正在加载视频...

视频加载失败

加载此视频时出现问题。这可能是由于临时网络问题，或视频可能不可用。

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... show more

Xiaochuang Han

1,107 subscribers

32,818 次观看 • 10 个月前 •via X (Twitter)

Anya Rossi• Live Now

Private livecam show

10 条评论

Xiaochuang Han 的头像

Xiaochuang Han10 个月前

TV2TV builds on Transfusion, LMFusion, and mixture-of-Transformers, but goes further by enabling interleaved text-video co-generation in a single model. It models text autoregressively and video in short chunks semi-autoregressively (flow matching within ~0.25s clips, autoregressive globally). Each text token or video chunk attends to all previous text tokens and clean video chunks.

Xiaochuang Han 的头像

Xiaochuang Han10 个月前

During inference, TV2TV dynamically alternates between planning in text and generating chunks of video frames, letting the model "think in words" about upcoming content before "acting in pixels" to produce those frames. This framework enables fine-grained and flexible control during generation, allowing users to intervene and steer the video trajectory at any point via text prompts.

Xiaochuang Han 的头像

Xiaochuang Han10 个月前

We first see the advantages of TV2TV in domains where ground-truth interleaved text is well defined. In controlled CS:GO (gaming) video generation experiments, TV2TV substantially outperforms competing approaches in both visual quality - preferred 91% of the time over a comparable T2V baseline - and controllability, with a 19-point improvement in fine-grained instruction-following accuracy over a "think-then-act" (Think2V) method that generates detailed text plans before video.

Xiaochuang Han 的头像

Xiaochuang Han10 个月前

Model generation below: TV2TV rolls out a world and plays within it in an interleaved manner, all within a single unified model.

Xiaochuang Han 的头像

Xiaochuang Han10 个月前

To scale TV2TV to real-world videos, which typically lack temporally aligned text, we design a pipeline that augments natural videos with interleaved language using VLMs. We curate 8K hours of interleaved text-video data in the sports domain - chosen for its dynamic motion and rich actions. Training TV2TV on this corpus yields strong generation quality relative to both established external video models and controlled T2V and Think2V baselines. See the paper for details! 📖

Xiaochuang Han 的头像

Xiaochuang Han10 个月前

Joint work with an amazing group of core contributors, authors, and collaborators at Meta FAIR! - @XiaochuangHan Youssef Emad @hall__melissa @__JohnNguyen__ Karthik Padthe Liam Robbins @_amirbar @Delong0_0 Michal Drozdzal @melbayad @huyushi98 @ShangwenLi1 @berylSreya Jakob Verbeek @XDWang101 @gh_marjan @LukeZettlemoyer @em_dinan

Miguel Twahirwa 的头像

Miguel Twahirwa10 个月前

interesting work, any insights on generating audio directly (video to speech), would that require significant changes to the current architecture?

Xiaochuang Han 的头像

Xiaochuang Han10 个月前

my hunch is that adding another audio tower will do. depending on your audio tokenizer (continuous or discrete), it could be very similar to video (diffusion) or text (autoregressive).

Aditya Kusupati 的头像

Aditya Kusupati10 个月前

Nice work Han!

robertus 的头像

robertus10 个月前

the autoregressive text decomposition is clever. treating video generation as iterative refinement through language bottlenecks forces the model to learn compositional structure

相关视频

New Kling, Runway, Luma competitor? Open text and image video generation model Pyramidal Flow Matching for Efficient Video Generative Modeling

New Kling, Runway, Luma competitor? Open text and image video generation model Pyramidal Flow Matching for Efficient Video Generative Modeling

AK

91,250 次观看 • 2 年前

Images, video, audio, actions — generative modeling has converged on one recipe: compress into continuous latents, generate in latent space, decode. Language is the lone exception: still generated token by token, as a long discrete stream. Latent Thought Flows: We compress 256 text tokens into 8 continuous latents, generate them with a one-step flow model, and read them out as text using an autoregressive decoder. Result: a better inference-compute vs. generation-quality Pareto frontier than a tuned autoregressive baseline — thanks to compression and one-step generation. Co-led w/ Zhengyang Geng 🧵

