1<!DOCTYPE html><html lang="en" data-tone="slate" data-astro-cid-hkbrpulz><head><meta charset="utf-8"><meta name="viewport" content="width=device-width, initial-scale=1"><title>AI Research · Tim Brooks</title><meta name="description" content="Selected AI research by Tim Brooks â Muse, Sora, InstructPix2Pix, video generation and computational photography."><link rel="canonical" href="https://timothybrooks.com/research/"><link rel="icon" href="/favicon.ico" sizes="32x32"><link rel="icon" href="/favicon.svg" type="image/svg+xml"><link rel="apple-touch-icon" href="/apple-touch-icon.png"><link rel="preload" href="/fonts/newsreader-latin-wght-normal.woff2" as="font" type="font/woff2" crossorigin><meta property="og:type" content="website"><meta property="og:title" content="AI Research · Tim Brooks"><meta property="og:description" content="Selected AI research by Tim Brooks â Muse, Sora, InstructPix2Pix, video generation and computational photography."><meta property="og:url" content="https://timothybrooks.com/research/"><meta name="twitter:card" content="summary_large_image"><meta name="color-scheme" content="light"><link rel="stylesheet" href="/_astro/Base.B3WHaZwm.css"><style>.intro[data-astro-cid-omha2cvd]{margin-bottom:var(--space-lg)}.projects[data-astro-cid-omha2cvd]{margin-top:var(--space-xl);padding:0;list-style:none}.projects[data-astro-cid-omha2cvd]>li[data-astro-cid-omha2cvd].has-figure{grid-template-columns:minmax(0, var(--measure)) minmax(0, 1fr);gap:var(--space-sm);align-items:start;display:grid}.projects[data-astro-cid-omha2cvd]>li[data-astro-cid-omha2cvd]+li[data-astro-cid-omha2cvd]{margin-top:var(--space-2xl)}.text[data-astro-cid-omha2cvd]{max-width:var(--measure)}h2[data-astro-cid-omha2cvd]{font-size:1.45rem}.meta[data-astro-cid-omha2cvd]{margin-top:var(--space-2xs)}.body[data-astro-cid-omha2cvd]{margin-top:var(--space-sm);color:var(--ink-soft)}.links[data-astro-cid-omha2cvd]{margin-top:var(--space-sm);gap:var(--space-md);flex-wrap:wrap;display:flex}.links[data-astro-cid-omha2cvd] a[data-astro-cid-omha2cvd]{text-decoration:none}.links[data-astro-cid-omha2cvd] a[data-astro-cid-omha2cvd]:hover{color:var(--ink)}.figure[data-astro-cid-omha2cvd] img,.figure[data-astro-cid-omha2cvd] video[data-astro-cid-omha2cvd]{object-fit:cover;width:100%;height:auto;max-height:32rem;display:block}@media (width<=64rem){.projects[data-astro-cid-omha2cvd]>li[data-astro-cid-omha2cvd].has-figure{gap:var(--space-md);grid-template-columns:1fr}.figure[data-astro-cid-omha2cvd] img,.figure[data-astro-cid-omha2cvd] video[data-astro-cid-omha2cvd]{max-height:none}} 2</style><link rel=âpreconnectâ href=âhttps://www.google-analytics.comâ>
vendor: 64 bytes, lines 2-3
2 3<script async src="https://www.googletagmanager.com/gtag/js?id=
3UA-146060534-1
vendor: 12 bytes, line 3
3"></script>
4<script> 5
vendor: 133 bytes, lines 5-8
5window.dataLayer = window.dataLayer || []; 6 function gtag(){dataLayer.push(arguments);} 7 gtag('js', new Date()); 8 gtag('config', '
8UA-146060534-1
vendor: 4 bytes, line 8
8');
9</script>
