A 2023 Reddit post describes disappointing results from the Veras plugin for Revit. Today's search results place that old report beside current Veras product pages, a Veras 4.5 video, and recent release material. A reader can absorb all of them as if they describe one stable tool.
They do not. Chaos says Veras 4.5 added Smart Selection, shareable gallery links, background removal, upload history, revised sketch tools, and source-image descriptions. Its current update page now presents Veras 5.0, including image-to-3D functionality in beta, new image and video engines, reference-image insertion, depth of field, Rhino sync changes, and interface updates.
That product movement does not prove an old critic would like the current release. It proves the old test and current product are different objects. A fair answer requires a retest designed to expose both improvement and persistent failure.
Preserve the old result as a baseline
Do not erase the previous verdict or mock the operator. Record the date, host application and version, Veras release if known, source model, view, settings, prompt, number of attempts, selected outputs, and the complaint in concrete terms. “Disappointing” might mean geometry drift, weak materials, generic styling, unusable resolution, slow iteration, or a poor fit with the intended stage.
If the original files are unavailable, label the baseline incomplete. A public screenshot can identify a concern, but it cannot reconstruct hidden settings or rejected generations. Treat the community report as a reason to test, not a measured benchmark.
Version drift is not a rebuttal to old evidence. It is the reason to collect new evidence beside it.
Build one small scene with five frames
| Frame | Test | Pass question |
|---|---|---|
| 01 | Clean exterior model view | Does primary geometry remain legible? |
| 02 | Same view, named material change | Does the edit stay within its target? |
| 03 | Interior with repeated elements | Are bays, lights, and openings consistent? |
| 04 | Thin details and signage | Which small facts drift or disappear? |
| 05 | One approved output handed onward | Can another person continue the work? |
Use a non-confidential model that resembles studio work without becoming a beauty contest. Include repeated windows, one difficult junction, glass, planting, a readable sign, thin railings, and a background edge. Save the host view, raw export, prompt, mode, engine, settings, credits consumed, generation time, and every output. Do not show only the winner.
Run the old release again only if it remains safely available and the comparison matters enough to justify the setup. Otherwise preserve the historical result and execute the protocol on the current release. The aim is not to manufacture a clean A/B test from missing evidence. It is to state exactly what changed between observations.
Frame 01: test the model before the style
Begin with a restrained prompt and the native host connection the office would actually use. Check the silhouette, opening count, floor lines, roof edge, major material zones, camera, and context boundary. Record the first output as well as the best of a fixed attempt count.
Chaos describes Veras as integrated with design tools and with Enscape, V-Ray, and Corona. Its current product page lists Revit, SketchUp, Rhino, Archicad, Vectorworks, Forma, and Allplan, plus a web app. Confirm the exact office host and supported version during the trial. A broad compatibility list is not evidence that one office configuration works correctly.
Do not score realism yet. If the building facts move before any targeted edit, the later visual polish cannot rescue the workflow for fidelity-sensitive work.
Frame 02: test selection as a boundary
Make one named change, such as converting a selected cladding zone from pale stone to dark metal. Veras 4.5 documentation says Smart Selection can target categories, objects, materials, or a prompt-defined subject, and pairs selection with Modify and Replace modes.
Capture the selection map, selected mode, prompt, source, and result. Inspect the target, its edge, reflections of the target, and everything outside it. Modify should be assessed for preservation as well as whether the new material looks attractive. Replace should be judged as a more permissive operation, not compared as if it promised the same constraint.
Repeat the edit on one easy object and one ambiguous boundary. The second case shows whether selection reduces manual work or merely moves masking work into cleanup.
Frames 03 and 04: count repeated and thin facts
Repeated architecture exposes local plausibility that breaks globally. Count window bays, downlights, chair legs, balusters, mullions, and ceiling modules in source and output. Mark additions, deletions, mergers, and spacing changes.
Then inspect thin details and literal content. Signage, handrails, parapets, door hardware, joints, cables, and distant rooflines should each receive a pass, repair, or fail status. Separate problems that affect design meaning from cosmetic artifacts. A bent plant stem is not equal to a lost guardrail.
Current vendor pages claim faster idea generation and better design continuity. The protocol does not accept those claims as results. It turns them into local questions the trial can answer for one office scene.
Frame 05: test the exit, not only the image
Choose one acceptable output and hand it to a colleague who did not run the session. Ask them to locate the source view, settings, prompt, selected engine, alternatives, and approval note. Have them make one controlled revision and return it without private instructions from the first operator.
If the office plans to use share links, test access from an account-free browser, revocation, cloning behavior, and what project information appears. Chaos says Veras 4.5 introduced shareable gallery links that can be viewed without an account and cloned into another gallery. That is a product claim to verify under the office's own confidentiality rules, not permission to share client work.
If image-to-3D is part of the reason for retesting, evaluate it as a separate beta workflow. Chaos says Veras 5.0 can match an image object to a Cosmos asset or generate geometry, then place it in Revit, Rhino, or SketchUp at a stated scale and position. Check category, dimensions, orientation, materials, editability, file weight, and office standards before calling the result BIM-ready.
Score the delta, not the release number
Compare the new record with the specific old complaints. Mark each concern improved, unchanged, worse, or untestable. Add new costs introduced by the current workflow: credits, cloud handling, setup, review time, cleanup, dependency on a particular host, and training.
A newer feature that does not address the original failure is not an improvement on that measure. A product can gain video, sharing, and 3D functions while still missing the geometry threshold for one team's final imagery. It can also become useful for early material options even if it remains unsuitable for documented design development.
The conclusion should name the approved role: reject, monitor, ideation only, targeted editing, client options, or production candidate with controls. Add the exact release and review date. Schedule another look only if a named future capability could change the decision.
Our take: old skepticism deserves a better test
Software marketing wants every release to reset the conversation. Online memory wants one bad screenshot to settle it forever. Neither position helps a practice decide.
Keep the disappointing result. Put a current version through the same five frames. Count what survives, name what changed, and approve only the role the evidence supports.
The version number gets no vote.
Editorial basis: the 29 September 2026 ArchiGen AI intel sweep, including a historical community report about Veras for Revit, the official Chaos overview of Veras 4.4 and 4.5 features, the current Veras update page, and the official Veras 5.0 release article. Vendor sources establish advertised features, not independent performance. This article reports no hands-on test. The five-frame protocol and acceptance labels are editorial recommendations.