Blender Developers Meeting Notes: 18 August 2025 By n8n on August 19, 2025 Blender Development Notes for weekly communication of ongoing projects and modules. This is a selection of changes that happened over the last week. For a full overview including fixes, code only changes and more visit projects.blender.org. Reset various runtime data for writing files (commit) - (Hans Goudey) Improve RNA performance tests flexibility. (commit) - (Bastien Montagne) Move VSync from an environment variable to an argument (commit) - (Campbell Barton) Recognize ACES config un-tone-mapped view (commit) - (Brecht Van Lommel) Allow empty names in File Output node (commit) - (Omar Emara) Support strings sockets (commit) - (Omar Emara) Allow menu sockets for pixel nodes (commit) - (Omar Emara) Removing Sun Beams node (commit) - (Mohamed Hassan) Don’t get/set PWD env var for working directory functions (commit) - (Jesse Yurkovich) Parallelize NURBS basis cache evaluation with O(n) complexity (commit) - (Mattias Fredriksson) Add cyclic curve offsets cache (commit) - (Hans Goudey) Do not sample direct light when ray segment is invalid (commit) - (Weizhen Huang) Always add world as object (commit) - (Weizhen Huang) Create one box for vdb mesh instead of many (commit) - (Weizhen Huang) Render volume by ray marching through octrees (commit) - (Weizhen Huang) Compute volume transmittance using telescoping (commit) - (Weizhen Huang) Shade volume with null scattering (commit) - (Weizhen Huang) Volume Scattering Probability Guiding (commit) - (Weizhen Huang) Use RGBE for denoised guiding buffers to reduce memory usage (commit) - (Weizhen Huang) Use analytic formula for homogeneous volume (commit) - (Weizhen Huang) Add and update volume test files (commit) - (Weizhen Huang) Store octree parent nodes in a stack (commit) - (Weizhen Huang) oneAPI: Disable L0 copy optimization for several dGPUs (commit) - (Nikita Sirgienko) Use deterministic linear interpolation for velocity (commit) - (Weizhen Huang) Use one-tap stochastic interpolation for volume (commit) - (Weizhen Huang) Add material name collision mode (commit) - (Oxicid) Shader: Add support for full template specialization (commit) - (Clément Foucault) Rewrite default_argument_mutation using parser (commit) - (Clément Foucault) Fix parser not being consistent (commit) - (Clément Foucault) Replace template macro implementation by copy paste (commit) - (Clément Foucault) Preprocess: Improve error reporting (commit) - (Clément Foucault) Remove Shader Draw Parameter workaround (commit) - (Clément Foucault) Add flag for shader debug info generation (commit) - (Christoph Neuhauser) Improve cyclical end cap rendering (commit) - (Casey Bianco-Davis) Export other curve types to SVG (commit) - (Casey Bianco-Davis) Edit Mode Pen Tool (commit) - (Casey Bianco-Davis) Support extracting Vulkan & OpenGL args even when disabled (commit) - (Campbell Barton) Add Apply Transforms option to obj exporter (commit) - (Thomas Hope) Fix recursive resync incorrectly clearing hierarchy info. (commit) - (Bastien Montagne) Prevent matching collection items only by their index if a name and ID are provided. (commit) - (Bastien Montagne) Avoid quadratic vertex valence complexity for corner normals (commit) - (_илья __) Improve performance and compression, always compress (commit) - (Aras Pranckevicius) Allow Preferences editor to be opened in Maximized Area (commit) - (Jonas Holzman) Gray out or hide asset shelf toggle if not available (commit) - (Julian Eisel) Center-align header modal status text (commit) - (Pablo Vazquez) Prevent automatic mode switching in certain scenarios (commit) - (Sean Kim) Remove liboverride UI dead code, improve UI messages. (commit) - (Bastien Montagne) Prevent ‘liboverride’ ‘decorator’ button to control keyframes. (commit) - (Bastien Montagne) Widen Preferences Window (commit) - (Pablo Vazquez) Theme: Move curve handle properties in common (commit) - (Nika Kutsniashvili) Generalized Alert and Popup Block Error Indication (commit) - (Harley Acheson) Tree View: Operator to delete with X key (commit) - (Pratik Borhade) Use UI_alert For Vulcan Fallback Warning (commit) - (Harley Acheson) Remove unused theme properties (commit) - (Nika Kutsniashvili) Warning When Dragging Non-Blend File Onto Executable (commit) - (Harley Acheson) “Duplicate Strips” also duplicates referenced IDs (commit) - (Falk David) Add copy and paste operators to preview keymap (commit) - (Ramon Klauck) Improve Histogram scope for HDR content (commit) - (Aras Pranckevicius) Add scene assets through strip add menu (commit) - (Falk David) Clear Strip Keyframes from Preview (commit) - (Ramon Klauck) Enable Pie Menu on Drag for Preview Keyframe Insert (commit) - (Ramon Klauck) Add “Mirror” menu to preview strip menu (commit) - (Ramon Klauck) Disable descriptor buffers (commit) - (Jeroen Bakker) Swap to system memory for device local memory (commit) - (Jeroen Bakker) Destroy resources in submission thread (commit) - (Jeroen Bakker) Update VMA to 3.3.0 (commit) - (Jeroen Bakker) Remove MoltenVK (commit) - (Jeroen Bakker) Enable maintenance4 in VMA (commit) - (Jeroen Bakker) Add message type for remote downloader messages to message bus (commit) - (Julian Eisel)

