Miju Labs

← All buyers · Meta (Meta Superintelligence Labs)

H

Huiwen Chang

VP of AI Research, Meta TBD Lab (secondary source) · Meta (Meta Superintelligence Labs)

low confidenceTier A · Talk to firstFrontier labsimage

One of the most accomplished image-generation researchers working anywhere: MaskGIT, Muse, Imagic, Palette, StyleDrop and MAGVIT at Google, then GPT-4o image generation at OpenAI, now a research leader for multimodal work at Meta's TBD Lab. With Jiahui Yu's departure in August 2026 she is plausibly the most senior image person left at Meta Superintelligence Labs.

Background

Chang did her PhD at Princeton under Adam Finkelstein after Tsinghua, and spent 2018-2023 at Google developing image-generation models — MaskGIT and Muse introduced masked generative transformers for parallel image synthesis, Imagic showed text-based editing of real photographs by optimising text embeddings inside a diffusion model, and StyleDrop and Palette extended style-driven and image-to-image generation. Her work is consistently about control: getting a model to do a specific thing to a specific image rather than sample something plausible.

She moved to OpenAI in 2023 and was a contributor to GPT-4o including its image generation, then joined Meta in the 2025 Superintelligence Labs wave. A secondary research-profile source lists her as VP of AI Research at Meta TBD Lab as of 2025, focused on multimodal intelligence and generative computer vision — a materially more senior title than 'researcher'.

What they run now

Multimodal intelligence and generative computer vision at TBD Lab — the research underneath Meta's Muse Image and Muse Video line. The control problems she has worked on throughout her career (precise editing, style transfer, identity preservation) are the ones where human judgement of the result is the only reliable measure.

What they buy

Inference. A research leader over generative vision needs preference data on editing fidelity and style adherence, style and aesthetic reference sets, and identity-preservation judgements — the same shopping list as the Google DeepMind and OpenAI image teams. Meta's Scale AI relationship makes external specialist purchase less likely but not impossible, particularly for niche creative-professional judgement that a general labelling vendor cannot staff.

Career

  1. 2018 – 2023Research scientist, Google — MaskGIT, Muse, Imagic, Palette, StyleDrop, MAGVIT
  2. 2023 – 2025Researcher, OpenAI — GPT-4o contributor including image generation
  3. 2025 – presentVP of AI Research (multimodal / generative vision), Meta Superintelligence Labs (TBD Lab) — Title from a secondary research-profile source
  4. 2013-2017Princeton University, PhD, Computer Science (advisor Adam Finkelstein)
  5. —Tsinghua University, Bachelor's
Why it matters here

Medium, rising. She is the most credible remaining image-research leader at Meta and, if the VP title is right, senior enough to commission data. The counterweight is Meta's internalised data supply. She is worth a dossier and a watch; she is not where Julian should spend his first ten hours.

How to reach

Research register, via her published work. Speculation: her career-long theme is controllable editing, so the artefact that would interest her is a human-graded dataset on edit fidelity — did the edit change exactly what was asked and nothing else — across a hard set of real photographs, with designer and retoucher rationale. Approach only after the Google DeepMind and Anthropic conversations are live.

What we could not establish
  • Her VP title comes from a secondary research-profile aggregator and should be confirmed before use in outreach.
  • No verified handle, LinkedIn or personal site found.
  • Whether she inherited Jiahui Yu's multimodal remit after his August 2026 departure.
  • Whether Meta procures any creative human data outside Scale AI.

Why they are on the map

Hired for GPT-4o image generation work; invented text-to-image architectures. Current role not confirmed after 2025. source — evidence dated 2025-07

Needs the data for a model; writes the spec.

How Meta (Meta Superintelligence Labs) buys data

Aug 2025: TBD Lab researchers reportedly preferred Surge and Mercor data over Scale's, despite Meta's $14.3B Scale stake. Meta has also partnered with Mercor on RL environments (TechCrunch, Sep 2025). In Aug 2025 Meta licensed Midjourney's 'aesthetic technology' to make its models 'more visually appealing', which is a direct signal of paying for visual taste. Muse Image/Video (Jul 2026) are marketed on human-preference Elo from internal ablations and Arena. SemiAnalysis (Jul 2026) reports about 3,000 engineers working on RL task/environment creation.

Alexandr Wang (Chief AI Officer, org context only) heads MSL. Since Aug 2025 it has four parts: TBD Lab (frontier models), FAIR (Rob Fergus), Products and Applied Research (Nat Friedman) and MSL Infra (Aparna Ramani, Amir Frenkel). Shengjia Zhao is Chief Scientist and owns the Muse roadmap; Jiahui Yu leads multimodal, including Muse Image/Video. No human-data owner is public; data-vendor decisions appear to sit with TBD Lab researchers and ex-Scale staff.

Known vendors: scale-ai, surge-ai, mercor, lmarena, Midjourney (aesthetics tech licence)

Sources

  1. Huiwen Chang research profile
  2. Meta TBD Labs researchers
  3. Meta Superintelligence Labs