The wildest demo on Hugging Face currently is this "Camera Control" LoRa trained on Qwen Image Edit 🤯 It lets you upload an image and generate a new scene by rotating the camera view. You can pan left/right, move forward or zoom in, and switch between bird’s-eye and worm’s-eye perspectives - all with just a few clicks. In other words, it adds full camera angle control to an image editing model, something Google's Nano Banana does not have. LoRA (Low-Rank Adaptation) is a technique for training only a few parameters on top of a frozen pre-trained model - in this case, Qwen Image Edit. This enables users to personalize models or add specific capabilities like camera control without retraining the entire model. The Space also leverages a variant called Rapid-AIO, which significantly speeds up inference, delivering great results in just four steps. Below someone tried it out on the "distracted boyfriend" meme, pretty cool right? Try it out here: https://lnkd.in/et3Yheug Source: https://lnkd.in/eaTZzHPr

The wildest demo on Hugging Face currently is this "Camera Control" LoRa trained on Qwen Image Edit 🤯 It lets you upload an image and generate a new scene by rotating the camera view. You can pan left/right, move forward or zoom in, and switch between bird’s-eye and worm’s-eye perspectives - all with just a few clicks. In other words, it adds full camera angle control to an image editing model, something Google's Nano Banana does not have. LoRA (Low-Rank Adaptation) is a technique for training only a few parameters on top of a frozen pre-trained model - in this case, Qwen Image Edit. This enables users to personalize models or add specific capabilities like camera control without retraining the entire model. The Space also leverages a variant called Rapid-AIO, which significantly speeds up inference, delivering great results in just four steps. Below someone tried it out on the "distracted boyfriend" meme, pretty cool right? Try it out here: https://lnkd.in/et3Yheug Source: https://lnkd.in/eaTZzHPr

This is what I miss sometimes in the discussions on AI: people lamenting on the accuracy of results produced by ChatGPT ("it's hallucinating all the time, the hype is not warranted") and missing the bigger picture of generative AI. The progress on signal processing and generation is super impressive and the easily available tools are often already "good enough" to replace a good amount of human work. As an example: of course you can pay a professional speaker to dub videos. Truth is: AI voice-overs might still lack the last 10% of authenticity and human touch to make same on par with a really good professional, but for most people it is just good enough and not worth it to pay more. Same will be for generating royalties-free background music, translations of documents etc. Of course this is not positive for the artists and professionals in these fields, but we won't be able to turn back time. We are living in very interesting times.

Virtual Gesture Studio! This is an advanced computer vision application that transforms your webcam into an interactive digital canvas. Forget your mouse and keyboard draw, design, and manage your artwork using only hand gestures. This project leverages OpenCV for real-time image processing and Google MediaPipe for precise 21-point hand landmark detection, supporting intuitive two-hand control.

The wildest demo on Hugging Face currently is this "Camera Control" LoRa trained on Qwen Image Edit 🤯 It lets you upload an image and generate a new scene by rotating the camera view. You can pan left/right, move forward or zoom in, and switch between bird’s-eye and worm’s-eye perspectives - all with just a few clicks. In other words, it adds full camera angle control to an image editing model, something Google's Nano Banana does not have. LoRA (Low-Rank Adaptation) is a technique for training only a few parameters on top of a frozen pre-trained model - in this case, Qwen Image Edit. This enables users to personalize models or add specific capabilities like camera control without retraining the entire model. The Space also leverages a variant called Rapid-AIO, which significantly speeds up inference, delivering great results in just four steps. Below someone tried it out on the "distracted boyfriend" meme. Try it out here: https://lnkd.in/et3Yheug Source: https://lnkd.in/eaTZzHPr 👉 Follow Ultan O. for more

🌍✨ Get ready to dive into WorldExplorer! This tool generates fully navigable 3D scenes from text. First, it creates a stunning image, then adds more angles and voila—your scene comes to life with Gaussian magic! Check out the interactive viewer on their site where you can stroll through examples. It might not have the glam of some others, but hey, it's real! 💯 If you're rocking an RTX 3090, here’s the scoop: ⏱️ Frame generation: ~5 mins (speed mode) ⏱️ Full scene expansion: 6-7 hours Time to get exploring! 🚀 #3DArt #TechMagic

🚀 Excited to share our new work at #ICCV2025 — T2Bs: Text-to-Character Blendshapes via Video Generation! T2Bs is a framework for generating high-quality, animatable 3D character heads directly from text, bridging the gap between text-to-3D generation (great geometry but static) and video diffusion models (dynamic but geometrically inconsistent). By leveraging deformable 3D Gaussian splatting, T2Bs aligns static 3D assets with dynamic video outputs — enabling coherent, expressive, and view-consistent 4D character generation. ✨ Highlights: Outperforms existing 4D generation methods in both accuracy and expressiveness Produces smooth, fully registered 3D geometries with realistic motion Paves the way for scalable morphable head models from text input 📍 Poster: Wed Oct 22, 11:15 a.m. – 1:15 p.m. (HST) Exhibit Hall I #1256 🔗 Project: https://lnkd.in/eZuGwvin 📄 Paper: arxiv.org/pdf/2509.10678 💻 Code: coming soon!

A new AI tool called Marble builds 3D worlds from images and text prompts using Gaussian splats. You get a quick, reusable static scene for games and simulations, reducing GPU requirements. Marble supports up to eight images and blends graphics with language for more accessible 3D design. - How does Marble create 3D scenes from images and text? - What are Gaussian splats in 3D modeling? - Where does Marble fit within game development or robotics tools? #ai #artificialintelligence #ainews #CarolEstrano #aiavatar

FlashWorld revolutionizes 3D scene generation with its cutting-edge technology. By leveraging a groundbreaking 3D-oriented generative model, FlashWorld can swiftly produce high-quality 3D scenes at remarkable speeds, surpassing traditional methods by 10–100×. This innovative approach directly generates 3D Gaussian representations, prioritizing realism and coherence. Thanks to its dual-mode pre-training and cross-mode distillation techniques, FlashWorld harmonizes the capabilities of both 2D and 3D frameworks. The result is outstanding rendering precision, robust adaptability, and instantaneous 3D scene creation from any input. Experience the future of 3D world generation with FlashWorld's unparalleled efficiency and realism. Project: https://lnkd.in/gzjtYG6g GitHub: https://lnkd.in/gMzUc4dw Paper: https://lnkd.in/g43iqFhm

AR isn’t just about overlaying graphics on a camera feed. It’s about contextualizing the real world: lighting, geometry, user movement, interaction. In our lab we work with shaders and spatial mapping to bring AR experiences that feel natural, not gimmicky. Imagine your phone knowing where you stand, what you look at, and adjusting AR accordingly. That’s what we’re building.