Hao Li

🎶 🧑 OmniTalker: One-shot Real-time Text-Driven Talking Audio-Video Generation With Multimodal Style Mimicking TLDR: Realtime joint audio and visual generation using reference clips of both and lightweight DiTs with an audio-visual fusion module to achieve sync. 📽️ Project Page: https://lnkd.in/eh4qMrz2 📜 Paper: https://lnkd.in/erzyYpKk

71

4 Comments

Luma AI just dropped Ray3 with an impressive demo promising 4K HDR video. I decided to put it to the test. No cherry-picking, just raw results from 2 hours of use. Site: https://lumalabs.ai/ray 4k download link to watch without compression: https://lnkd.in/eBfte2gX ✨ What stood out: -Interface is smooth you can queue shots and run multiple variants at once. -Cheap to run half of my $30 credit gave me ~100–150 generations, draft mode is a great feature! -Camera motion looks strong. -Fast turnaround. ⚠️ Where it struggles: -720p looks solid, HD is fine but 4K falls apart. The upscale looks more like low-poly CGI than real 4K. -Prompts aren’t followed well. Even simple prompts like “static camera” took multiple retries. -Every usable shot took 6-10 variants. -Close-ups look good, but medium/wide shots suffer from the usual AI quirks. 🔥 What really annoyed me The tech demo on their website looks like absolute fire. Trying to recreate those shots was a bit of a failure. Sure, I managed to get the camera motion and scene in there but only after a shotgun approach of endless iterations. And even then, the image quality never came close to their demo material. It suffers from the same issue as Veo3: amazing marketing, but the actual results require endless “slot machining.” 💡 Takeaway Ray3 has potential the speed and pricing are great but right now, Seedream still delivers stronger results overall for a cheaper price. Curious if anyone else here has tried Ray3 yet. What’s your take? 4k link: https://lnkd.in/eBfte2gX https://lumalabs.ai/ray https://x.com/lumalabsai

76

37 Comments

Real-time rendering has always fascinated me. Seeing light, material, and motion resolve instantly transforms how we think and iterate. The barrier has never been conceptual, but economic. High-end systems have historically required expensive hardware, proprietary engines, and significant infrastructure, limiting access to institutions and well-funded studios. Generative AI presents itself as a potential solution. If it truly lowers that threshold, it does more than reduce cost. It redistributes creative power. It shifts real-time rendering from a privileged capability to an accessible workflow, with implications for authorship, experimentation, and who gets to participate in high-fidelity visual storytelling. This may not just be a technical upgrade. It may be a structural correction in who gets to think, prototype, and author in real time. Let’s see how this plays out.

2

2 Comments

The authors observe that existing stateful recurrent neural networks for 3D reconstruction perform well on static scenes — but struggle when parts of the scene move. In dynamic (non‑rigid) scenarios, moving regions corrupt the internal feature propagation between spatial memory and image features, producing artifacts and unstable reconstructions. By inspecting the inner behaviour of a pretrained transformer‑based model, they find a consistent latent signal: attention maps aggregated across layers naturally down‑weight dynamic regions. In other words, the model already encodes “motion cues” implicitly, but does not use them explicitly in reconstruction. Building on this insight, the paper proposes MUT3R, a lightweight, training‑free framework that at inference time applies an “attention‑derived motion gating” — using the implicit motion cues to suppress influence of dynamic regions early in the transformer. By doing so, the method prevents dynamic content from propagating through the feature hierarchy and destabilizing the reconstruction. The result is improved temporal consistency and more robust camera‑pose estimation across multiple dynamic benchmarks, offering a simple and effective path toward motion‑aware streaming 3D reconstruction without retraining or fine‑tuning. https://lnkd.in/gh3zfZ3r

1

Blueprints vs. C++ in Unreal Engine : Which One Should You Choose? One question I hear constantly from Unreal Engine learners is this: “Should I focus on Blueprints or C++?” With Unreal Engine 5-series, the answer is clearer than ever- it’s not a competition, it’s a collaboration. Blueprints: The Visual Powerhouse Blueprints make Unreal incredibly approachable. Visual scripting allows you to focus on game logic and creativity, not syntax. They’re perfect for: ·        Beginners learning core concepts ·        Rapid prototyping and iteration ·        Designers and artists tweaking gameplay without touching code Blueprints help ideas come to life fast. That instant feedback is powerful, especially in learning and early development stages. C++: The Power Beneath the Engine C++is the backbone of Unreal Engine, gives you full control. It’s where Unreal’s core systems live, and when you work in C++, you’re speaking the engine’s native language. This level of access is invaluable for performance-critical systems, advanced AI, networking, or large-scale projects. ·        Maximum performance ·        Full access to engine systems ·        Better scalability for large or complex projects Yes, the learning curve is steeper, but the control and optimization it provides are unmatched, especially for AI, networking, and core systems. So… Which One Wins? Here’s the honest truth: There is no “Blueprints vs. C++.” It’s Blueprints and C++. In real-world development, and especially in professional studios, both are used together. Blueprints shine when you need speed, clarity, and creativity. C++ excels when you need performance, control, and scalability. The smartest workflow uses each where it makes the most sense. Prototype quickly in Blueprints. Build core systems in C++. Expose C++ logic back to Blueprints for designers. This hybrid workflow combines speed with strength, and it’s how most professional Unreal projects are built. One helps you think visually and move fast. The other helps you go deep and build strong foundations. Final Thought: I often tell students: “Blueprints are your quick brushstrokes. C++ is your fine detailing. Together, they create the masterpiece.” Instead of asking which one is better, ask: What does my project need right now? That mindset changes everything. Once you start thinking that way, Unreal Engine truly opens up, and that’s where the real fun begins..

201

46 Comments

Hot: Procedural splats are strong primitives for prompt-driven procedural authoring (now available in Arrival.Space) They abstract away many complex patterns needed for fire, volume fog, emissive particles and ground cover. You shape probability, not polygons. Generate. Tweak. Re-prompt - Directly in the browser. Scene link in comments. 🔥

64

16 Comments