Images, video, audio, actions — generative modeling has converged on one recipe: compress into continuous latents, generate in latent space, decode. Language is the lone exception: still generated token by token, as a long discrete stream. Latent Thought Flows: We compress 256 text tokens into 8 continuous latents, generate them with a one-step flow model, and read them out as text using an autoregressive decoder. Result: a better inference-compute vs. generation-quality Pareto frontier than a tuned autoregressive baseline — thanks to compression and one-step generation. Co-led w/ Zhengyang Geng 🧵

Mihir Prabhudesai

83,120 次观看 • 3 个月前

Our research team just released Flex-Forcing: a video generation method that lets a single model switch between generation methods at inference time. Right now there are two main approaches to video generation. Bidirectional diffusion models attend to every frame at once, holding structure well at the cost of speed. Autoregressive models generate frame by frame, so they stream fast and scale to long clips, but accumulate error and drift over time. Flex-Forcing trains a single model to do both, letting you choose from the range at inference based on your compute budget.

Our research team just released Flex-Forcing: a video generation method that lets a single model switch between generation methods at inference time. Right now there are two main approaches to video generation. Bidirectional diffusion models attend to every frame at once, holding structure well at the cost of speed. Autoregressive models generate frame by frame, so they stream fast and scale to long clips, but accumulate error and drift over time. Flex-Forcing trains a single model to do both, letting you choose from the range at inference based on your compute budget.

NVIDIA AI

31,412 次观看 • 3 个月前

We introduce W.A.L.T, a diffusion model for photorealistic video generation. Our model is a transformer trained on image and video generation in a shared latent space. 🧵👇

We introduce W.A.L.T, a diffusion model for photorealistic video generation. Our model is a transformer trained on image and video generation in a shared latent space. 🧵👇

Agrim Gupta

431,392 次观看 • 2 年前

ByteDance just announced MagicVideo-V2 Multi-Stage High-Aesthetic Video Generation paper page: The growing demand for high-fidelity video generation from textual descriptions has catalyzed significant research in this field. In this work, we introduce MagicVideo-V2 that integrates the text-to-image model, video motion generator, reference image embedding module and frame interpolation module into an end-to-end video generation pipeline. Benefiting from these architecture designs, MagicVideo-V2 can generate an aesthetically pleasing, high-resolution video with remarkable fidelity and smoothness. It demonstrates superior performance over leading Text-to-Video systems such as Runway, Pika 1.0, Morph, Moon Valley and Stable Video Diffusion model via user evaluation at large scale

ByteDance just announced MagicVideo-V2 Multi-Stage High-Aesthetic Video Generation paper page: The growing demand for high-fidelity video generation from textual descriptions has catalyzed significant research in this field. In this work, we introduce MagicVideo-V2 that integrates the text-to-image model, video motion generator, reference image embedding module and frame interpolation module into an end-to-end video generation pipeline. Benefiting from these architecture designs, MagicVideo-V2 can generate an aesthetically pleasing, high-resolution video with remarkable fidelity and smoothness. It demonstrates superior performance over leading Text-to-Video systems such as Runway, Pika 1.0, Morph, Moon Valley and Stable Video Diffusion model via user evaluation at large scale

AK

181,049 次观看 • 2 年前

StreamDiT Real-Time Streaming Text-to-Video Generation StreamDiT enables real-time text-to-video generation at 16 FPS on a single GPU (H100)

StreamDiT Real-Time Streaming Text-to-Video Generation StreamDiT enables real-time text-to-video generation at 16 FPS on a single GPU (H100)

