Maket just rolled out Text to Furniture, a new tool that turns text prompts into 3D furniture models. You can describe specific pieces like an oval burl wood dining table, and it creates a scaled 3D model that drops right into your floor plan.
Ant Group just unveiled Ling-3.2-flash, a 560B-parameter native multimodal MoE model that only activates 25B parameters per token. It natively processes text, image, and video inputs across a 1M context window, handling tasks from multi-agent workflow coordination to 3D spatial reconstruction.
A two-person lab called Linum just open-sourced Pyramid-JiT, a 2B-parameter text-to-image model trained entirely from scratch for only $21,000. It ditches the standard VAE to generate images directly in pixel space, delivering solid visuals on a shoestring budget. While it struggles with text and complex anatomy, the Apache 2.0 release proves training diffusion models doesn't require millions.
Pangram just launched its flagship Image 1.0 and Image Flash 1.0 detection models, built to spot synthetic images and videos from frontier generators. Image 1.0 outputs heatmaps that highlight exact manipulated spots, while Flash 1.0 runs frame-by-frame video analysis with 98.7% accuracy on the DF26 benchmark. Both models are available via API and a Chrome extension for real-time feed checking.
OpenAI just launched Images 2.6 across ChatGPT, Codex, and its API, boosting generation speeds by 50% and adding native support for animated GIFs. Developers get two API variants—including the six-second Flare model—plus SVG exports and fine-grained scene complexity controls. It is live now for all subscription tiers.
OpenAI just revealed it's working with Chip Ganassi Racing engineers to turn 3D scans into editable CAD models and manufacturing drawings. The company is also collaborating with Mazda to explore new vehicle designs through simulation, relying on partner engineers to review outputs and sharpen the models.
Researchers just open-sourced NitroFusion 2, slashing image diffusion down from 8 steps to just 1 or 2 steps. By introducing adversarial distillation directly into the multimodal text stream, it preserves fine details and legible typography even in a single pass. The weights are available on Hugging Face and run locally via ComfyUI.
Pruna AI just released P-H3-Image, an open-weights image model built by reworking the MiniMax-H3 video model. By distilling components and slashing diffusion steps from 50 to 8, it runs 5x faster and shrinks down to 20 GB while rendering up to 16 megapixels. ComfyUI already supports it natively.
Sandflow AI just open-sourced sand3d, a Python library built on Blender that lets LLMs construct parametric 3D objects and robotic joints in readable code. The entire API fits in a 3,600-token prompt—using up to 6x fewer tokens than raw Blender script—and launches alongside a massive 100,000-object open dataset.
NVIDIA just introduced Cosmos Framework and Cosmos NIM microservices to build and deploy physical AI world models for robotics, AVs, and surgical simulation. It features new Cosmos 3 Nano checkpoints capable of generating synchronized multi-camera video and LiDAR data. With speculative decoding, the NIM containers deliver 2–3x faster inference on H200 GPUs.
Sperid Labs just open-sourced Idol, a new model that turns a single 2D image into a standard 3D Gaussian splat with up to a million points. It handles background removal, accurately guesses the hidden backside, and caps out at just 7.2 GB of VRAM. Why it matters: High-fidelity 3D reconstruction is finally fast and lean enough for local desktop rigs.
Developer Nutlope just open-sourced Bygga, a web app that turns flat IKEA paper instructions into interactive, animated 3D guides. It covers 1,171 products completely offline, reads steps aloud, and listens for voice commands so you can say "next" without dropping your screwdriver. Unknown manuals can be scanned and drafted via Together AI.
Microsoft just released MAI Image 2.7, its latest image editing model that debuted at No. 2 on Artificial Analysis and No. 3 on Arena Edit. The update outperforms its predecessor across all evaluated editing tasks, showing the biggest jump in scene and style edits with a 24-Elo lead. Microsoft is rolling out editing first, with text-to-image coming soon.
AI 3D platform Hi3D just open-sourced its V1.0 model weights and code under a commercial-friendly license. Sized at just 7.74GB, it runs on consumer GPUs to turn single images into detailed meshes up to 1536³ resolution. It hands creators a solid local baseline, even though the flagship V3 remains closed-source.
DeepMind just unveiled Genie 4, a real-time world simulation model that pairs interactive environments with native sound effects, music, and spoken dialogue. Backed by Gemini models for physical realism and continuous logical reasoning, it lets users steer simulated scenes on the fly using standard keyboard controls or text prompts. It is currently rolling out as a research preview to trusted testers.
SpAItial just launched Echo-3, a physically grounded world model that turns casual iPhone video scans into accurate, metric-scale 3D environments in minutes. It intelligently fills in blind spots your camera missed while eliminating floating artifacts, alongside an enterprise API for robotics and simulation. It basically turns casual smartphone footage into production-ready spatial data without professional gear.
Designer AmirMushich just open-sourced a clever technique to create interactive, rotatable web product pages without 3D modeling. The method generates a spinning video from reference images using LTX-2.5 Pro, splits it into about 30 frames, and controls playback via JavaScript based on cursor movement. Why it matters: E-commerce brands can build sleek pseudo-3D product showcases with zero 3D modeling overhead or runtime AI costs.
Researchers just introduced MatAnyone, a video matting model capable of isolating complex elements like individual hair strands and semi-transparent edges. It runs on around 3GB of VRAM and hits real-time 4K matting on a single H200 GPU by selectively allocating compute to tricky regions. Open-source code and ComfyUI node support are planned soon.
Bria just released open weights for Fibo 2.0 Edit Turbo, an 8B-parameter image editing model under a non-commercial license. Distilled down to just 4 steps without classifier-free guidance, it uses structured prompts to swap objects or alter styles while keeping untouched areas pixel-stable.
Alibaba just rolled out commercial Pro and Turbo APIs for Qwen-Image-2.1 on its Model Studio platform. The single 7B model collapses separate design steps by handling generation, 10-image reference edits, native RGBA transparency, and 2K typography. Turbo starts at around $0.014 per image, giving production teams a full pipeline without self-hosting GPUs.
