iiinova — where your digital self keeps evolving

Michael J. Black @michael-j-black.bsky.social · 09/06/2025

InteractVLM: 3D Interaction Reasoning from 2D Foundational Models interactvlm.is.tue.mpg.de InterDyn: Controllable Interactive Dynamics with Video Diffusion Models interdyn.is.tue.mpg.de Reconstructing Animals and the Wild raw.is.tue.mpg.de Workshop paper: Generative Zoo genzoo.is.tue.mpg.de

030

Michael J. Black @michael-j-black.bsky.social · 09/06/2025

ChatGarment: Garment Estimation, Generation and Editing via Large Language Models chatgarment.github.io ChatHuman: Chatting about 3D Humans with Tools chathuman.github.io PICO: Reconstructing 3D People In Contact with Objects pico.is.tue.mpg.de

140

Michael J. Black @michael-j-black.bsky.social · 09/06/2025

Here are all the CVPR projects that I’m part of in one thread. Conference papers: PromptHMR: Promptable Human Mesh Recovery yufu-wang.github.io/phmr-page/ DiffLocks: Generating 3D Hair from a Single Image using Diffusion Models radualexandru.github.io/difflocks/

2102

Reposted by Michael J. Black

Meshcapade @meshcapade.com · 05/06/2025

Final drop in our #CVPR2025 video series: PICO 🤝📦 Watch how we reconstruct realistic human-object interaction from just one image—with dense contact and mesh fitting! 👋 Visit our booth 1333 at CVPR. 🔗 Paper link in the thread. #3DBody #AI #SMPL

253

Michael J. Black @michael-j-black.bsky.social · 28/05/2025

My dream has been to take a photo of a person and extract the 2D sewing pattern of their clothing and then turn it into a 3D garment. ChatGarment does exactly this plus it lets you edit the garment, or create a completely new one, using text prompts. chatgarment.github.io At CVPR2025!

chatgarment.github.io

ChatGarment: Garment Estimation, Generation and Editing via Large Language Models

A novel approach that leverages large vision-language models (VLMs) to automate the estimation, generation, and editing of 3D garment sewing patterns from images or text descriptions.

0105

Michael J. Black @michael-j-black.bsky.social · 23/05/2025

To estimate 3D humans from video in world coordinates we add side information to prompt the process. Prompts include bounding boxes, face detections, segmentation masks, and text descriptions. It's currently the most accurate video-based method out there. See us at CVPR2025.

041

Michael J. Black @michael-j-black.bsky.social · 20/05/2025

Check out DiffLocks, appearing at #CVPR2025. From a single image, we estimate about 100K hair strands that you can then physically simulate. We use a dataset of 40K synthetic hair images with ground truth strands. It's all available for research purposes.

1195

Reposted by Michael J. Black

Meshcapade @meshcapade.com · 15/05/2025

Yay, @cvprconference.bsky.social, we’re in! 🎉#CVPR2025 5 papers accepted, 5 going live 🚀 Catch PromptHMR, DiffLocks, ChatHuman, ChatGarment & PICO at Booth 1333, June 11–15. Details about the papers in the thread! 👇 #3DBody #SMPL #GenerativeAI #MachineLearning

554

Reposted by Michael J. Black

Meshcapade @meshcapade.com · 13/05/2025

🔥 Heading to the #ICRA2025? Join us on May 23, 2pm (Room 316) for MoCapade: markerless motion capture from any video! Powered by PromptHMR (CVPR 2025). No suits, no markers—just motion. 🕺💻 #AI #3DMotion #SMPL #Robotics #ICRA #Meshcapade

153

Michael J. Black @michael-j-black.bsky.social · 08/05/2025

Improved foot-ground contact coming soon to MoCapade3.0.

