Video wird geladen...

Video konnte nicht geladen werden

Beim Laden dieses Videos ist ein Problem aufgetreten. Dies könnte an einem vorübergehenden Netzwerkproblem liegen oder das Video ist möglicherweise nicht verfügbar.

our new model Atlas also happens to be a capable text-to-image generator, providing it with a strong foundation of “world knowledge”... this generalizes across both single and multi view domains, meaning you can step into even highly stylized scenes with precise 3D control

Ben Mildenhall

11,625 subscribers

103,214 Aufrufe • vor 10 Tagen •via X (Twitter)

Kunst Wissenschaft & Technologie Bildung

Anya Rossi• Live Now

Private livecam show

36 Kommentare

Profilbild von Keenan Crane

Keenan Cranevor 10 Tagen

Question purely out of curiosity: can you recover an explicit 3D representation (say, a textured mesh) from Atlas, such that renderings of the explicit representation closely reproduce the video (say, for example “hard surface” models and relatively simple materials)? This kind of question often gets side stepped (e.g., “meshes are old news”), but for me it’s a scientific question rather than an application question: what information is the model predicting, and which variables are still underdetermined?

Profilbild von Thariq

Thariqvor 10 Tagen

Incredible, congrats Ben!

Profilbild von Mazeyar

Mazeyarvor 10 Tagen

Can you let me play GTA VI in Atlas 😌🙏🏼

Profilbild von Ben Mildenhall

Ben Mildenhallvor 10 Tagen

yes

Profilbild von Mazeyar

Mazeyarvor 10 Tagen

Hahaha thanks 🤣

Profilbild von Mathew Tizard

Mathew Tizardvor 10 Tagen

Incredible work. If I have a variety of viewpoints of the same scene, can Atlas ingest all that ground truth and use it to synthesise the missing perspectives? Would this work with synchronised video frames too?

Profilbild von Ben Mildenhall

Ben Mildenhallvor 9 Tagen

exactly!

Profilbild von POM

POMvor 10 Tagen

Beautiful style adherence! The fabric on the octopus is amazing

Profilbild von bone

bonevor 10 Tagen

very cool, can you show it doing a couple 360s? would like to see if it keeps persistence

Profilbild von Ian Curtis

Ian Curtisvor 10 Tagen

this one 🫠 so good

Profilbild von Nicholas Bardy

Nicholas Bardyvor 10 Tagen

nice work

Profilbild von Troy Kirwin

Troy Kirwinvor 10 Tagen

Sick

Profilbild von XGEN Labs

XGEN Labsvor 4 Tagen

The move from image generation to controllable 3D worlds is especially exciting. Curious how far this kind of world knowledge can generalize across unseen environments and viewpoints.

Profilbild von Doug Thompson

Doug Thompsonvor 10 Tagen

I was waiting for this. Marble etc were awesome but I can't wait to experiment. Working on history projects (London UK, really top people) and we have such deep ideas but needed something like Atlas to get us there. Storytelling has a new tool kit and you guys rocking it. EPIC

Profilbild von Yücel İnanoğulları

Yücel İnanoğullarıvor 9 Tagen

Very impressive. I wonder if one day, AI would read a novel and produce a cinematic version of it.

Profilbild von Nicholas H

Nicholas Hvor 10 Tagen

wow that 2D japanese scene though

Profilbild von ritwik

ritwikvor 10 Tagen

Takes one back to NeRF days

Profilbild von Amara Wolfe

Amara Wolfevor 10 Tagen

Claiming Atlas has 'world knowledge' from 2D text-to-image training is a massive stretch. How does it handle actual physical consistency and occlusion when you try to navigate those stylized scenes?

Profilbild von Peachy J

Peachy Jvor 10 Tagen

That world knowledge foundation generalizing across views is clever.

Profilbild von COSMOS

COSMOSvor 9 Tagen

wf

Profilbild von man ki Hui

man ki Huivor 9 Tagen

Great 👍🏻

Profilbild von Jason Farris

Jason Farrisvor 10 Tagen

@venturetwins How do we get started using this?

