the demo narration — never play a silent clip again.
- Filler drill + memorize the open and close: 9.6 fillers per 100 words ("uh" ×26, "like" ×22) and a restarted first sentence.