051

Reposted by Michael J. Black

Meshcapade @meshcapade.com · 28/04/2025

🎬 Join Meshcapade at FMX! See how to turn video or text into ready-to-use 3D motion—no suits or markers needed. Workshop: May 8, 10:00 AM. Perfect for animation, VFX & game dev! 📍 Info: fmx.de/en/program/p... #FMX2025 #3DMotion

022

Reposted by Michael J. Black

Zhenjun Zhao @ericzzj.bsky.social · 18/04/2025

St4RTrack: Simultaneous 4D Reconstruction and Tracking in the World Haiwen Feng, @junyi42.bsky.social, @qianqianwang.bsky.social, Yufei Ye, Pengcheng Yu, @michael-j-black.bsky.social, Trevor Darrell, @akanazawa.bsky.social DUSt3R-like framework arxiv.org/abs/2504.13152

1104

Reposted by Michael J. Black

Meshcapade @meshcapade.com · 18/04/2025

Missed us at GDC? Watch Part 1 of our talk here 👉 youtu.be/0jCTiQMutow 🚶 Motion capture with MoCapade 🎮 Import directly into Unreal Engine 🎭 Retarget to any character 👀 Bonus: sneak peek at 3D hair & realtime single-cam mocap 🕴️ No suits. 📍 No markers. 🤳 Just one camera.

youtu.be

Meshcapade: From zero to game-ready assets in seconds (GDC 2025 Presentation)

YouTube video by Meshcapade

143

Michael J. Black @michael-j-black.bsky.social · 11/04/2025

Will there be an AI tariff? TL;DR: The societies that will "win" the AI race will not be those that develop the technology first. It will be those that are best able to manage the long-term social disruption AI will cause. perceiving-systems.blog/en/news/what... on Medium medium.com/@black_51980...

perceiving-systems.blog

The AI tariff?

3133

Reposted by Michael J. Black

Meshcapade @meshcapade.com · 28/03/2025

One shot. One town. Endless dancing. 🕺💃🎶 This one-take Unreal animation uses Meshcapade to bring every character to life. No mocap suits, just seamless motion from start to finish. Stylized, cinematic, and full of vibes; exactly how digital animation should feel 📹✨ #3DAnimation #MarkerlessMocap

0103

Michael J. Black @michael-j-black.bsky.social · 25/03/2025

The arXiv paper for PRIMAL is now online. It's a data-driven, interactive avatar, that can be controlled by varied commands, runs in a game engine, and is responsive to perturbations (without physics simulation). arxiv.org/abs/2503.17544

arxiv.org

PRIMAL: Physically Reactive and Interactive Motor Model for Avatar Learning

To build a motor system of the interactive avatar, it is essential to develop a generative motion model drives the body to move through 3D space in a perpetual, realistic, controllable, and responsive...

0150

Reposted by Michael J. Black

Meshcapade @meshcapade.com · 21/03/2025

👨‍🎤 Ever seen an Interactive Generative avatar running inside @unrealengine.bsky.social? Check out our latest work, PRIMAL, in collaboration with Max Planck Institute for Intelligent Systems & Stanford University - live demo at @officialgdc.bsky.social! 🎮 www.youtube.com/watch?v=-Gcp...

youtube.com

PRIMAL: Physically Reactive and Interactive Motor Model for Avatar Learning

YouTube video by Yan Zhang

163

Reposted by Michael J. Black

Artur Grigorev @arturgrigorev.bsky.social · 20/03/2025

🎉🎉🎉 Happy to announce that the code for our paper Gaussian Garments is now public! Link: github.com/eth-ait/Gaus... Gaussian Garments uses a combination of 3D meshes and Gaussian splatting to reconstruct photorealistic simulation-ready digital garments from multi-view videos. 🧵

1142

Michael J. Black @michael-j-black.bsky.social · 17/03/2025

The CameraHMR video is now on Youtube. This is currently the most accurate single-image method for estimating 3D human shape and pose. The paper will be presented at 3DV. The code and data are all on-line here: camerahmr.is.tue.mpg.de youtu.be/v3WzpjXpknc

youtu.be

CameraHMR: Aligning People with Perspective (3DV 2025)

YouTube video by Michael Black

0143

Reposted by Michael J. Black

Meshcapade @meshcapade.com · 13/03/2025

🎮 Motion Generation in Unreal, made SMPL. Experience real-time reactive behavior—characters adapt instantly to your input. With controllable generative 3D motion, every move is unique. Real-time motion blending keeps it smooth. 🚀 See it at Booth C1821! #3DAnimation #MotionGeneration #GameDev #GDC

093

Michael J. Black @michael-j-black.bsky.social · 05/03/2025

And it just keeps coming from @meshcapade.com -- next up, real-time markerless motion capture from a single camera. Check it out at #GDC2025. Yes, the founders will be there to dance for you (as in this video) but it's more fun to try it yourself!