Of around 7,000 languages in the world, a tiny fraction are supported by AI language models. NVIDIA is tackling the problem with a new dataset and models that support the development of high-quality speech recognition and translation AI for 25 European languages — including languages with limited available data like Croatian, Estonian and Maltese. These tools will enable developers to more easily scale AI applications to support global users with fast, accurate speech technology for production-scale use cases such as multilingual chatbots, customer service voice agents and near-real-time translation services. They include: Granary, a massive, open-source corpus of multilingual speech datasets that contains around a million hours of audio, including nearly 650,000 hours for speech recognition and over 350,000 hours for speech translation. NVIDIA Canary-1b-v2, a billion-parameter model trained on Granary for high-quality transcription of European languages, plus translation between English and two dozen supported languages. NVIDIA Parakeet-tdt-0.6b-v3, a streamlined, 600-million-parameter model designed for real-time or large-volume transcription of Granary’s supported languages. The paper behind Granary will be presented at Interspeech, a language processing conference taking place in the Netherlands, Aug. 17-21. The dataset, as well as the new Canary and Parakeet models, are now available on Hugging Face. How Granary Addresses Data Scarcity To develop the Granary dataset, the NVIDIA speech AI team collaborated with researchers from Carnegie Mellon University and Fondazione Bruno Kessler. The team passed unlabeled audio through an innovative processing pipeline powered by NVIDIA NeMo Speech Data Processor toolkit that turned it into structured, high-quality data. This pipeline allowed the researchers to enhance public speech data into a usable format for AI training, without the need for resource-intensive human annotation. It’s available in open source on GitHub. With Granary’s clean, ready-to-use data, developers can get a head start building models that tackle transcription and translation tasks in nearly all of the European Union’s 24 official languages, plus Russian and Ukrainian. For European languages underrepresented in human-annotated datasets, Granary provides a critical resource to develop more inclusive speech technologies that better reflect the linguistic diversity of the continent — all while using less training data. The team demonstrated in their Interspeech paper that, compared to other popular datasets, it takes around half as much Granary training data to achieve a target accuracy level for automatic speech recognition (ASR) and automatic speech translation (AST). Tapping NVIDIA NeMo to Turbocharge Transcription The new Canary and Parakeet models offer examples of the kinds of models developers can build with Granary, customized to their target applications. Canary-1b-v2 is optimized for accuracy on complex tasks, while parakeet-tdt-0.6b-v3 is designed for high-speed, low-latency tasks. By sharing the methodology behind the Granary dataset and these two models, NVIDIA is enabling the global speech AI developer community to adapt this data processing workflow to other ASR or AST models or additional languages, accelerating speech AI innovation. Canary-1b-v2, available under a permissive license, expands the Canary family’s supported languages from four to 25. It offers transcription and translation quality comparable to models 3x larger while running inference up to 10x faster. https://blogs.nvidia.com/wp-content/uploads/2025/08/Canary-demo.mp4 NVIDIA NeMo, a modular software suite for managing the AI agent lifecycle, accelerated speech AI model development. NeMo Curator, part of the software suite, enabled the team to filter out synthetic examples from the source data so that only high-quality samples were used for model training. The team also harnessed the NeMo Speech Data Processor toolkit for tasks like aligning transcripts with audio files and converting data into the required formats. Parakeet-tdt-0.6b-v3 prioritizes high throughput and is capable of transcribing 24-minute audio segments in a single inference pass. The model automatically detects the input audio language and transcribes without additional prompting steps. Both Canary and Parakeet models provide accurate punctuation, capitalization and word-level timestamps in their outputs. Read more on GitHub and get started with Granary on Hugging Face.

