Melissa Hall

Excited to release EvalGIM, an easy-to-use evaluation library for generative image models. EvalGIM ("EvalGym") unifies metrics, datasets, & visualizations, is customizable & extensible to new benchmarks, & provides actionable insights. Check it out! github.com/facebookresear…

Super excited about our new paper. If you are working on reward models, judges or post-training pipelines for omni-models, we hope MMRB2 helps you benchmark progress. See Yushi's thread for full details. Paper: arxiv.org/abs/2512.16899 Code: github.com/facebookresear…

6/n How big improvement can a good reward model give? We simply use it for best-of-N selection, and see big performance gap on four downstream tasks, even for Nano Banana. MMRB2 acc strongly correlate with the performance gain. Even bigger gain may be achieved through RL

Reward models make or break post-training for multimodal omni models (e.g., nano banana), yet there’s surprisingly little research on that‼️ We’re releasing MMRB2: new reward benchmark focusing on omni models, spanning T2I, editing, interleaved, and thinking with images 🧵1/n

Can we simplify video generation by decomposing it into interleaved text-video co-generation? Would explicit, repeated thinking in language improve generation in pixels? We introduce TV2TV: a unified model that jointly learns - language modeling (next-token prediction) - video flow matching (next-frame prediction) At inference, TV2TV dynamically alternates between textual thinking and video generation. Model generations below: interleaved text plans and video slices (~1–2s) are co-generated over time, conditioned on a single frame per sport. 📖 arxiv.org/abs/2512.05103

Happening today at 11AM PST! Pass by poster 4300 to have a chat 🌊

"Increasing the Utility of Synthetic Images through Chamfer Guidance" has been accepted at NeurIPS 2025! 🎉 We're happy to share our approach for generating synthetic images that are both diverse and high-quality. #NeurIPS2025 Paper: arxiv.org/abs/2508.10631

1/ My model can’t generate “full” refrigerators when I ask for it?? — but it does by default anyway when I don’t specify!? At #ICCV2025, we’re presenting DIMCIM — a new framework to quantify diversity and capacity in T2I models, that enables pinpointing such failure modes

📢 PhD Students in CV/GenAI! 📢 Our team at FAIR is hiring a Research Intern for Summer 2026 to push the boundaries of Multimodal Generative AI. Projects include Diffusion, Video/Image Generation, and Multimodal Transformers. Learn more and apply: metacareers.com/jobs/218993527…

Transfusion combines autoregressive with diffusion to train a single transformer, but what if we combine Flow with Flow? 🤔 🌊OneFlow🌊 the first non-autoregressive model to generate text and images concurrently using a single transformer—unifying Edit Flow (text) with Flow Matching (images). Performance boost while unlocking new capabilities: 🔥 Scales better than Transfusion (AR), 50% fewer FLOPS than Transfusion for similar performance ⚡ Mixed-modal training boosts both generation & understanding 😑 Mask diffusion: fixed-length + extra FLOPS for masked tokens. Transfusion: sequential generation ✅ OneFlow: variable-length via token deletion + concurrent mixed-modal generation with fewer FLOPS 🧵1/n

✨See you this afternoon 1:30-5:30 at Room 106A for the 2nd Workshop on Responsible GenAI @CVPR✨ Check out the program and contributed works here: sites.google.com/view/cvpr-resp… #CVPR2025

Join us Thursday afternoon in Room 106A for a great set of speakers, panelists, and contributed works focused on Responsible Generative AI @CVPR!! #CVPR2025

Join us at #CVPR2025 Demographic Diversity in Computer Vision workshop tomorrow! 📅 Wednesday, June 11, 9am-6pm 📍 room 213 (main session) + Hall D (poster sessions), the Music City Center We have an amazing lineup of speakers and panelists! Can't wait to meet you all there :)

Less than one week away from the 2nd Workshop on Responsible Generative AI @CVPR! It’ll be on Thursday afternoon, and we’ve got a great set of presenters and participants! Check out the program & accepted work: sites.google.com/view/cvpr-resp… See you in Nashville! 🎶 #cvpr2025

Deliberate practice is accepted to #ICML2025 as a spotlight (top 2.6%!) 🚀

BLT weights are out!

📣 Reminder that today is the last day to submit to the Responsible GenAI Workshop @CVPR !! Works focused on responsible AI + generative models w/ at least one modality that is visually grounded are welcome! Submit here: cmt3.research.microsoft.com/ReGenAI2025 #CVPR2025

Just 3⃣ more day until the submission deadline for the Responsible GenAI workshop @CVPR !! Submit here: cmt3.research.microsoft.com/ReGenAI2025 #CVPR2025

✨Updated submission deadline: Friday March 14! 5 more days to submit to the 2nd edition of the Responsible GenAI workshop @CVPR! We welcome works about responsible AI & generative models where at least one modality is visually grounded. cmt3.research.microsoft.com/ReGenAI2025 #CVPR2025

🔥Submission link to ReGenAI workshop #CVPR2025 is open at CMT: cmt3.research.microsoft.com/ReGenAI2025 Kind reminder that the due date is 10th March, and we look forward to your submissions! @CVPR

If you're interested in how to keep challenging neural networks throughout training, check out our latest preprint! #sample_efficiency #scaling_laws