All updates

VidDub is a two-step diffusion video dubbing model that uses replacement-aware cross-audio supervision to avoid conflicting source-mouth targets and achieves strong lip synchronization with high visual fidelity.

TaoMate-HD is an autoregressive streaming model that generates high-resolution audiovisual content in chunks with progressive denoising, history-based continuation, and optimized parallel deployment for faster, coherent long-form output.

Recent advances in generative AI have enabled multiple pathways for high-fidelity video synthesis: text-to-video, image-to-video animation, and video outpainting. Empirical side-by-side evaluations of how untrained viewers perceive and distinguish these outputs from real footage…

Autoregressive video generation has achieved impressive results on short clips, yet generating long videos remains challenging due to error accumulation over extended horizons, where each predicted frame depends on potentially imperfect previous frames. Existing approaches…

Recent advances in video generation have enabled generative video compression to reconstruct details lost during coding using pretrained spatiotemporal priors. However, a substantial gap remains between codec-induced degradation and the perturbation trajectories learned by…

Video inpainting for object removal is scored over the whole hole, although the background behind the object is visible in other frames of the shot for some hole pixels and in no frame for others. A whole-hole score therefore mixes copying with generation and cannot show which…

Visually realistic video generation does not guarantee physically consistent motion. Generated objects may exhibit implausible collision responses or rebounds without energy dissipation, while even physically valid motion predictions may not be faithfully realized during…

Autoregressive video generators produce long videos block by block, using historical context to maintain temporal consistency. After denoising, a committing pass encodes each clean block into keys and values (KV) that later blocks read, and the same clean block can be encoded…

Real-time long-video generation requires continuous output with consistent appearance and natural motion. Existing chunk-wise autoregressive methods face challenges in both generation efficiency and long-video quality: serialized denoising and history-cache refresh require two…

We report the end-to-end deployment and measurement of MiniMax H3, a 20.38-billion-parameter video diffusion transformer (DiT) - the count measured from the safetensors headers of the Q8 checkpoint as deployed locally (Section 3.2) - running entirely on consumer Apple Silicon…

We present \textbf{UniCaMo}, an unified framework for joint control of camera and object motion in image-and-text-to-video diffusion models. Given a reference image, text prompt, sparse 3D object trajectories, and a 3D camera trajectory, UniCaMo constructs motion-consistent…

In this video, we'll share a new workflow for relighting your films that will change Filmmaking forever. The results are incredibly impressive. The workflow is simple, but there are a few things you can do to ensure the best quality possible! Check out a full breakdown here…

Text and image to video · 3–5 s · 480p, 720p · 16:9 · from $0.005/s on OpenRouter

Your AI characters can now stand in your actual room, and react to what's in it! Simulon is back: turn a character image into a 3D "neural asset," place it in the real world through your phone's camera, act out the scene with your own camera move, and the app regenerates the…

We just wrapped our 4th annual AI Horror Film competition, and we are SHOOK. Not only are we terrified by the stories (all our judges are now sleeping with the lights on), but the quality of these films is mind-blowing. Once the premiere has finished, check out the winners here…

In this video of ComfyUI Nodes Explained, you will learn how ComfyUI reads your prompt: what CLIP Text Encode does, how positive and negative prompts work, why the negative prompt does nothing at CFG 1, what Conditioning Zero Out is for, how to write a prompt for new models, why…

Kling 4.0 makes up to 30 seconds of 1080p video from a prompt, and it can thread one continuous shot through up to 10 keyframes.

A little "Inside Baseball" on the Chinese AI Video models, and while I don't normally cover International Business News, I thought this was pretty interesting. The Voltron of AI Video models, next year?

CozyClay 3.0: a free, open-source 3D previs tool built for AI video. One-click install through Pinokio; it runs in your browser and doesn't need a GPU. Full details in this video: https://youtu.be/qlSI_ZbRdEc

- Iron Goddess: an unclaimed AI video model that showed up in battle mode on the Arena this week. Big, fast, swing-for-the-fences motion; frame by frame, the fastest moments get smudgy. Seedance 2.5 holds together better, with less dynamic movement. - Who made it? Nobody has…

A free, open-source tool that gets you the exact shot before you generate, plus the fix for when 3D-guided AI video comes out stiff. And a secret AI video model just turned up in blind tests on the Arena, swinging big on motion. Also this week: Kling's reported IPO, Ideogram's…

In this video of ComfyUI Nodes Explained, you will learn what a model is, what the file names mean (bf16, fp8, int8, GGUF), where each file goes, and how to load it, with Load Checkpoint or with Load Diffusion Model, Load CLIP and Load VAE. Everything is shown with Z-Image…

Seedance faked a video game that doesn’t exist, so I had Claude Opus 5.5 rebuild it as a playable browser shader in 45 minutes, using gokaygokay’s video-to-shader skill. Then I pushed further: Claude wrote a new skill and rebuilt an AI-generated pirate cabin in Unreal Engine…

ByteDance's unreleased world model took a scene from Stephen Chow's King of Comedy and re-shot it from new camera angles off one drawn camera path, at a recent showcase event. More details on the World Model in this video: https://youtu.be/VTrc346ea30

Try AIQORA: https://aiqora.ai/empfehlung?ref=MATTVIDP Start free and get 10 extra credits with code MATTVIDPRO. This segment is sponsored by AIQORA, an AI video platform from Germany. I was out sick for five days and the AI world did not wait. Gemini 4 Argon is out, and it's…

In this video, we'll team up with Albert Bozesan to show you the best AI Camera control hack we have ever seen. Claude can not control Blender, and this makes a once-complicated software WAY easier to utilize in your pipeline. Explore a full breakdown of the tutorial…

ByteDance's new world model, Sora returns?, Krea 3 is here, OmniHuman 2.0 is back, Pico Space Pro, World Gacha, and LTX-2.5's new post tools. ByteDance's unreleased world model took a scene from Stephen Chow's King of Comedy and re-shot it from new camera angles off one drawn…

This is the first video of ComfyUI Nodes Explained, a series where I explain the ComfyUI nodes one by one. Before we look at any single node, this video covers the basics every beginner needs: what a node is, its inputs, outputs and settings, and what the wire colors mean. ⬇️⬇️…

Check out our full tutorial here: https://www.youtube.com/watch?v=BFsVW1DrLng&t=1s

In this video, we'll compare two popular AI character consistency methods to see which one actually works better for AI filmmaking. Check out the full written breakdown and prompts here: curiousrefuge.com/blog/how-to-create-pro-ai-character-sheets Check out our courses…

Upscale, interpolate, and convert to HDR locally on your RTX GPU, with better EXR support and color management built for VFX pipelines.

Here's one of my favorite hacks for saving credits while generating for your AI Film. Nowadays, you don;t have to fully commit, but there are a handful of workflows and processes you can set up so that you can ensure you don't waste tons of money on bad generations. Check out…

Out of Memory is a song about ComfyUI, for everyone who has watched the progress bar crawl to 99 percent and then seen the red box. The whole music video happens inside the real ComfyUI: real workflows, real runs with the live preview, real out of memory errors, real missing…