Reimagining Photoshop as a 2030 AI editor you can talk to.
Adobe’s brief: design a lightweight, mobile-first AI image editor for 2030 that stays fast on low-compute devices and keeps the human in control. I was the sole designer on a six-person team. We placed 6th of 23 IITs.
Problem
By 2030, editing moves to the phone, and to conversation.
Photoshop’s users have always relied on desktop-class hardware and dense toolbars.
Adobe’s brief looked ahead: by 2030, creators will expect mobile-first, AI-assisted editing on limited compute, where you say “brighten only the left half of the sky” instead of hunting for a tool.
The hard part Adobe named explicitly: keep the human in control while using small, efficient models. The user has to see what the AI did and be able to refine it.
The obvious response is to bolt AI onto the existing tool (Photoshop), plus a prompt box. I didn’t want that.
The bet
I didn’t want Photoshop 2030. I wanted an app that feels two-way.
Most AI editors work like a vending machine: type an input, get a result.
My stand was that the 2030 editor should feel like a conversation layered on top of fast direct actions: you talk to it when you want guidance, and go direct when you don’t.
The product should feel alive and context-aware, not like a form you submit.
That single decision drove these two things: Orby, an assistant that makes the app a partner rather than a passive tool, and a home experience built to feel like a 2030 AI product, not a phone-sized copy of the desktop app.
Intent over tools
You describe what you want; the app figures out which operation that is.
Human in the loop
You always see what changed and can refine it. AI assists, it doesn’t decide.
Conversation on top of speed
Talk to it for guidance; bypass it entirely when you already know.
The assistant
Orby is a guide you summon, not an assistant that interrupts.
Orby is what makes the app feel human: an emotive navigator that gives context-aware suggestions (“Remove tourists”, “Fix lighting”) instead of leaving you to find the right tool.
But an always-active AI companion is one wrong move away from Clippy: charming in a demo, exhausting in daily use.
So Orby earns its place by staying out of the way. It sits minimised at the edge of the screen, blinking, and only comes forward when touched or called.
And it’s never a gate: Prompt is direct-to-result, and Stylize and Auto-Enhance are one tap, so a power user can bypass Orby entirely.
It’s a guide for when you can’t express what you want, not a toll you pay on every edit.
Control
Two ways to remove the birds: quick, or precise.
“Human in the loop” only means something if the user can choose how much control they want. Ask to remove the birds from a photo and Orby offers two paths.
The Prompt tab: just type it, done.
The Select tab: mask the birds with stylus or touch and erase or inpaint, for precise results. Same goal, two levels of control, the user’s call.
That logic shapes the whole editor. I reduced the cluttered toolbar to a floating capsule of four decisive options: Select (one linear workflow that absorbs Heal, Patch and Move, where you select and then pick the intent: Erase, Move, or Inpaint), Prompt (conversational generation), Edit (legacy manual tools, tucked away), and Layers (non-destructive control).
AI is the default. Manual is always one tap away.
The constraint
On-device or cloud: the user’s choice, not a hidden default.
The brief’s hard constraint was low compute. Most solutions would hide that decision in the backend. I made it a visible toggle: Atom runs locally, private and fast on limited hardware, while PhotoFlux V30 runs in the cloud for high-fidelity results.
The constraint becomes a feature: privacy and speed when you want them, fidelity when you need it, and the user always knows which one they’re using.
Two smaller calls round it out. One-tap generative shortcuts (Stylize, Auto-Enhance) sit in the top navigation for instant access.
And every exported, AI-generated layer carries C2PA content credentials: provenance built in, because a 2030 AI tool should be honest about what it made.
Result & reflection
3rd highest score bracket in Design & UX. Overall 6th of 23 IITs, and a sharper view of AI-UX.
We got placed in the 3rd highest score bracket in design & UX. And for the overall, AI + Dev + Design Problem Statement, 6th of 23 IITs.
It was a team result across three tracks (product design, market research, and building the AI pipelines), and I led the product design. This is a concept built under competition time pressure, not a shipped product.
What I took from it: the hard part of designing for AI isn’t the generation. It’s control and trust.
The decisions that mattered weren’t the impressive outputs; they were the off-ramps: letting the user bypass Orby, the precise Select path next to the quick Prompt one, the visible model toggle.
Designing where the human stays in control taught me more than any single AI feature could.