9</head><body data-astro-cid-hkbrpulz><a class="skip" href="#main" data-astro-cid-hkbrpulz>Skip to content</a><div class="page" data-wide data-astro-cid-hkbrpulz><header class="site-header" data-astro-cid-fzpbxy5g><a class="wordmark" href="/" data-astro-cid-fzpbxy5g>Tim Brooks</a><nav aria-label="Primary" data-astro-cid-fzpbxy5g><a class="label nav-link" href="/" data-astro-cid-fzpbxy5g>About</a><a class="label nav-link" href="/research" aria-current="page" data-astro-cid-fzpbxy5g>AI Research</a><a class="label nav-link" href="/photography" data-astro-cid-fzpbxy5g>Photography</a></nav></header><main id="main" data-astro-cid-hkbrpulz><header class="intro" data-astro-cid-omha2cvd><h1 data-astro-cid-omha2cvd>AI Research</h1></header><ol class="projects" data-astro-cid-omha2cvd><li class="has-figure" data-astro-cid-omha2cvd><div class="text" data-astro-cid-omha2cvd><h2 data-astro-cid-omha2cvd>Muse Image and Muse Video</h2><p class="label meta" data-astro-cid-omha2cvd>Meta Superintelligence Labs · 2026 · Lead</p><p class="body" data-astro-cid-omha2cvd>Muse Image generates and edits images agentically â it calls search and coding tools to ground facts and get details like QR codes and plots right, then refines its own output. Muse Video is built on the same foundation, with native audio. Muse Image ships across the Meta AI app, meta.ai, Instagram Stories and WhatsApp; Muse Video is in early preview.</p><p class="links" data-astro-cid-omha2cvd><a class="label" href="https://ai.meta.com/blog/introducing-muse-image-muse-video-msl/" target="_blank" rel="noopener" data-astro-cid-omha2cvd>Research Blog â</a></p></div><div class="figure" data-astro-cid-omha2cvd><video src="/video/muse.mp4" poster="/video/muse.jpg" autoplay loop muted playsinline preload="metadata" aria-label="Muse Image and Muse Video â generated video sample" data-astro-cid-omha2cvd></video></div></li><li class="has-figure" data-astro-cid-omha2cvd><div class="text" data-astro-cid-omha2cvd><h2 data-astro-cid-omha2cvd>Sora</h2><p class="label meta" data-astro-cid-omha2cvd>OpenAI · 2024 · Co-lead</p><p class="body" data-astro-cid-omha2cvd>A generative model for video. I co-led the project, which trained a diffusion transformer on video and images of varying durations, resolutions and aspect ratios.</p><p class="links" data-astro-cid-omha2cvd><a class="label" href="https://openai.com/index/video-generation-models-as-world-simulators/" target="_blank" rel="noopener" data-astro-cid-omha2cvd>Research Blog â</a></p></div><div class="figure" data-astro-cid-omha2cvd><video src="/video/sora-tokyo.mp4" poster="/video/sora-tokyo.jpg" autoplay loop muted playsinline preload="metadata" aria-label="Sora â generated video sample" data-astro-cid-omha2cvd></video></div></li><li class="has-figure" data-astro-cid-omha2cvd><div class="text" data-astro-cid-omha2cvd><h2 data-astro-cid-omha2cvd>InstructPix2Pix</h2><p class="label meta" data-astro-cid-omha2cvd>Berkeley AI Research · 2023 · First author</p><p class="body" data-astro-cid-omha2cvd>Learning to follow image editing instructions. Given a photograph and a written instruction â "make it winter", "turn him into a cyborg" â the model performs the edit directly, in a single forward pass, with no per-image fine-tuning, masks or extra descriptions. With Aleksander Holynski and Alyosha Efros.</p><p class="links" data-astro-cid-omha2cvd><a class="label" href="https://arxiv.org/abs/2211.09800" target="_blank" rel="noopener" data-astro-cid-omha2cvd>Paper â</a><a class="label" href="https://instruct-pix2pix.timothybrooks.com/" target="_blank" rel="noopener" data-astro-cid-omha2cvd>Project page â</a></p></div><div class="figure" data-astro-cid-omha2cvd><img src="/_astro/instructpix2pix.D0K_Y2-y_NAYxM.webp" srcset="/_astro/instructpix2pix.D0K_Y2-y_xWedl.webp 600w, /_astro/instructpix2pix.D0K_Y2-y_ZsBw8z.webp 900w, /_astro/instructpix2pix.D0K_Y2-y_16ijUF.webp 