01 / 08

Now · Adobe Research

Bryan Wang

I build AI that helps people create the stories they want to tell.

A scroll-through of how I got here, what I've made, and what I'm still trying to figure out.

Scroll

Bryan playing guitar in his college band

2013–2018 · NTU, Taipei

Origin

I was a CS major. And I was in a band.

For a while I really couldn't tell which one I was. Computer scientist? Musician? Both pulled equally hard.

Then I found out there were people training AI to generate music. Wait — that's a thing? Suddenly the two halves of me had something to do together. I had to try it.

Scroll

PerformanceNet — score-to-audio music generation

2019 · Academia Sinica

First taste

I spent a summer making AI play music.

Sheet music in, performed audio out. One of the first deep-learning models that could play a written score. AAAI 2019.

This was before ChatGPT. Before Suno. Before anyone called this gen AI. And for the first time, I felt this technology could touch something I loved.

PerformanceNetAAAI '19

Scroll

But —

Pure automation never quite satisfied me.

Watching the model play was wonderful. Watching it play without me felt empty.

I didn't want AI that replaced the creator. I wanted AI that collaborated with them. I needed a field that cared about humans, not just models. So I went looking.

Scroll

2018–2024 · University of Toronto

I found HCI

My PhD chased one question: how do humans and AI make things together?

I studied human-AI interaction, building systems across video, music, sound, and images. All trying to figure out the same thing: how creators can collaborate with AI to do more and do better. A few of them:

LAVE

IUI '24 · Video editing agent

LAVE

IUI '24 · Video editing agent

A video editor you talk to. Plans, finds clips, and edits through natural conversation.

SynthScribe

IUI '24 · Synth companion

SynthScribe

IUI '24 · Synth companion

A synthesizer assistant that helps you find and shape the sound you're chasing.

Promptify

UIST '23 · Image-gen prompts

Promptify

UIST '23 · Image-gen prompts

Turns text-to-image from a guessing game into a guided conversation with the model.

Soloist

CHI '21 · Guitar tutor

Soloist

CHI '21 · Guitar tutor

Turns YouTube guitar lessons into structured, step-by-step practice you can actually follow.

ROPE

UIST '22 · Audio storytelling

ROPE

UIST '22 · Audio storytelling

Automatically shortens long voice recordings into social-ready audio stories without losing the thread.

Stargazer

CHI '23 · How-to video robot

Stargazer

CHI '23 · How-to video robot

A camera robot that captures how-to videos by reading the instructor's subtle cues.

2024–present · Adobe Research

Now

After graduation, I kept pushing for this mission at Adobe Research.

I ship and I publish. Two Firefly features in production. Six papers at CHI, IUI, and UIST in '25 and '26. Four PhD interns through the door. I work with filmmakers, building bespoke tools to push the boundary of what's possible. The work, in tiles:

Precision Flow

Firefly · Image editing

Precision Flow

Firefly · Image editing

Every Firefly image edit becomes a continuous spectrum. Slide from A to B and land anywhere between.

Generate Soundtrack

Firefly · Music

Generate Soundtrack

Firefly · Music

Generates commercially-safe instrumental music that fits a video, from a prompt or the footage itself.

Vidmento

CHI '26 · Generative video

Vidmento

CHI '26 · Generative video

Fills the gaps in your video story by generating new context-aware footage between existing shots.

VidTune

CHI '26 · Video soundtracks

VidTune

CHI '26 · Video soundtracks

Helps video creators find the right soundtrack by generating music that matches the footage's mood.

Script2Screen

IUI '26 · Audiovisual scripts

Script2Screen

IUI '26 · Audiovisual scripts

Brings dialogue scripts to life with interactive audio + visual generation as you write.

MapStory

UIST '25 · LLM map agents

MapStory

UIST '25 · LLM map agents

Prototypes editable map animations with LLM agents. Sketch a story, get a map sequence.

SimTube

IUI '25 · Simulated viewers

SimTube

IUI '25 · Simulated viewers

Simulates how an audience would react to your video, with AI-generated personas and comments.

Today

What I keep returning to

Two tradeoffs I keep wrestling with.

Too much →

Automation.

Delegating feels great — until everything you make is slop. The output is fine, but the work stops feeling like yours. You lose ownership. Eventually, the skill itself.

Too much →

Control.

Steering every detail also feels great — until the tool becomes the work. Photoshop is a skill before it's a brush. Most people never get past the learning curve.

Tomorrow

What I'm working on

Where's the right balance?

The answer is more nuanced than one rule, and it shifts with the creator, the task, the moment. That's what I'm trying to figure out: AI that's there when you're stuck, out of the way when you're flowing.

If you're working on the same thing, let's talk.

See the work