
Research @OpenAI, Sora videogen. Previously @UmichCSE, @GoogleDeepMind, @MetaAI, @berkeley_ai
Joined August 2016
- Tweets145
- Following998
- Followers1.5K
- Likes1K
Pinned Tweet
Can you make a jigsaw puzzle with two different solutions? Or an image that changes appearance when flipped?
We can do that, and a lot more, by using diffusion models to generate optical illusions!
Continue reading for more illusions and method details 🧵
Daniel Geng retweeted
🧐A question I've long been interested in: how can we learn from human hands and transfer that directly to robots?
Our new work, HUG, makes it possible in three simple steps: (1) collect human grasps at scale, (2) learn from them, and (3) retarget for deployment.
Daniel Geng retweeted
We present MotionStream — real-time, long-duration video generation that you can interactively control just by dragging your mouse.
All videos here are raw, real-time screen captures without any post-processing. Model runs on a single H100 at 29 FPS and 0.4s latency.
Daniel Geng retweeted
(1/n)
Can pretrained video diffusion models be prompted to track pixels — without any retraining?
We introduce Point-Prompting, a zero-shot point tracking method that simply prompts video models to visually mark and propagate points across time.
🌐 point-prompting.github.io
Daniel Geng retweeted
it's an honor to work with such an incredible group of researchers.
massive shoutout to @LiLiunian, @DmytroOk , avi, @ezhang887 , @ryw999 @troyluhman and eric, eric mintun, @bram_wallace @JureZbontar @mmmbchang @gabriel1 @dangengdg and @cmikeh2 for their incredible work on sora 2.
the sora research team is very special.
Daniel Geng retweeted
On the left is a rabbit. On the right is an elephant. But guess what: They’re the *same image* rotated 90°!
In @currentbiology, @chazfirestone & I show how these images—known as “visual anagrams”—can help solve a longstanding problem in cognitive science. bit.ly/45BVnCZ
Daniel Geng retweeted
Excited to share our CVPR 2025 paper on cross-modal space-time correspondence!
We present a method to match pixels across different modalities (RGB-Depth, RGB-Thermal, Photo-Sketch, and cross-style images) — trained entirely using unpaired data and self-supervision.
Our approach learns correspondences through contrastive random walks across visual modalities.
#CVPR2025 (1/6)
Daniel Geng retweeted
Ever wondered how a scene sounds👂 when you interact👋 with it?
Introducing our #CVPR2025 work "Hearing Hands: Generating Sounds from Physical Interactions in 3D Scenes" -- we make 3D scene reconstructions audibly interactive!
yimingdou.com/hearing_hands/
Daniel Geng retweeted
Can AI image detectors keep up with new fakes?
Mostly, no. Existing detectors are trained using a handful of models. But there are thousands in the wild!
Our work, Community Forensics, uses 4800+ generators to train detectors that generalize to new fakes.
#CVPR2025 🧵 (1/5)
Hello! If you like pretty images and videos and want a rec for CVPR oral session, you should def go to Image/Video Gen, Friday at 9am:
I'll be presenting "Motion Prompting" @RyanBurgert will be presenting "Go with the Flow" and @ChangPasca1650 will be presenting "LookingGlass"
Motion Prompting: motion-prompting.github.io/
Go with the Flow: eyeline-research.github.io/G…
LookingGlass: lookingglass-lpw.github.io/
2
420
Really cool paper about training interactivity into a video model with surprisingly little data! Great work!
Daniel Geng retweeted
Ever wish YouTube had 3D labels?
🚀Introducing🎥DynPose-100K🎥, an Internet-scale collection of diverse videos annotated with camera pose!
Applications include camera-controlled video generation🤩and learned dynamic pose estimation😯
Download: huggingface.co/datasets/nvid…
Daniel Geng retweeted
A sister team to ours at Google DeepMind is looking for student researchers this summer. Please reach out if you are a PhD student working on media generation (diffusion models), or if you are a professor with students to recommend! 😀
Daniel Geng retweeted
Our workshop at #ICLR2025 is now open to submissions until 02/03 🥳 check our website if you are interested: sites.google.com/view/icbinb…
Our #ICLR2025 @iclr_conf workshop is looking for submissions on unexpected outcomes and hard-earned lessons in #DeepLearning
Submission Deadline: 03 February 2025
Workshop Dates: 27 or 28 April 2025
Location: Singapore
More info:
shorturl.at/OpQns
Daniel Geng retweeted
Decentralized Diffusion Models power stronger models trained on more accessible infrastructure.
DDMs mitigate the networking bottleneck that locks training into expensive and power-hungry centralized clusters. They scale gracefully to billions of parameters and generate photorealistic images with just a week of training on eight independent GPU nodes. They’re easy to implement, adopt DiT hyperparameters directly and outperform standard models FLOP-for-FLOP.
Hey all, I'll be answering questions about our "Motion Prompting" paper on alphaXiv (@askalphaxiv) (it's like arXiv, but adds a discussion section, and I think is quite well built!):
alphaxiv.org/abs/2412.02700v…
I noticed I've been habitually using the same LLM over and over and I'm a bit worried I'm "overfit" so I made a website that randomly chooses from [Claude, Gemini, ChatGPT] for you:
imfeelinglucky . chat (actual link + repo in thread)
2
23
2,151
url: imfeelinglucky.chat
repo: github.com/dangeng/imfeeling…
hopefully you find this useful! let me know if there are problems, if somebody has done this before, or if you have any other comments!
4
443
