070

Reposted by Michael J. Black

Meshcapade @meshcapade.com · 03/03/2025

Faces, Expressions, Hair—Brought to Life with Meshcapade. From facial animation to 3D hair, creating digital humans has never been easier. See it at #GDC2025! Visit Booth C1821 and experience motion, detail, expression. #UnrealEngine #3DAnimation #MotionCapture #FacialAnimation #genAI #Meshcapade

0105

Michael J. Black @michael-j-black.bsky.social · 19/02/2025

I'll be at #GDC2025 so if you are there and want to meet, you can probably find me at the @meshcapade.com booth.

031

Reposted by Michael J. Black

Meshcapade @meshcapade.com · 11/02/2025

Meshcapade was just featured again in DER SPIEGEL and their Spiegel-Shortcut podcast! 🎙️✨ The podcast highlights the diversity of ideas shaping the future - something we’re very proud to part of! ▶️ Watch on Youtube: www.youtube.com/watch?v=gN0X...

youtube.com

KI-Innovation: Hat Deutschland Antworten auf ChatGPT, DeepSeek & Co.? – Shortcut | DER SPIEGEL

YouTube video by DER SPIEGEL

182

Michael J. Black @michael-j-black.bsky.social · 06/02/2025

Turning videos into 3D humans has never been easier or more accurate. MoCapade 3.0 improves everything over 2.0 while adding 3D camera tracking (and output) and multi-person capture. Try out the new state-of-the-art. Of course, there's more to come.

0346

Reposted by Michael J. Black

Meshcapade @meshcapade.com · 05/02/2025

MoCapade 3.0 goes LIVE today 🥳 🎉 👯 Multi-person capture 🎥 3D camera motion 🙌 Detailed hands & gestures ⏫ New GLB, MP4 and SMPL export formats All from a single camera. Any camera! 📹 🤳 📸 Ready, set, CAPTURE! 🏂 youtu.be/jizULlZTAR8

youtu.be

MoCapade 3.0: Meshcapade's markerless mocap is live 🥳 🎉

YouTube video by Meshcapade

1115

Michael J. Black @michael-j-black.bsky.social · 31/01/2025

Thanks, but this version of CameraHMR estimates SMPL and not SMPL-X. The default in SMPL is flat hands. So we can't really take credit for this! For a commercial version with SMPL-X hands, try out meshcapade.me

010

Reposted by Michael J. Black

Naureen Mahmood @naureenm.bsky.social · 28/01/2025

"AI is like electricity: the decisive factor is not who invented it, but who knows how to use it best." ⚡

152

Reposted by Michael J. Black

Meshcapade @meshcapade.com · 28/01/2025

Such a great comparison video showing what's possible with @meshcapade.com using a single video compared to using a mocap suit. From 3D artist, Alejandro de Pasquale. www.youtube.com/watch?v=ess6...

youtube.com

Mocap Suit - Mocap Video - TEST (subtitles)

YouTube video by alerender

163

Reposted by Michael J. Black

Meshcapade @meshcapade.com · 28/01/2025

A fantastic article from @spiegel.de about the importance of the AI technologies we're building at Meshcapade & the wider German AI startup space, including groundbreaking work from startups like OSORA medical, EndoSurge, Mirelo AI, Maddox AI, Polybot. www.spiegel.de/wissenschaft...

spiegel.de

(S+) Künstliche Intelligenz: In Deutschland entsteht eine innovative KI-Szene

Künstliche Intelligenz erobert die Welt: Start-ups in Württemberg versuchen, auf dem Weltmarkt mitzumischen. Sie entwickeln ferngesteuerte Herzkatheter, KI-generierte Filmmusik und Agrarroboter.

152

Michael J. Black @michael-j-black.bsky.social · 21/01/2025

We iterate training CameraHMR and running CamSMPLify on the training set initialized with CameraHMR. This results in much improved pGT for 4DHumans and a SOTA single-image HMR method.