Profilbild von Pablo Antonio

Pablo Antoniovor 10 Tagen

very cool!

Profilbild von THC Humor 💹🧲

THC Humor 💹🧲vor 9 Tagen

World knowledge plus precise 3D control is a powerful combination

Profilbild von Nathan Quantum

Nathan Quantumvor 9 Tagen

you did more for open research with the original nerf code release than most labs manage in a decade. really hoping atlas follows the same path

Profilbild von Shawn Neal

Shawn Nealvor 9 Tagen

So in the future, video editing will become world editing? like world editor in RTS game?

Profilbild von Neek

Neekvor 10 Tagen

nice!

Profilbild von Manaen

Manaenvor 9 Tagen

The cinematography and real world simulation is unmatched!

Profilbild von 김란지

김란지vor 9 Tagen

😍😍😍😍

Profilbild von Alex AI Home

Alex AI Homevor 9 Tagen

Precise 3D control across single and multi-view domains solves the biggest headache in AI home transformation workflows. When passing raw balcony photos into Atlas, does the underlying world knowledge lock structural dimensions (like railings and glass planes) when pulling dynamic camera orbits?

Profilbild von AI Mastery Guide

AI Mastery Guidevor 10 Tagen

Precise 3D control from a side feature, wow

Profilbild von Logerhaus

Logerhausvor 10 Tagen

Yep were in a simulation

Profilbild von kaen_

kaen_vor 9 Tagen

The 3D consistency here is really impressive. Feels like a meaningful step toward truly controllable world models

Profilbild von Callum Marley

Callum Marleyvor 10 Tagen

Extremely cool, congrats to the team

Profilbild von Owx

Owxvor 9 Tagen

Does that work with anime ?

Profilbild von Ben

Benvor 9 Tagen

when we revisit the same place twice are the details identical or it's being generated differently? How long does the same details persist? I think I can find some fun use cases with this new tool

Ähnliche Videos

MVDream: Multi-view Diffusion for 3D Generation paper page: propose MVDream, a multi-view diffusion model that is able to generate geometrically consistent multi-view images from a given text prompt. By leveraging image diffusion models pre-trained on large-scale web datasets and a multi-view dataset rendered from 3D assets, the resulting multi-view diffusion model can achieve both the generalizability of 2D diffusion and the consistency of 3D data. Such a model can thus be applied as a multi-view prior for 3D generation via Score Distillation Sampling, where it greatly improves the stability of existing 2D-lifting methods by solving the 3D consistency problem. Finally, we show that the multi-view diffusion model can also be fine-tuned under a few shot setting for personalized 3D generation, i.e. DreamBooth3D application, where the consistency can be maintained after learning the subject identity.

MVDream: Multi-view Diffusion for 3D Generation paper page: propose MVDream, a multi-view diffusion model that is able to generate geometrically consistent multi-view images from a given text prompt. By leveraging image diffusion models pre-trained on large-scale web datasets and a multi-view dataset rendered from 3D assets, the resulting multi-view diffusion model can achieve both the generalizability of 2D diffusion and the consistency of 3D data. Such a model can thus be applied as a multi-view prior for 3D generation via Score Distillation Sampling, where it greatly improves the stability of existing 2D-lifting methods by solving the 3D consistency problem. Finally, we show that the multi-view diffusion model can also be fine-tuned under a few shot setting for personalized 3D generation, i.e. DreamBooth3D application, where the consistency can be maintained after learning the subject identity.

