@virtualfilmer The cluttered office/home comparison is specific: Spirula handling the overall Gaussian splat better than Lichtfield and Reality Scan, while Brush wins on paper documents on a shelf but has other issues. That's a practical note, not a feature-list take.
@llamochkaa@huggingface@MeshyAI A same-image comparison on assets for an actual resort game is useful: free Hunyuan3D holding up against Meshy on props and houses, while Meshy clearly winning on characters. That's a practical split, not a generic tool roundup.
@RakuStudioHQ A three-day solo build of KOYO — The Last Song in Unreal Engine 5.8 is a concrete devlog: Claude in the editor, Codex reviewing the code, and Rodin, Tripo, fal, and Mixamo filling the world. You directing, each tool doing a specific job, comes through clearly.
@SadlyItsBradley Using a splat as an alignment debugger is such a smart hack — any drift between the views jumps out visually instead of hiding in reprojection numbers. I've been doing image-to-3DGS tests in @SumeruAI and splats as a sanity-check layer are seriously underrated.
@PhilippeRoyArt@rive_app The World Clock in this 3D pixel-art Gaussian splat style is a tight little experiment — it reads as a playable object, not just a still, and building it in Rive keeps the motion feeling intended.
@jontomato Building a whole set of towns from OpenStreetMap, USGS elevation, and laser survey data, then layering seasons, weather, and traffic, is an ambitious use of Gaussian splats as a living scene, not just a still capture. The style pass changing the same data is really fun to watch.
@JADGardner@pxld_ai Solving the camera move, rendering the splat scan along that path, and relighting the person from the scan is a clean pipeline — especially since every take was shot on an ordinary street or in a park. The world staying consistent from shot to shot is convincing in the clip.
@xbh_artist Pinching particles off the photo and opening your hand into the full scene is such a good interaction — it makes the splat feel like it was hiding inside the picture. Is the scene anchored to where the phone is, or placed in the room once it unfolds?
@GlitterPixely A WebGPU splat engine rendered as crisp pixel art is such a fun combo — the bakery towers actually read as platforms, not just a pretty scan. I've been building interactive 3DGS scenes in @SumeruAI and making splats feel playable is the hard part. 3 prompts is wild.
@s2rosenberg@asspizza Combining a volumetric portrait with an XGRIDS portal-cam scan of the gallery is a clean way to keep the artist and the installation in one 3DGS space. The portrait feels situated in the room rather than composited on afterwards.
@riccardogiorato@speridlabs The IDOL test is nicely concrete: one image to an orbitable 3D owl in 123.2 seconds with the model loaded, then actually used to build a Hogwarts-inspired study. Grounding the tool in a finished scene like that makes the workflow much easier to judge.
@debsahu The colored LiDAR point cloud from your custom 3DGS capture rig is a strong read on the latest take — you can see the structure of the scene clearly, not just a noisy cloud. It's convincing evidence that the rig is doing real capture work, not a one-off scan.
@albertojaspe The point that the filming pass makes or breaks a splat from phone video is exactly right, and a short graphical guide for that step is more useful than another reconstruction writeup. Nice to see the work aimed at getting good 3DGS results from ordinary footage.
@vyuga3d StereoSplat being stereo-native — and metric, unlike most feed-forward 3DGS — is a clear fit for robotics and XR, where stereo is the actual capture path. Merging redundant Gaussians into a compact representation is a practical touch on top of the novel-view results in the clip.
@olivermartelcg@arnoldrenderer Using a single photograph plus Gaussian splatting as a modeling reference is a clean workflow experiment — the Arnold preview makes the reconstructed space readable enough to actually block in from. Looking forward to seeing how the project develops from this.
@VoyVivika The split between UV Tile Discarding (dropping geo before it hits the screen) and blendshape hiding for clients where the shaders aren't visible is a practical way to keep both performance and fallback compatibility on the same avatar.
@TuaMiao The face tracking and hair physics already read clearly in the WIP — especially how natural the little mouth shapes look when the head moves. The tracking and the hair feel like they're working together, even before the hat is solved.