041

Michael J. Black @michael-j-black.bsky.social · 21/01/2025

4. But SMPLify only uses sparse 2D keypoints, which do not capture body shape. So we train a dense surface keypoint detector, DenseKP, on BEDLAM and run it on 4DHumans, resulting in improved body shape. The resulting method is CamSMPLify.

151

Michael J. Black @michael-j-black.bsky.social · 21/01/2025

2. We introduce CameraHMR, which integrates HumanFOV into HMR2.0 to exploit the estimated focal length. 3. To get accurate pseudo ground truth (pGT) training data, we compute the focal length for images in 4DHumans dataset and modify SMPLify to take this into account.

110

Michael J. Black @michael-j-black.bsky.social · 21/01/2025

There are 4 key contributions that make it so accurate and robust: 1. To get accurate 3D shape and pose as well as good alignment to image features, you need to know the focal length of the camera. To solve this, we train HumanFOV to compute the field of view.

110

Michael J. Black @michael-j-black.bsky.social · 21/01/2025

Code and data are now online for CameraHMR, our state-of-the-art parametric 3D human pose and shape (HPS) estimation method that will appear at hashtag#3DV2025. github.com/pixelite1201...

2313

Michael J. Black @michael-j-black.bsky.social · 18/01/2025

How can LLMs help computers see, how human motion is the dark matter of computer vision, what's artificial good-enough intelligence, & more? With Ben Kaplan at Deep Venture Partners. @meshcapade.com @mpi-is.bsky.social APPLE: lnkd.in/ez8cKuNs SPOTIFY: lnkd.in/eWtmtHd5

lnkd.in

This link will take you to a page that’s not on LinkedIn

2296

Reposted by Michael J. Black

Meshcapade @meshcapade.com · 15/01/2025

And now? From Mocapade to Motion Art! 🎥✨ Our ballerina's motion + #RunwayML = pure art! Just one export and endless ways to reimagine movement. Have you tried mixing Mocapade motions with your favorite creative tools? Show us your experiments! 🌟 #3DAvatars #DigitalCreation #genAI

083

Michael J. Black @michael-j-black.bsky.social · 04/01/2025

Here's a link to a recent talk that I gave at the University of Aukland Bioengineering Institute, entitled "Pixels to Skeletons: Towards Internet-Scale Biomechanics." t.co/c1LODMUL5t

t.co

https://cassyni.com/events/NTLKjJ1FF9HXipZwY3XnrZ/seminar/abstract

0154

Michael J. Black @michael-j-black.bsky.social · 29/12/2024

I’ve come to terms (mostly) with SVD. Life is change…

020

Michael J. Black @michael-j-black.bsky.social · 08/12/2024

The core problem of the Polar Express was that the mocap didn’t capture the nuances of the human motions. I’m guessing that Runway will improve the visuals but won’t fix the motions (yet). Though there does seem to be more natural facial motion in the “restoration”, which is interesting.

130

Michael J. Black @michael-j-black.bsky.social · 06/12/2024

We don’t have that kind of geometric accuracy nor is there any explicit alignment. For our early work on using 3d body models and scanners to track skin cancer, see files.is.tue.mpg.de/fbogo/papers...

files.is.tue.mpg.de

010

Michael J. Black @michael-j-black.bsky.social · 05/12/2024

Remember Photo Tourism? It reconstructed rigid scenes from a photo collection. Can we do the same for humans, creating an avatar from your personal photo album? Come see Yuliang Xiu present PuzzleAvatar today #SIGGRAPHAsia2024, 2:45-3:55 PM, Hall B7 (1). puzzleavatar.is.tue.mpg.de

1253

Reposted by Michael J. Black

Meshcapade @meshcapade.com · 04/12/2024

✨ Even higher accuracy for 3D body from images! Our newest "from image" model is here - delivering unmatched precision in 3D body shape and pose accuracy from any photo. #3DAvatars #DigitalCreation #genAI

133

Michael J. Black @michael-j-black.bsky.social · 04/12/2024

If I agree to do something (eg write a reference letter), the time needed to do it goes in my calendar. If can’t block the time, then I have to say no.

1100

Michael J. Black @michael-j-black.bsky.social · 04/12/2024

I know. Right? How can people live with themselves? ;)

030