AK

294,519 Aufrufe • vor 3 Jahren

Today, we launch Viggle V4 + Character Refine. - Strong character consistency with precise motion control - Upload 1 image → auto-generate multi-angle references - 1:1 motion transfer from reference videos - Multi-character swap in complex scenes - Generate in SECONDS at lower cost - Up to 60s videos Powered by our proprietary 3D foundation model, built for speed, reliability, and full creative control. Try it for free! 👇

Today, we launch Viggle V4 + Character Refine. - Strong character consistency with precise motion control - Upload 1 image → auto-generate multi-angle references - 1:1 motion transfer from reference videos - Multi-character swap in complex scenes - Generate in SECONDS at lower cost - Up to 60s videos Powered by our proprietary 3D foundation model, built for speed, reliability, and full creative control. Try it for free! 👇

ViggleAI

400,530 Aufrufe • vor 6 Monaten

📹The Hailuo I2V-01-Director Model is now available to everyone! Experience cinematic storytelling with precise camera control—powered by both our text and image models. Try it here:

📹The Hailuo I2V-01-Director Model is now available to everyone! Experience cinematic storytelling with precise camera control—powered by both our text and image models. Try it here:

Hailuo AI (MiniMax)

319,991 Aufrufe • vor 1 Jahr

Introducing Frames: An image generation model offering unprecedented stylistic control. Frames is our newest foundation model for image generation, marking a big step forward in stylistic control and visual fidelity. With Frames, you can begin to architect worlds that represent very specific points of view and aesthetic characteristics. See below for examples. World 1089: Mise-en-scène (1/11)

Introducing Frames: An image generation model offering unprecedented stylistic control. Frames is our newest foundation model for image generation, marking a big step forward in stylistic control and visual fidelity. With Frames, you can begin to architect worlds that represent very specific points of view and aesthetic characteristics. See below for examples. World 1089: Mise-en-scène (1/11)

Runway

263,530 Aufrufe • vor 1 Jahr

🚨 Google releases its most advanced world model Genie 3. It lets you build fully interactive 3D worlds you can explore in real-time with a text prompt, or even upload your own image, and turning it into your own playable world.

🚨 Google releases its most advanced world model Genie 3. It lets you build fully interactive 3D worlds you can explore in real-time with a text prompt, or even upload your own image, and turning it into your own playable world.

Alvaro Cintas

23,059 Aufrufe • vor 7 Monaten

Rolling out today we are launching Nano Banana Pro, the world’s best image model built to move beyond casual creation and into a new era of studio-quality, functional design. Nano Banana Pro enables a new level of precision and creative control, transforming the way you bring ideas to life. Here are a couple of our favorite new features: — Text rendering and translation: Generate crystal-clear text directly within your images. With the model’s advanced language understanding, you can even translate and regenerate visuals with localized text. — World knowledge: By connecting to Search’s vast knowledge base, Nano Banana Pro generates factually accurate diagrams and realistic product placements, making it an invaluable tool for learning and communication.

Rolling out today we are launching Nano Banana Pro, the world’s best image model built to move beyond casual creation and into a new era of studio-quality, functional design. Nano Banana Pro enables a new level of precision and creative control, transforming the way you bring ideas to life. Here are a couple of our favorite new features: — Text rendering and translation: Generate crystal-clear text directly within your images. With the model’s advanced language understanding, you can even translate and regenerate visuals with localized text. — World knowledge: By connecting to Search’s vast knowledge base, Nano Banana Pro generates factually accurate diagrams and realistic product placements, making it an invaluable tool for learning and communication.

Google AI

337,937 Aufrufe • vor 9 Monaten

#ICCV2025 🤩3D world generation is cool, but it is cooler to play with the worlds using 3D actions 👆💨, and see what happens! — Introducing *WonderPlay*: Now you can create dynamic 3D scenes that respond to your 3D actions from a single image! Web: 🧵1/7

#ICCV2025 🤩3D world generation is cool, but it is cooler to play with the worlds using 3D actions 👆💨, and see what happens! — Introducing WonderPlay: Now you can create dynamic 3D scenes that respond to your 3D actions from a single image! Web: 🧵1/7

