Access
Web App | Windows | Mac Sketchup | Mac Vectorworks or ArchiCAD
SketchUp Mac Update Instructions: Updating Veras for SketchUp on Mac
SketchUp Mac First Install Instructions: Installing Veras for SketchUp on Mac
Vectorworks Mac Install and Update Instructions: Getting Started with Veras for Vectorworks
ArchiCAD Mac Install and Update Instructions: Getting Started with Veras for ArchiCAD
Veras 5.0 introduced the MODEL tab and the ability to turn a rendering into real 3D geometry. Veras 5.1 builds on it: a lighter way to get that geometry into your host model, the ability to reuse assets you have already generated, a new asset engine, and a substantially faster 3D view.
Proxy Bake — Real Assets, Light Models
A Veras asset can carry a million triangles and its own textures. Baking several of those into a host model adds significant file size and slows the model down.
Proxy bake writes a small, untextured placeholder into your host model instead of the full mesh, stamped so that Enscape renders the full asset where the proxy stands. You get the same result in an Enscape render at a fraction of the file weight.
- Revit — the Bake format dropdown gains a Proxy row, and it is the new default. OBJ and DXF are still available.
- SketchUp and Rhino — a new Bake geometry dropdown in the same spot: Proxy (default) or Full geometry. Veras remembers your choice.
- Cosmos assets bake as the Cosmos proxy — the same low-poly outline Enscape places itself. Generated meshes bake as a simplified silhouette of the placed model.
- The Veras 3D view always shows the full model. The proxy is what goes into the file, not what you design against.
- Cards read “Baked to model (proxy)”, and re-baking with Full geometry selected replaces the proxy with the full mesh.
- Refresh the model and Veras still recognizes its own proxies, re-links them to the rendering they came from, and loads the full mesh back into the 3D view so you can keep editing.
Proxy baking requires the Veras plugin at 5.1 — install this release for your host and the option becomes available.
NOTE: Baking assets into the host model is supported in Revit, Rhino and SketchUp. Vectorworks and ArchiCAD support is not available yet.
Place from Gallery — Reuse a Generated Asset
Place from Gallery is a third Asset source, alongside Generate and Search Cosmos library. Point a detected object at a 3D model already in your Gallery and Veras places it the same way it places a Cosmos pick, instead of generating a new one.
- Free, and it does not use a queue slot — no generation, no wait
- Choose the model yourself from the same window the Cosmos results use, with a search box that filters your models as you type by name, category and engine
- Load more pages through your full 3D model history
- Placed at the detected spot at the correct size and facing, with the usual rotate, scale, move and bake behavior
- Every asset now gets a name and a category on every host, so models you generate today can be found and baked later. Older models without a category are updated the first time you select them.
NOTE: Click an object’s thumbnail circle to open the picker on any row — the same action that swaps a Cosmos match
Microsoft TRELLIS 2.0 — A New Asset Engine
Microsoft TRELLIS 2.0 joins the 3D GEN engine list. It produces sharper geometry with full PBR materials than TRELLIS 1.0.
The 3D GEN engines also now use the names Chaos uses in its other products:
| Before | Now |
|---|---|
| Trellis | Microsoft TRELLIS 1.0 |
| (new) | Microsoft TRELLIS 2.0 |
| Hunyuan 3D 3.1 | Tencent HY 3D 3.1 |
- No new agreement to accept. TRELLIS 2.0 uses the same 3D GEN agreement you have already accepted.
- Quota limits are now explained clearly. Some asset engines are rate-limited per user. Instead of a generic error about 30 seconds after you click, you get a dialog with a countdown, how much of your allowance you have used, and a Generate button that becomes available when the limit resets and retries every object that was refused as a single batch.
A Faster 3D View
Two performance problems in the 3D viewer are fixed in this release.
- Hovering over a heavy asset no longer slows the viewer. Every mouse move tested the cursor against every triangle of every placed asset, which costs about 140ms per mouse move on a million-triangle mesh — roughly 7fps. Veras now builds a spatial index for heavy geometry when it loads, and the same test costs 0.03ms. Hover and selection stay responsive regardless of asset size.
- The old “hide geometry” optimization is replaced. It hid geometry on almost every frame, on models of any size. The new interaction LOD engages only on genuinely heavy models, only while the camera is moving, and only on geometry that is small on screen. It always restores exactly what it hid.
What this changes in use:
- No visible pop when switching between orbit and scroll-wheel zoom
- The viewer no longer freezes after a scroll during a drag, or after releasing the mouse outside the canvas
- Renders started during camera movement no longer capture frames with geometry missing
- Middle-click pan and touch drag end correctly
- If your graphics driver drops the 3D context, Veras now reports it instead of showing a blank canvas
More Accurate Asset Placement
- Assets are placed on the surface under the cursor, not on geometry above it. Dragging an object across muti level floors could place it on the roof directly overhead. This happened anywhere there was geometry above the cursor, and is fixed for both drag-to-move and click-to-place.
- Imported GLBs use the size stored in the file. An imported file was previously sized from the current camera distance, so the same file could land at a different size each time.
- Imported GLBs are now fully editable — rotate, scale, move, reposition and delete, the same as a generated asset. Previously an import could not be adjusted at all.
- Transforms moved onto the 3D canvas, so assets are adjusted directly in the view.
More Reliable Baking
Large assets could report that the bake timed out even though the asset had actually been written into the model, and retrying then produced a second copy.
Veras now gives a bake the time it actually needs based on the size of the asset. Large assets bake successfully instead of reporting a false failure, and a bake that genuinely fails is reported sooner.
GPT Image Models by OpenAI
- GPT Image 2.5 — a new image engine with its own quality options: Fast, High (the default), Extra High and Max, at 1K or 2K. Fast runs a lighter model at High quality, so it costs the same as High and returns in roughly half the time.
- GPT Image 2 is now available
Grok Imagine 2.0 Model by xAi
- Grok Imagine 2 is now available
- Reference images are supported on both Grok Imagine engines — up to 4 on Grok Imagine 1 Fast and on Grok Imagine 2, and 2 on Grok Imagine 1’s Quality tiers. Design Lock remains a Nano Banana feature.
More Improvements
- The object row’s busy indicator moved outside the thumbnail — a progress ring around the circle rather than a gray overlay on the image, so the thumbnail stays visible while generating
- A selected object’s thumbnail now has a blue ring, matching how an applied preset is marked
- Renaming an object during generation updates the pill on the rendering immediately, instead of waiting until generation finishes
- You can now share 3D models. Sharing a generated 3D model previously returned an error. It now works, and the share dialogs show 3D models with their own icon and label instead of listing them as images.
- Sketches are hidden when you switch renderings. Sketch visibility is now remembered per rendering, so a sketch drawn to guide a render no longer covers the result, and returns if you left it visible.
- The Cosmos search progress bar no longer overlaps the bottom edge of the search field
- The update prompt checks the plugin version you actually have installed, per host application, so web-only updates no longer prompt desktop users and each host is compared against its own version