AI News for 3D and Video Generators
Short news notes about what has changed in the tools: a new version released, a license changed, support for another platform added. No press-release retelling: if a claim cannot be tied to a repository, model page, or official page, there will be no note.
Articles
38 articles
VA-Judger: Reward Model for Joint Video and Audio GenerationResearchers introduced VA-Judger, a reward model for post-training of joint video and audio generation systems, along with the VAPref-10K dataset of 1…
SparsePR accelerates video generators with training-free sparse attentionResearchers introduced SparsePR, a training-free block-sparse attention method for video generators and world models. The paper was published on arXiv…
HunyuanVideo gets a generation acceleration method with near-lossless quality at 5–7x speedupResearchers introduced LinCa, a method for accelerating diffusion models via learnable feature caching with component-wise decomposition. Instead of a…
New h3-vbvr model on HuggingFace tackles image-to-videoOn August 18, 2026, the Patarapoom/h3-vbvr model was published on HuggingFace with the image-to-video task. At the time of publication, it had 2,434 d…
SQuad accelerates video generation in Wan 2.2, reducing attention by 67xThe authors of SQuad introduced a sub-quadratic attention distillation method for Video Diffusion Transformers. Instead of training a model from scrat…
KeyID introduces training-free method for identity-preserving video generationA paper on arXiv presents KeyID, a training-free framework for identity-preserving video generation that separates video dynamics synthesis from ident…
AnyTalk generates 3D speech animation for arbitrary characters without animation dataResearchers introduced AnyTalk, a method for generating 3D speech animation for arbitrary characters that requires neither animation data nor manual r…
Audio-visual generation gets an acceleration method that preserves audio-video synchronizationThe authors of the paper Efficient Audio-Visual Generation via Synchrony-Aware Cross-Modal Sparse Attention propose a method to speed up inference for…- Qwen-Video-Edit turns an image editing model into a video editorAlibaba introduced Qwen-Video-Edit, an approach where the instruction-based image editing model Qwen-Image-Edit is applied to video without training a…
MegaParts scales part-based 3D object generation to 300 partsResearchers introduced MegaParts, a framework for generating 3D objects decomposed into semantic parts. The approach is based on a vector-quantized to…
V-RAE proposes a new approach to latent spaces for video generationA paper on V-RAE has been published on arXiv, in which the authors rethink the structure of video autoencoder latent spaces. Instead of optimizing the…
Alaya-EVOKE presents a method for generating long interactive worlds with linear scalingA paper by Alaya-EVOKE has been published on arXiv, in which the authors tackle the task of generating long interactive worlds. They externalize the s…
LTX-2.5 released on HuggingFace with text-to-video taskThe model comfyicu/LTX-2.5 has been published on HuggingFace with the text-to-video task. Publication date: 13.08.2026, the model has 3630 downloads…
LTX-2.5 text-to-video model released on HuggingFaceThe yuvraj108c/LTX-2.5 model with a text-to-video task has been published on HuggingFace. Publication date: August 12, 2026. As of this news note, the…
News Note on a New Diffusion-Based Method for Removing Reflections from VideoResearchers presented the work From Synthesis to Removal: Physics-Grounded Reflection Simulation and Diffusion-Based Video Dereflection, published on…
LTX-Video and CogVideoX Get a New Zero-Shot Captioning Method via Synthetic DataResearchers have published a paper on arXiv titled Watching Synthetic Videos: Aligning Cross-modal Representations with Visual Synthesis for Zero-shot…
Minimax-H3-fl2va-ref2va-hybrid-models: new text-to-video model on HuggingFaceOn August 11, 2026, the model smhfacct/Minimax-H3-fl2va-ref2va-hybrid-models was published on HuggingFace with the text-to-video task. At the time of…- Latent-to-4D generates 4D scenes directly from video latents without intermediate RGBThe authors of the paper Beyond Pixels: From Video Priors to 4D Worlds propose the Latent-to-4D method, which bypasses the RGB video reconstruction st…
Sci-VBench: A Benchmark for Scientific Accuracy of Video from Generative ModelsThe authors of Sci-VBench introduced a benchmark for evaluating video generation in scientific domains. It includes 1,253 expert-annotated examples ac…
Hunyuan3D 2.1 without CUDA: update adds ROCm and PBR texturingThe port of Hunyuan3D 2.1 to hardware without CUDA was updated on August 9, 2026, and it now includes something that was previously missing entirely…
New Minimax_h3_hybrid model on HuggingFace tackles image-to-videoThe abakanai/Minimax_h3_hybrid model has been published on HuggingFace for image-to-video generation. The model appeared on August 9, 2026, and at the…
Robust-WAM adds semantic robustness to World-Action Models built on video generatorsResearchers from arXiv presented Robust-WAM, a post-training method for World-Action Models built on video generative models. The main problem the wor…
Vorch-Streamer turns avatar generation into real-time streaming videoarXiv has published Vorch-Streamer, a post-training method that enables real-time text-to-video generation with audio. The authors address two issues…
VideoArgus offers a unified rubric-based evaluation system for video generation and editingResearchers have introduced VideoArgus, a framework for evaluating the quality of generated and edited videos, described in a paper on arXiv. Unlike e…
HelloWorld — a model for social interaction with characters in video worldsResearchers have introduced HelloWorld, a video world model in which the user can interact with characters on screen. With the press of a button, the…- GVCCTurbo proposes bitrate planning for generative compression of video and images withoutarXiv Researchers introduced GVCCTurbo, a planner that separates costly updates of the generative prior from transmitting corrections through a codebo…
- SPADE accelerates attention in video diffusion transformers without fine-tuningResearchers have introduced SPADE, a sparse attention engine that accelerates inference of video diffusion transformers without model fine-tuning. The…
SUV: Scene Future as Video Generation for End-to-End DrivingOn August 4, 2026, a paper on arXiv introduced SUV — an end-to-end driving framework that frames future scene understanding as a video generation task…
Token Radius Attention Speeds Up Video Generation in Diffusion Transformers Without Fine-TuningResearchers from Peking University have introduced Token Radius Attention (TRA), a method that reduces the computational load in Video Diffusion Trans…
EchoCache accelerates sound-driven video generation using audio signal energyarXiv reports on EchoCache, a caching method for audio-driven video generation (A2V). The authors identified two types of mismatch in existing approac…