Hong-Xing (Koven) Yu

57,951 Aufrufe • vor 1 Jahr

V3D Video Diffusion Models are Effective 3D Generators Automatic 3D generation has recently attracted widespread attention. Recent methods have greatly accelerated the generation speed, but usually produce less-detailed objects due to limited model capacity or 3D data. Motivated by recent advancements in video diffusion models, we introduce V3D, which leverages the world simulation capacity of pre-trained video diffusion models to facilitate 3D generation. To fully unleash the potential of video diffusion to perceive the 3D world, we further introduce geometrical consistency prior and extend the video diffusion model to a multi-view consistent 3D generator. Benefiting from this, the state-of-the-art video diffusion model could be fine-tuned to generate 360degree orbit frames surrounding an object given a single image. With our tailored reconstruction pipelines, we can generate high-quality meshes or 3D Gaussians within 3 minutes. Furthermore, our method can be extended to scene-level novel view synthesis, achieving precise control over the camera path with sparse input views. Extensive experiments demonstrate the superior performance of the proposed approach, especially in terms of generation quality and multi-view consistency

V3D Video Diffusion Models are Effective 3D Generators Automatic 3D generation has recently attracted widespread attention. Recent methods have greatly accelerated the generation speed, but usually produce less-detailed objects due to limited model capacity or 3D data. Motivated by recent advancements in video diffusion models, we introduce V3D, which leverages the world simulation capacity of pre-trained video diffusion models to facilitate 3D generation. To fully unleash the potential of video diffusion to perceive the 3D world, we further introduce geometrical consistency prior and extend the video diffusion model to a multi-view consistent 3D generator. Benefiting from this, the state-of-the-art video diffusion model could be fine-tuned to generate 360degree orbit frames surrounding an object given a single image. With our tailored reconstruction pipelines, we can generate high-quality meshes or 3D Gaussians within 3 minutes. Furthermore, our method can be extended to scene-level novel view synthesis, achieving precise control over the camera path with sparse input views. Extensive experiments demonstrate the superior performance of the proposed approach, especially in terms of generation quality and multi-view consistency

AK

31,997 Aufrufe • vor 2 Jahren

📢📢 𝐏𝐞𝐫𝐜𝐇𝐞𝐚𝐝: 𝐏𝐞𝐫𝐜𝐞𝐩𝐭𝐮𝐚𝐥 𝐇𝐞𝐚𝐝 𝐌𝐨𝐝𝐞𝐥 𝐟𝐨𝐫 𝐒𝐢𝐧𝐠𝐥𝐞-𝐈𝐦𝐚𝐠𝐞 𝟑𝐃 𝐇𝐞𝐚𝐝 𝐑𝐞𝐜𝐨𝐧𝐬𝐭𝐫𝐮𝐜𝐭𝐢𝐨𝐧 & 𝐄𝐝𝐢𝐭𝐢𝐧𝐠📢📢 PercHead reconstructs realistic 3D heads from a single image and enables disentangled 3D editing via geometric controls and style inputs from images or text. At its core is a generalized 3D head decoder trained with perceptual supervision from DINOv2 and SAM 2.1. We find that our new perceptual loss formulation improves reconstruction fidelity compared to commonly-used methods such as LPIPS. Our trained reconstruction model is able to generate 3D-consistent heads from a single input image. Even with challenging side-view inputs, the model robustly infers missing regions for a coherent, high-fidelity output. In addition, our architecture seamlessly adapts to downstream tasks: by swapping the encoder, we can transform the model into a disentangled 3D editing pipeline. In this scenario, we can control geometry through - potentially hand-drawn - segmentation maps, and condition style via image or text prompt. We also provide an interactive GUI to enable the exploration of our editing pipeline. 🌍 📽️ Great work by Antonio Oroz and Tobias Kirschstein