Work Reimagining Photoshop as a 2030 AI editor you can talk to.
Adobe’s brief: design a lightweight, mobile-first AI image editor for 2030 that stays fast on low-compute devices and keeps the human in control. I was the sole designer on a six-person team. We placed 6th of 23 IITs.
Problem
By 2030, editing moves to the phone, and to conversation.
Photoshop’s users have always relied on desktop-class hardware and dense toolbars.
Adobe’s brief looked ahead: by 2030, creators will expect mobile-first, AI-assisted editing on limited compute, where you say “brighten only the left half of the sky” instead of hunting for a tool.
The hard part Adobe named explicitly: keep the human in control while using small, efficient models. The user has to see what the AI did and be able to refine it.
The obvious response is to bolt AI onto the existing tool (Photoshop), plus a prompt box. I didn’t want that.
The bet
I didn’t want Photoshop 2030. I wanted an app that feels two-way.
Most AI editors work like a vending machine: type an input, get a result.
My stand was that the 2030 editor should feel like a conversation layered on top of fast direct actions: you talk to it when you want guidance, and go direct when you don’t.
The product should feel alive and context-aware, not like a form you submit.
That single decision drove these two things: Orby, an assistant that makes the app a partner rather than a passive tool, and a home experience built to feel like a 2030 AI product, not a phone-sized copy of the desktop app.
Intent over tools
You describe what you want; the app figures out which operation that is.
Human in the loop
You always see what changed and can refine it. AI assists, it doesn’t decide.
Conversation on top of speed
Talk to it for guidance; bypass it entirely when you already know.
The assistant
Orby is a guide you summon, not an assistant that interrupts.
Orby is what makes the app feel human: an emotive navigator that gives context-aware suggestions (“Remove tourists”, “Fix lighting”) instead of leaving you to find the right tool.
But an always-active AI companion is one wrong move away from Clippy: charming in a demo, exhausting in daily use.
So Orby earns its place by staying out of the way. It sits minimised at the edge of the screen, blinking, and only comes forward when touched or called.
And it’s never a gate: Prompt is direct-to-result, and Stylize and Auto-Enhance are one tap, so a power user can bypass Orby entirely.
It’s a guide for when you can’t express what you want, not a toll you pay on every edit.
Control
Two ways to remove the birds: quick, or precise.
“Human in the loop” only means something if the user can choose how much control they want. Ask to remove the birds from a photo and Orby offers two paths.
The Prompt tab: just type it, done.
The Select tab: mask the birds with stylus or touch and erase or inpaint, for precise results. Same goal, two levels of control, the user’s call.
That logic shapes the whole editor. I reduced the cluttered toolbar to a floating capsule of four decisive options: Select (one linear workflow that absorbs Heal, Patch and Move, where you select and then pick the intent: Erase, Move, or Inpaint), Prompt (conversational generation), Edit (legacy manual tools, tucked away), and Layers (non-destructive control).
AI is the default. Manual is always one tap away.
The constraint
On-device or cloud: the user’s choice, not a hidden default.
The brief’s hard constraint was low compute. Most solutions would hide that decision in the backend. I made it a visible toggle: Atom runs locally, private and fast on limited hardware, while PhotoFlux V30 runs in the cloud for high-fidelity results.
The constraint becomes a feature: privacy and speed when you want them, fidelity when you need it, and the user always knows which one they’re using.
Two smaller calls round it out. One-tap generative shortcuts (Stylize, Auto-Enhance) sit in the top navigation for instant access.
And every exported, AI-generated layer carries C2PA content credentials: provenance built in, because a 2030 AI tool should be honest about what it made.
Result & reflection
3rd highest score bracket in Design & UX. Overall 6th of 23 IITs, and a sharper view of AI-UX.
We got placed in the 3rd highest score bracket in design & UX. And for the overall, AI + Dev + Design Problem Statement, 6th of 23 IITs.
It was a team result across three tracks (product design, market research, and building the AI pipelines), and I led the product design. This is a concept built under competition time pressure, not a shipped product.
What I took from it: the hard part of designing for AI isn’t the generation. It’s control and trust.
The decisions that mattered weren’t the impressive outputs; they were the off-ramps: letting the user bypass Orby, the precise Select path next to the quick Prompt one, the visible model toggle.
Designing where the human stays in control taught me more than any single AI feature could.