AK

43,694 次观看 • 1 年前

🚀 Introducing Emu3.5 — a large-scale multimodal world model that natively predicts the next vision-language state. 🔥 Trained on over 10T interleaved vision-language tokens and enhanced with reinforcement learning, Emu3.5 achieves powerful multimodal reasoning and generation. ⚡ Powered by our new Discrete Diffusion Adaptation (DiDA) for 20× faster inference. 🔥 Emu3.5 outperforms Nano Banana across image generation, editing, interleaved tasks and more. 🌍 Explore Emu3.5: Github: #Emu3 #MultimodalAI #WorldModel #NextTokenPrediction

🚀 Introducing Emu3.5 — a large-scale multimodal world model that natively predicts the next vision-language state. 🔥 Trained on over 10T interleaved vision-language tokens and enhanced with reinforcement learning, Emu3.5 achieves powerful multimodal reasoning and generation. ⚡ Powered by our new Discrete Diffusion Adaptation (DiDA) for 20× faster inference. 🔥 Emu3.5 outperforms Nano Banana across image generation, editing, interleaved tasks and more. 🌍 Explore Emu3.5: Github: #Emu3 #MultimodalAI #WorldModel #NextTokenPrediction

BAAI

51,880 次观看 • 1 年前

Make Pixels Dance: High-Dynamic Video Generation paper page: Creating high-dynamic videos such as motion-rich actions and sophisticated visual effects poses a significant challenge in the field of artificial intelligence. Unfortunately, current state-of-the-art video generation methods, primarily focusing on text-to-video generation, tend to produce video clips with minimal motions despite maintaining high fidelity. We argue that relying solely on text instructions is insufficient and suboptimal for video generation. In this paper, we introduce PixelDance, a novel approach based on diffusion models that incorporates image instructions for both the first and last frames in conjunction with text instructions for video generation. Comprehensive experimental results demonstrate that PixelDance trained with public data exhibits significantly better proficiency in synthesizing videos with complex scenes and intricate motions, setting a new standard for video generation.

Make Pixels Dance: High-Dynamic Video Generation paper page: Creating high-dynamic videos such as motion-rich actions and sophisticated visual effects poses a significant challenge in the field of artificial intelligence. Unfortunately, current state-of-the-art video generation methods, primarily focusing on text-to-video generation, tend to produce video clips with minimal motions despite maintaining high fidelity. We argue that relying solely on text instructions is insufficient and suboptimal for video generation. In this paper, we introduce PixelDance, a novel approach based on diffusion models that incorporates image instructions for both the first and last frames in conjunction with text instructions for video generation. Comprehensive experimental results demonstrate that PixelDance trained with public data exhibits significantly better proficiency in synthesizing videos with complex scenes and intricate motions, setting a new standard for video generation.

AK

102,824 次观看 • 2 年前

The first truly open-source audio-video model. LTX-2 is a DiT-based foundation model with all core video generation capabilities in one unified model. Designed to run locally on consumer GPUs. - text-to-video - image-to-video - and video-to-video modes 100% open-source.

The first truly open-source audio-video model. LTX-2 is a DiT-based foundation model with all core video generation capabilities in one unified model. Designed to run locally on consumer GPUs. - text-to-video - image-to-video - and video-to-video modes 100% open-source.

Akshay 🚀

66,268 次观看 • 9 个月前

A breakthrough in real-time video generation. As a research preview developed with NVIDIA and shared at NVIDIAGTC this week, we trained a new real-time video model running on Vera Rubin. HD videos generate instantly, with time-to-first-frame under 100ms. Unlocking an entirely new creative paradigm and bolstering the foundations of our General World Model, GWM-1. Real-time generation opens a fundamentally different design space for video models and world simulation. We're investing in co-designing our models alongside advances in hardware to keep pushing this frontier.

A breakthrough in real-time video generation. As a research preview developed with NVIDIA and shared at NVIDIAGTC this week, we trained a new real-time video model running on Vera Rubin. HD videos generate instantly, with time-to-first-frame under 100ms. Unlocking an entirely new creative paradigm and bolstering the foundations of our General World Model, GWM-1. Real-time generation opens a fundamentally different design space for video models and world simulation. We're investing in co-designing our models alongside advances in hardware to keep pushing this frontier.