📢📢 𝐏𝐞𝐫𝐜𝐇𝐞𝐚𝐝: 𝐏𝐞𝐫𝐜𝐞𝐩𝐭𝐮𝐚𝐥 𝐇𝐞𝐚𝐝 𝐌𝐨𝐝𝐞𝐥 𝐟𝐨𝐫 𝐒𝐢𝐧𝐠𝐥𝐞-𝐈𝐦𝐚𝐠𝐞 𝟑𝐃 𝐇𝐞𝐚𝐝 𝐑𝐞𝐜𝐨𝐧𝐬𝐭𝐫𝐮𝐜𝐭𝐢𝐨𝐧 & 𝐄𝐝𝐢𝐭𝐢𝐧𝐠📢📢 PercHead reconstructs realistic 3D heads from a single image and enables disentangled 3D editing via geometric controls and style inputs from images or text. At its core is a generalized 3D head decoder trained with perceptual supervision from DINOv2 and SAM 2.1. We find that our new perceptual loss formulation improves reconstruction fidelity compared to commonly-used methods such as LPIPS. Our trained reconstruction model is able to generate 3D-consistent heads from a single input image. Even with challenging side-view inputs, the model robustly infers missing regions for a coherent, high-fidelity output. In addition, our architecture seamlessly adapts to downstream tasks: by swapping the encoder, we can transform the model into a disentangled 3D editing pipeline. In this scenario, we can control geometry through - potentially hand-drawn - segmentation maps, and condition style via image or text prompt. We also provide an interactive GUI to enable the exploration of our editing pipeline. 🌍 📽️ Great work by Antonio Oroz and Tobias Kirschstein

Matthias Niessner

18,899 Aufrufe • vor 10 Monaten

We've officially released and open-sourced HunyuanImage 2.1, our latest text-to-image model. The new model delivers on our commitment to balancing performance and quality. With native 2K image generation, HunyuanImage 2.1 is an advanced open-source text-to-image model.🎨 ✨ New in 2.1: 🔹Advanced Semantics: Supports ultra-long and complex prompts of up to 1000 tokens, and precisely controls the generation of multiple subjects in a single image. 🔹Precise Chinese and English Text Rendering with seamless image–text integration: The model naturally integrates text into images, making it suitable for a wide range of applications such as product covers, illustrations, and poster design to meet the needs of various fields. 🔹Rich Styles and High Aesthetic: Capable of generating images in various styles—including photorealistic portraits, comics, and vinyl figures—it delivers outstanding visual appeal and artistic quality. 🔹High-Quality Generation: Efficiently produces ultra-high-definition (2K) images in the same time other models take to generate a 1K image. HunyuanImage 2.1 uses two text encoders: a multimodal large language model (MLLM) to improve the model's image and text alignment capabilities, and a multi-language character-aware encoder to improve text rendering capabilities. The model is a single- and double-stream diffusion transformer with 17B parameters. We've also open-sourced the weights of the the accelerated version with meanflow which reduces inference steps from 100 to just 8, and PromptEnhancer, the first industrial-grade rewriting model that enhances your prompts for more nuanced and expressive image generation. Now, creators turn complex ideas—like posters with slogans or multi-panel comics—into visuals faster than ever. We’re just getting started. Stay tuned for our native multimodal image generation model coming soon. 🌐Website: 🔗Github: 🤗Hugging Face: ✨Hugging Face Demo:

We've officially released and open-sourced HunyuanImage 2.1, our latest text-to-image model. The new model delivers on our commitment to balancing performance and quality. With native 2K image generation, HunyuanImage 2.1 is an advanced open-source text-to-image model.🎨 ✨ New in 2.1: 🔹Advanced Semantics: Supports ultra-long and complex prompts of up to 1000 tokens, and precisely controls the generation of multiple subjects in a single image. 🔹Precise Chinese and English Text Rendering with seamless image–text integration: The model naturally integrates text into images, making it suitable for a wide range of applications such as product covers, illustrations, and poster design to meet the needs of various fields. 🔹Rich Styles and High Aesthetic: Capable of generating images in various styles—including photorealistic portraits, comics, and vinyl figures—it delivers outstanding visual appeal and artistic quality. 🔹High-Quality Generation: Efficiently produces ultra-high-definition (2K) images in the same time other models take to generate a 1K image. HunyuanImage 2.1 uses two text encoders: a multimodal large language model (MLLM) to improve the model's image and text alignment capabilities, and a multi-language character-aware encoder to improve text rendering capabilities. The model is a single- and double-stream diffusion transformer with 17B parameters. We've also open-sourced the weights of the the accelerated version with meanflow which reduces inference steps from 100 to just 8, and PromptEnhancer, the first industrial-grade rewriting model that enhances your prompts for more nuanced and expressive image generation. Now, creators turn complex ideas—like posters with slogans or multi-panel comics—into visuals faster than ever. We’re just getting started. Stay tuned for our native multimodal image generation model coming soon. 🌐Website: 🔗Github: 🤗Hugging Face: ✨Hugging Face Demo:

Tencent Hy

89,257 Aufrufe • vor 1 Jahr

Seedance 2.5 has landed on Luma. 🚀 Create up to 30 seconds of multi-shot video, high-definition in a single generation using up to 50 references (text, image, video, and audio). With Luma Agents, you can control scenes, camera angles, and pacing to produce your next masterpiece.

Seedance 2.5 has landed on Luma. 🚀 Create up to 30 seconds of multi-shot video, high-definition in a single generation using up to 50 references (text, image, video, and audio). With Luma Agents, you can control scenes, camera angles, and pacing to produce your next masterpiece.

Luma

26,457 Aufrufe • vor 1 Monat

Introducing LTX-2.5, the world model the world builds on. One of the biggest upgrades yet to the model already powering film, robotics, and real-time workflows. Higher pixel fidelity, multishot scenes that hold together across cuts, and a pretrained foundation built to be finetuned across domains. Plus: Diffusion Fidelity Rendering, a new rendering approach built for pixel quality that holds up frame by frame, even on a cinema screen. This isn't a tool you rent. It's a foundation you build on, open and yours.

Introducing LTX-2.5, the world model the world builds on. One of the biggest upgrades yet to the model already powering film, robotics, and real-time workflows. Higher pixel fidelity, multishot scenes that hold together across cuts, and a pretrained foundation built to be finetuned across domains. Plus: Diffusion Fidelity Rendering, a new rendering approach built for pixel quality that holds up frame by frame, even on a cinema screen. This isn't a tool you rent. It's a foundation you build on, open and yours.

LTX.io

13,726,933 Aufrufe • vor 1 Monat

Designing an Encoder for Fast Personalization of Text-to-Image Models TL;DR: use an encoder to personalize a text-to-image model to new concepts with a single image and 5-15 tuning steps abs: project page:

Designing an Encoder for Fast Personalization of Text-to-Image Models TL;DR: use an encoder to personalize a text-to-image model to new concepts with a single image and 5-15 tuning steps abs: project page:

AK

165,168 Aufrufe • vor 3 Jahren

GPT-6 coded an entire world: a busy harbor you can step inside. From a single prompt and an asset pack, it assembled a container port with towering cranes, docked ships, rows of containers, and forklifts across the yard, all within an explorable 3D scene. A first look at our upcoming Code-for-3D benchmark, featuring the text-to-scene setting.

GPT-6 coded an entire world: a busy harbor you can step inside. From a single prompt and an asset pack, it assembled a container port with towering cranes, docked ships, rows of containers, and forklifts across the yard, all within an explorable 3D scene. A first look at our upcoming Code-for-3D benchmark, featuring the text-to-scene setting.

SimWorld

23,957 Aufrufe • vor 2 Tagen

Introducing Cube 2.0, a significant step towards controllable 3D game-world and video generation from any input. Powered by: 🌟 Image-to-3D in seconds w/ a new foundation model 🐎 Text-to-animation 🌎 AI rendering to create beautiful worlds (mesh+splats) Web app → Discord → Blog →

Introducing Cube 2.0, a significant step towards controllable 3D game-world and video generation from any input. Powered by: 🌟 Image-to-3D in seconds w/ a new foundation model 🐎 Text-to-animation 🌎 AI rendering to create beautiful worlds (mesh+splats) Web app → Discord → Blog →

Common Sense Machines

411,413 Aufrufe • vor 2 Jahren