One control vocabulary can unify photo, video, and live creative workflows.
Pixel Rotator
A creative workspace for photo, video, and live production, with shared controls and a future agent layer that can learn from each creator's editing process.
Key takeaways
An agent layer could operate the same creative controls as human editors across devices.
The longer-term thesis is that edit traces could support creator-specific style models rather than one shared visual default.
Summary
Pixel Rotator begins with a familiar creative-workflow problem: filmmakers, photographers, and online creators often move among specialized applications for color, effects, still images, video, and live capture. The project aims to consolidate the overlapping parts of those workflows in one studio.
The interface is meant to use a consistent vocabulary for photo and video operations. Controls such as grain, masking, typography, and related adjustments should behave predictably instead of changing meaning from one medium or application to another.
Agent control is a stated next layer rather than a demonstrated finished feature. The presenter described natural-language skills operating the same knobs a person can manipulate by hand across iOS, iPad, and laptop versions.
Consolidation also creates a proposed learning loop: capture the intermediate steps between an original and final edit, use those traces as training data, and eventually build a style model for each artist. The talk did not establish the data-governance, model-quality, or production-readiness details of that plan.
How it works
Unify creative workflows
Photo, video, effects, and live-creation tools share one studio instead of requiring separate applications.
Use one control vocabulary
The same creative concepts and knobs are intended to behave consistently across media.
Give agents the same controls
Planned natural-language skills would operate the same editing controls as the creator across iOS, iPad, and laptop.
Train a personal style model
Before, intermediate, and final edit traces would become training data for a creator-specific image or video model.
Transcript
So I'm presenting Pixel Rotator, and, I'm presenting Pixel Rotator, and, it is essentially like a, studio for creators, rather filmmakers or photographers or influencers, that want to speed up their workflow. So as you use it, learns how you work and it learns your style.
Essentially one of the big problems I'm trying to solve with this is creators very commonly move between a lot of different apps. They'll go to like like if they're working on video, they might go to one specific one that they like to do color grading in, or have a specific effect that only this one app does, you know, really authentically, like a VHS kind of, scan lines or something like that. So, there's a lot of app, app hopping and just like fragmentation between these things.
So we're making one product that has enough of the, features that overlap for video and photo and some of the live stuff, like if you're recording yourself with a, like a forward-facing or a backward-facing camera really, for any kind of like live stream, maybe a podcast. And, we're aggregating that all into one spot.
It uses the same, let's say vocabulary for all of these features. So like most of the time, video and video and photos are separated, and the features in it are a little different. Like the grain over here will work, will work differently. One of the effects isn't in video, but it's in photo. And yeah, so we're consolidating everything kind of in the language itself, how the actual knobs work, what they do across both of these spaces for consistency.
And of course, we're going to enable agents for this, so all of the knobs in this application can be controlled and driven by agents. You can use skills that better understand, you know, natural language on what you're what story you're trying to tell. And anything you can do with your hands in the app by masking or, typography, any of that stuff, the agents will be able to do that stuff and drive the application entirely on iOS, iPad, and, on laptop also.
Yeah, so also part of the style capture thing essentially is when you consolidate this kind of feature set into one application, you can actually collect all the traces from it, kind of like Cursor does. Cursor got really good with what they offer by, having like a high quality stream of these traces from editor, people that were developing and stuff, so as they edit and whatnot, they got a good data set. When you bring it all together, like you can actually get a data set that learns how to do the edits directly. It'll have like before and after and middle, steps between a final product, so you can train like image generation and video generation LoRAs that aren't just like a part of some common, like common engine everybody uses, right?
When you use a big model for these things, there's like a common design language that does kind of come out in a lot of this. And this is different because it's not about like sharing a common model. It's going to be more about every artist has their own, style model. So if they if you want to use it to make your own workflow faster, or if you want to enable other people to use it, like and it... Oh, yes. Yeah, sorry. I'll just go to this. So we have some examples of some, some of this type of stuff that you can make in this. I'll just play one.