Introduction My name is Ayush Banik, a 3D artist from India. I got into 3D through a random Blender tutorial back in late 2022, starting with that classic donut. I have worked on weapons for two games that are still under NDA, as well as some foliage and environment props for Dekogon. I'm still in engineering college, so working full-time is a bit difficult.EOTECH EXPS3 & G45The EOTECH EXPS3 scope and the G45 magnifier are part of my Full Internal AR-15 project. For references, I collected many high-quality images from EOTECH's official website as well as additional ones for texturing from The Weapon Room Discord channel. The aim of this project was to improve my texturing skills for black anodized and Cerakote coatings.I modeled this scope the same way I model all my other work: using a CAD/Blender live boolean workflow. For these two scopes in particular, I used Plasticity for modeling, MOI3D for exporting, and Blender for the low poly. The high-poly was derived in ZBrush using the process: DynaMesh > Polish > Decimate.For the topology, I exported the mesh using MOI3D and fixed the low poly in Blender. Some of the add-ons I used were Resample Mesh, LoopTools, etc. I always unwrap in Blender, straighten UVs using TexTools, and pack them with UVPackmaster. As for texture sets, each sight has its own UV set, since this is part of a weapon with removable attachments.MaterialsI love to talk about materials!The metal was very simple: a base with dark diffuse and specular, since it’s an anodized material, which is non-metallic and usually dark in nature. This was followed by a gloss and specular grain, and a metallic grain on the normal for the height.The next step is to put more normal details onto the surface, like milling.After that, I moved on to the color and gloss details. I added oil and fingerprint smudges layer by layer, as can be seen in the GIF below:After that, moving on to the dust, I did it in a way I really don’t recommend – manually. Every piece of dust was placed using a grainy noise stencil.Moving on to the sticker residue: I started with a little bit of glue and paper residue and worked my way up to dust.Parallax & Magnification EffectsFirst, I needed a height map, which can be created using just a simple black square. Next, for the depth, the following settings work very well, although depending on the size of the reticle, the “depth” option can be adjusted.Transmission set to Thin Surface works best, with “Use Albedo” turned on. Reflectivity should be set to Specular, with Conserve Energy turned on. And, of course, the alpha of the reticle should be in emission. Anything below 0.5 transparency works fine.One more important thing you will need in Marmoset Toolbag for this to work is in the Render settings: transmission needs to be at least 3 for the magnification to work (more on that later).Magnification:The single most important setting for magnification is the Refractive Index under Transmission. Refractive Index at 1 = no magnification. As you increase the RI, you get higher and higher magnification, along with more distortion.RenderingRendering the asset was a huge challenge. I didn’t want to do a flat render on a flat background. There was also the element of showing off my WIP gun, as the EOTECH project is just part of the full AR-15 project.Here’s my rendering setup: 20% of a SWAT guy holding a gun. A major mistake I made was not scaling the SWAT model – my man is 6’9” tall. I should have scaled him to 6’.I use a lot of directional lights to light my scene because I don’t like to depend on HDRIs, so here’s my light setup without any directional lights and using just a simple lighting HDRI.After this, I added my fill lights:Then I added a few more key lights to just brighten everything up:Here are my lights. ConclusionA part of the appeal of creating weapons is to stand out, and I think that’s the most important factor if you want to get more reach. Most of the weapons have already been made over the past 30 years of first-person shooter history. The trick is to either make the best version yet or at least add some elements of your own that no one has thought of before. I added the sticker and markers as part of that, not to mention the environment renders compared to renders on a flat white or black background.One of the biggest challenges is making your work stand out. For example, when someone searches for an M4 on ArtStation, there will be a thousand thumbnails that appear, but the one from KEAL will always stand out to both beginners and veterans alike.Ayush Banik, Hard Surface ArtistInterview conducted by Gloria Levine

Stack-O-Bot, originally released in 2022, has been completely rebuilt for Unreal Engine 5.6. This update introduces a range of new gameplay features, including dynamic cameras and intelligent NPCs powered by StateTree. It also expands the project with advanced systems like Procedural Content Generation (PCG) Chaos Physics and Destruction, and more.This free sample project is perfect for newcomers to Unreal Engine as well as seasoned developers eager to explore the latest features. Built entirely with Blueprints, Stack-O-Bot is fully customizable without writing a single line of code, and developer comment bubbles throughout the project provide clear insights into how everything works. At the heart of the sample is a fully rigged third-person character capable of moving, jumping, and interacting with the environment.Stack-O-BotStack-O-BotStack-O-Bot empowers you to build intelligently with minimal effort and uses Level Instances, Hierarchical Level of Detail (HLOD), smart UV-free materials, and integrates a variety of physics-based features, including telekinetic physics grab, destructible environments, and AI-driven NPCs designed for interaction. The sample also demonstrates lots of different ways to keep your players engaged, and it should run smoothly across everything from current-gen consoles to mobile devices.Epic Games is inviting everyone to join the Mega Stack-O-Jam, where you can work on an Unreal Engine project using the new Stack-o-Bot, learn from the experts behind it, and compete for prizes. Hurry and register by clicking this link.Download Stack-O-Bot 2.0 here and join our 80 Level Talent platform and our Discord server, follow us on Instagram, Twitter, LinkedIn, Telegram, TikTok, and Threads, where we share breakdowns, the latest news, awesome artworks, and more.

Check out the changes to the 3D scanning app, including new Auto Tiling and Fold Removal filters, and the option to flatten material layers.