Runway

1,166,467 次观看 • 7 个月前

New text and image to video generation AI model Open-Sora-Plan-v1.3.0

New text and image to video generation AI model Open-Sora-Plan-v1.3.0

AK

51,843 次观看 • 2 年前

🚀 We open-sourced LongLive — interactive, real-time long-video generation. 👥Generates video in real time as users enter text prompts. ⚡️20.7 FPS on a single H100,⏱️up to 240s per clip. 🎬Fine-tunes SOTA short-video models (e.g., Wan) into long-video generators. 🌍One step closer to World Models. All code for training & inference, model weights, demo page, and videos released! Paper: Code: Model: Demo Page: Introduction Video:

🚀 We open-sourced LongLive — interactive, real-time long-video generation. 👥Generates video in real time as users enter text prompts. ⚡️20.7 FPS on a single H100,⏱️up to 240s per clip. 🎬Fine-tunes SOTA short-video models (e.g., Wan) into long-video generators. 🌍One step closer to World Models. All code for training & inference, model weights, demo page, and videos released! Paper: Code: Model: Demo Page: Introduction Video:

Yukang Chen

11,943 次观看 • 1 年前

1/4 🚀We are launching Qwen-Image-2.0, a next-generation foundational image generation model. The key highlights of Qwen-Image-2.0 include: Professional Typography Rendering: Supports 1k-token instructions for direct generation of professional infographics, including PPTs, posters, comics, and more. Stronger Semantic Adherence: Native 2K resolution support for finely detailed realistic scenes, including people, nature, and architecture. Improved Text Rendering: Integrated understanding and generation capabilities, unifying image generation and editing in a single mode Lighter Model Architecture: Smaller model size with faster inference speed.

1/4 🚀We are launching Qwen-Image-2.0, a next-generation foundational image generation model. The key highlights of Qwen-Image-2.0 include: Professional Typography Rendering: Supports 1k-token instructions for direct generation of professional infographics, including PPTs, posters, comics, and more. Stronger Semantic Adherence: Native 2K resolution support for finely detailed realistic scenes, including people, nature, and architecture. Improved Text Rendering: Integrated understanding and generation capabilities, unifying image generation and editing in a single mode Lighter Model Architecture: Smaller model size with faster inference speed.

Tongyi Lab

164,097 次观看 • 8 个月前

Gen2Act: Casting language-conditioned manipulation as *human video generation* followed by *closed-loop policy execution conditioned on the generated video* enables solving diverse real-world tasks unseen in the robot dataset! 1/n

Gen2Act: Casting language-conditioned manipulation as human video generation followed by closed-loop policy execution conditioned on the generated video enables solving diverse real-world tasks unseen in the robot dataset! 1/n

Homanga Bharadhwaj

71,305 次观看 • 2 年前

Meta announces Movie Gen A Cast of Media Foundation Models We present Movie Gen, a cast of foundation models that generates high-quality, 1080p HD videos with different aspect ratios and synchronized audio. We also show additional capabilities such as precise instruction-based video editing and generation of personalized videos based on a user’s image. Our models set a new state-of-the-art on multiple tasks: text-to-video synthesis, video personalization, video editing, video-to-audio generation, and text-to-audio generation. Our largest video generation model is a 30B parameter transformer trained with a maximum context length of 73K video tokens, corresponding to a generated video of 16 seconds at 16 frames-per-second. We show multiple technical innovations and simplifications on the architecture, latent spaces, training objectives and recipes, data curation, evaluation protocols, parallelization techniques, and inference optimizations that allow us to reap the benefits of scaling pre-training data, model size, and training compute for training large scale media generation models. We hope this paper helps the research community to accelerate progress and innovation in media generation models