1400w" alt="InstructPix2Pix â example output" sizes="(max-width: 64rem) 100vw, 55vw" loading="lazy" data-astro-cid-omha2cvd="true" decoding="async" data-astro-image="constrained" data-astro-image-fit="cover" data-astro-image-pos="center" width="2000" height="1014"></div></li><li class="has-figure" data-astro-cid-omha2cvd><div class="text" data-astro-cid-omha2cvd><h2 data-astro-cid-omha2cvd>Long Video Generation</h2><p class="label meta" data-astro-cid-omha2cvd>NVIDIA · 2022 · First author</p><p class="body" data-astro-cid-omha2cvd>Research on generative models for video, before video generation was a crowded field.</p><p class="links" data-astro-cid-omha2cvd><a class="label" href="https://arxiv.org/abs/2206.03429" target="_blank" rel="noopener" data-astro-cid-omha2cvd>Paper â</a><a class="label" href="https://www.timothybrooks.com/tech/long-video-gan/" target="_blank" rel="noopener" data-astro-cid-omha2cvd>Project page â</a></p></div><div class="figure" data-astro-cid-omha2cvd><img src="/_astro/nvidia.DyoI-bRS_ZwDXwF.webp" srcset="/_astro/nvidia.DyoI-bRS_1VSIx4.webp 600w, /_astro/nvidia.DyoI-bRS_Z15yTUt.webp 900w, /_astro/nvidia.DyoI-bRS_ZA0n82.webp 1400w" alt="Long Video Generation â example output" sizes="(max-width: 64rem) 100vw, 55vw" loading="lazy" data-astro-cid-omha2cvd="true" decoding="async" data-astro-image="constrained" data-astro-image-fit="cover" data-astro-image-pos="center" width="2224" height="864"></div></li><li class="has-figure" data-astro-cid-omha2cvd><div class="text" data-astro-cid-omha2cvd><h2 data-astro-cid-omha2cvd>Pixel Smartphone Camera</h2><p class="label meta" data-astro-cid-omha2cvd>Google · 2018</p><p class="body" data-astro-cid-omha2cvd>Computational photography for the Pixel phone â the part of the camera that is a model rather than a lens.</p><p class="links" data-astro-cid-omha2cvd><a class="label" href="https://blog.google/products-and-platforms/devices/pixel/see-light-night-sight/" target="_blank" rel="noopener" data-astro-cid-omha2cvd>Research Blog â</a></p></div><div class="figure" data-astro-cid-omha2cvd><img src="/_astro/pixel.Czwg0Ro1_p3Q0C.webp" srcset="/_astro/pixel.Czwg0Ro1_I9OMV.webp 600w, /_astro/pixel.Czwg0Ro1_2qWlV6.webp 900w, /_astro/pixel.Czwg0Ro1_1Ul75T.webp 1400w" alt="Pixel Smartphone Camera â example output" sizes="(max-width: 64rem) 100vw, 55vw" loading="lazy" data-astro-cid-omha2cvd="true" decoding="async" data-astro-image="constrained" data-astro-image-fit="cover" data-astro-image-pos="center" width="1600" height="840"></div></li></ol></main><footer class="site-footer" data-astro-cid-nns7i3if><ul data-astro-cid-nns7i3if><li data-astro-cid-nns7i3if><a class="label" href="mailto:[email protected]" data-astro-cid-nns7i3if>Email</a></li><li data-astro-cid-nns7i3if><a class="label" href="https://github.com/timothybrooks" data-astro-cid-nns7i3if>GitHub</a></li><li data-astro-cid-nns7i3if><a class="label" href="https://scholar.google.com/citations?hl=en&user=l9iNFaMAAAAJ" data-astro-cid-nns7i3if>Google Scholar</a></li></ul></footer></div>
9<script type="module">var e=()=>document.documentElement.setAttribute(`data-ready`,``);addEventListener(`pointerdown`,e,{once:!0}),addEventListener(`keydown`,e,{once:!0});</script>
9</body></html>
Line numbers count LF bytes from the start of the resource, as the search results do. Vendor segments are library code the classifier recognised; they are stored but not indexed. Bytes are shown as Latin1 characters, one per byte.