Toronto researchers just open-sourced ProxyPose, a system that recovers full 6-DoF 3D pose trajectories from standard video using just a single clicked pixel. Instead of tracking directly, it translates the query into a colored cube via a video model and solves the pose geometrically. Why it matters: It seamlessly handles transparent, reflective, or occluded objects where classical trackers break down.
Researchers just released MIRA, an interactive multiplayer world model that simulates Rocket League without any physics or rendering engine. Running on a single B200 GPU, the 5B diffusion model generates consistent 20 fps views responding to real-time inputs from four players, and the code and dataset are fully open-sourced.
Researchers just introduced PixWorld, a unified model that handles both 3D scene reconstruction and generation directly in pixel space. Instead of relying on latent diffusion and VAEs that lose fine detail, it supervises diffusion on rendered images with a custom geometry perception loss. The approach matches top reconstruction models while beating prior latent generators in consistency.
Hi3D just launched an end-to-end 3D printing workflow that turns text prompts into manufacturable 3MF files in about five minutes. Unlike visual-only 3D tools, it generates watertight meshes, splits models into printable parts, and calculates press-fit and ball joints automatically. Why it matters: It removes hours of manual CAD cleanup, bringing AI manufacturing to anyone with a 3D printer.
Researchers just open-sourced RDM, a technique that generates high-quality images in a single forward pass instead of multi-step diffusion. By matching visual feature distributions across frozen encoders, it compressed a 4-step FLUX.2 into a 1-step model that actually beats its teacher. It cuts generation latency to virtually instant while keeping compute requirements modest.
Researchers just released Mr. Flow, a training-free framework that accelerates text-to-image models by 9x to 21x. It drafts quickly at low resolution, upscales with subtle noise, and refines high-frequency details in the final step, slashing wait times to around one second without hurting quality.
Meta just introduced LUNA, a neural animation model that ditches traditional skeletal skinning and body-fitting bottlenecks. By turning simple 2D signals like sketches, keypoints, or video clips directly into 3D Gaussian deformations, it breathes motion into static character photos without the usual jitter and mesh tearing.
Google just launched two new generative media models for developers: Nano Banana 2 Lite and Gemini Omni Flash. Nano Banana 2 Lite delivers text-to-image outputs in 4 seconds at an ultra-low $0.034 per 1,000 images, while Omni Flash brings conversational video generation and editing at $0.10 per second of video. Why it matters: developers can now cheaply chain rapid image drafts straight into animated, multi-turn video edits.
Researchers just open-sourced OVOW, a training-free system that turns a single video into simulation-ready 4D mesh scenes with separated objects. By chaining models like Qwen3-VL and SAM 3, it reconstructs motions and deformations directly for physics simulators. It is available for non-commercial research and recommends at least 40GB of VRAM.
Google just opened up personalized image generation in the Gemini app to eligible U.S. users for free. With your permission, it pulls context from Gmail, YouTube, and Google Photos, allowing you to create custom art featuring yourself without uploading photos manually. The feature is opt-in and easily controlled in your settings.
ComfyUI just launched Comfy MCP in public beta, turning AI agents like Claude and Cursor into automated creative operators. Instead of manually wiring complex node spaghetti, you can simply prompt your agent in plain English to build, edit, and run production pipelines. It is free to test during beta with a Comfy account.
Researchers just introduced ViDiHand, a system that taps into pretrained video diffusion models to reconstruct precise 4D hand poses directly from video. It operates on full frames without separate hand detectors, handling fast movement and severe occlusions to feed humanoid robots far better training data.
Developer eric-venti-seeds just released Sun-Direction-Lora-Flux2Klein9B on Hugging Face, letting creators pinpoint exact sun angles in outdoor generations using a simple 3D reference sphere. The workflow neutralizes original lighting into an overcast plate before casting sunlight from the chosen direction. Why it matters: precise light control has always been painful in diffusion models, and this brings an intuitive visual gizmo to the process.
Researchers just open-sourced PhysiFormer, a diffusion transformer that simulates real-world 3D mechanics directly on mesh coordinates. Instead of generating rigid, unresponsive shapes, it takes an object's initial mesh, velocity, and material stiffness to predict collisions and jelly-like deformations. It marks a solid step toward geometry-aware physics engines for robotics and graphics.
Researchers just introduced ShutterMuse, a unified multimodal LLM designed for real-time photography guidance right when you take a photo. Trained on 130K reasoning samples, it runs 20x faster than GPT-Image-2 to advise both photographers on framing and subjects on scene-matched poses. It basically turns an MLLM into a pocket photography coach.
NVIDIA just introduced NeuMatEx, an inverse rendering pipeline that extracts complex 3D neural materials straight from multi-view photos. By pairing a feed-forward prior with differentiable path tracing, it reconstructs tricky real-world surfaces like clearcoats and car paint in 4ms without baking in ambient lighting.
Figma just introduced a design agent at Config that builds custom, reusable plugins and WebGPU shaders directly from text prompts. Designers can now spin up tailored tools on the canvas without dev setups or coding skills. It effectively turns prompt writing into functional workflow automation for teams.
A creator just released XHS Visual Director, a dedicated skill designed to cure mismatched social media slide decks. Given a topic, it plans a 6-to-8 slide narrative, establishes unified visual rules, and delivers ready-to-run image generation prompts alongside captions and a QA checklist.
Google just added Maps Imagery Grounding to Google Flow, letting users generate photos and videos anchored to real-world Google Maps Street View data. Instead of conjuring generic scenes, the agent pulls actual street-level details whenever you name a specific location or landmark. It basically gives generative video actual GPS coordinates, though it is currently limited to U.S. locations.
Krea just released open weights for two new image models: Krea 2 Raw and Krea 2 Turbo. Raw is an undistilled, mid-training checkpoint built specifically for developers to fine-tune, while Turbo is a distilled, fast-inference version offering broad aesthetic variety. Why it matters: developers can now build and customize Krea's generation tech on their own hardware.
