Blender и Hunyuan - это революция в мире 3D!

AI for 3D is no longer a myth – neural networks can already assemble 3D models! Blender + Hunyuan Free Blender mini-course - https://cloudlessons.ru/mc/blender-mi... Useful links for the video: https://3d.hunyuanglobal.com/ https://extensions.blender.org/add-on... In this video, we'll show you what 3D modeling with neural networks looks like in practice: we'll take a concept image, upload it to Hunyuan 3.0, get a finished 3D model, retopologize it using a neural network and manually in Blender, bake textures, and polish everything to perfection in Substance Painter. First, we create a concept: we take the hyped-up idol emoji, find a reference, and refine it using ChatGPT. This approach demonstrates how to link different neural networks in a single pipeline: one generates a 2D image, the other a full-fledged 3D model. After that, we go to the Hunyuan website, use the image-to-3D mode, adjust the parameters (geometry + texture, up to 1.5 million triangles), and obtain a detailed high-poly model. We evaluate the mesh quality and symmetry. This is a great example of how AI for 3D is already saving hours of manual work. But the magic of neural networks is only half the battle. The mesh is heavy, so we need retopology. First, we try automatic retopology directly in Hunyuan using 3D Smart Topology. Next, it's the classic approach: manual retopology in Blender. We import the model, create a low-poly container, set up bindings, disable high-poly selection, and enable the polyquilt add-on. When the low-poly model is ready, we unwrap the UVs, set up the Mirror modifier, align the vertices along the axis, and get a neat, symmetrical model. After that, we move on to the best part: texture baking. In Blender, I show you how to set up the Cycles renderer, select the Bake mode, and bake the diffuse and normal maps from the high-poly neural mesh onto our optimized blank. We cover important nuances: extrusion distance, object selection order (high first, then low), working with Image Texture containers, correct color spaces for normal maps, and connecting them via a Normal Map node. Then we transfer the result to Substance Painter: import the low-poly model, load the maps, assign them via a fill layer, and set the color space. I demonstrate how to find and edit artifacts, work with projections, switch between 2D and 3D views, refine details, and mask problematic seams. This is an important step to ensure that the "neural network + Blender" combination produces a truly finished asset for a game or animation, not just a rough draft. As a result, you get a clear, coherent pipeline: from the idea and the neural network to an optimized model with diffuse and normal maps, ready for use in the project. This type of 3D modeling with neural networks is perfect for both 3D artists and game developers who want to speed up content production without losing control over quality. #blender #ai #3d