Meta announces Movie Gen A Cast of Media Foundation Models We present Movie Gen, a cast of foundation models that generates high-quality, 1080p HD videos with different aspect ratios and synchronized audio. We also show additional capabilities such as precise instruction-based video editing and generation of personalized videos based on a user’s image. Our models set a new state-of-the-art on multiple tasks: text-to-video synthesis, video personalization, video editing, video-to-audio generation, and text-to-audio generation. Our largest video generation model is a 30B parameter transformer trained with a maximum context length of 73K video tokens, corresponding to a generated video of 16 seconds at 16 frames-per-second. We show multiple technical innovations and simplifications on the architecture, latent spaces, training objectives and recipes, data curation, evaluation protocols, parallelization techniques, and inference optimizations that allow us to reap the benefits of scaling pre-training data, model size, and training compute for training large scale media generation models. We hope this paper helps the research community to accelerate progress and innovation in media generation models

AK

62,719 次观看 • 2 年前

Ovi is out on Hugging Face Twin Backbone Cross-Modal Fusion for Audio-Video Generation Ovi is a veo-3 like, video+audio generation model that simultaneously generates both video and audio content from text or text+image inputs. Video+Audio Generation: Generate synchronized video and audio content simultaneously Flexible Input: Supports text-only or text+image conditioning 5-second Videos: Generates 5-second videos at 24 FPS, area of 720×720, at various aspect ratios (9:16, 16:9, 1:1, etc)

Ovi is out on Hugging Face Twin Backbone Cross-Modal Fusion for Audio-Video Generation Ovi is a veo-3 like, video+audio generation model that simultaneously generates both video and audio content from text or text+image inputs. Video+Audio Generation: Generate synchronized video and audio content simultaneously Flexible Input: Supports text-only or text+image conditioning 5-second Videos: Generates 5-second videos at 24 FPS, area of 720×720, at various aspect ratios (9:16, 16:9, 1:1, etc)

AK

23,082 次观看 • 1 年前

Splatter a Video can turn a video into a 3D Gaussian representation! This allows for enhanced video tracking, depth prediction, motion and appearance editing, and stereoscopic video generation.

Splatter a Video can turn a video into a 3D Gaussian representation! This allows for enhanced video tracking, depth prediction, motion and appearance editing, and stereoscopic video generation.

Dreaming Tulpa 🥓👑

11,586 次观看 • 2 年前

Runway has unveiled a massive breakthrough in AI video generation at NVIDIA's GTC event, showcasing a new model capable of generating high-definition video in genuine real-time. The new model boasts a "time-to-first-frame" of under 100 milliseconds. This means HD video begins generating and playing almost the instant a command is given. This research preview was co-developed with NVIDIA and runs on their newly announced Vera Rubin architecture. This real-time capability is a foundational step for Runway's General World Model (GWM-1). It pushes the technology beyond simple video generation and into the realm of live, interactive environments.

Runway has unveiled a massive breakthrough in AI video generation at NVIDIA's GTC event, showcasing a new model capable of generating high-definition video in genuine real-time. The new model boasts a "time-to-first-frame" of under 100 milliseconds. This means HD video begins generating and playing almost the instant a command is given. This research preview was co-developed with NVIDIA and runs on their newly announced Vera Rubin architecture. This real-time capability is a foundational step for Runway's General World Model (GWM-1). It pushes the technology beyond simple video generation and into the realm of live, interactive environments.

Wes Roth

82,506 次观看 • 7 个月前

Collaborative Video Diffusion Consistent Multi-video Generation with Camera Control Research on video generation has recently made tremendous progress, enabling high-quality videos to be generated from text prompts or images. Adding control to the video generation

Collaborative Video Diffusion Consistent Multi-video Generation with Camera Control Research on video generation has recently made tremendous progress, enabling high-quality videos to be generated from text prompts or images. Adding control to the video generation

AK

29,278 次观看 • 2 年前