LookX came up because it was one of the few entries on a top-ten list we audited this morning that was both AI-native and actually about making images of buildings. We had never covered it. This is a first look from the vendor's own material and what is publicly available, not a tested review, and the distinction matters for what follows.
What it is
An architecture and interior design specific image generator, running in the browser, that takes text, sketches, massing studies, diagrams and reference images and returns rendered visuals. Alongside generation it offers video output, image upscaling, and style adaptation.
The integrations are the part that decides whether a tool gets used or admired. LookX publishes plugins for SketchUp and Rhino. That is the whole list. There is no Revit plugin, which we will come back to because for a lot of readers it is the end of the conversation.
There is also a gallery area, called Creations Plaza, where you can look at what other users have made and duplicate an image to work from its settings. That sounds like a marketing feature and mostly is, but the practical use is real: reading somebody else's parameters is the fastest way to learn what a generator responds to.
The differentiator is custom training
Most tools in this category give you prompt fields, style presets and a strength slider. You steer toward a look and accept some drift. LookX lets you upload a series of images with a shared style, train a model on them, and then generate against that model.
For a practice this is the difference between three interesting outcomes and one useful one.
House style becomes reproducible. If your firm has a visual signature, muted palette, particular skies, a certain restraint about entourage, a trained model holds it across a set without being re-argued in every prompt. We have written before about consistency across views being the thing that actually breaks AI renders in front of a client, and training is a more direct answer to it than prompt discipline.
Repeat clients get their own model. A developer with a standard material palette across sites is a good candidate. Train once, reuse across schemes.
Competition boards stop looking generic. The complaint that AI renders all look the same is largely a complaint about everyone sampling the same base model with the same adjectives. Training on your own material is the available escape from that.
A style preset borrows a look. A trained model builds one.
The caveat is that training needs a corpus. If you have a decade of consistent finished visuals, this is a strong feature. If you are two years old and your output is stylistically all over the place, you will train a model on inconsistency and get inconsistency back, faster.
What we could not confirm
The price.
The vendor site has a pricing link in its navigation. When we fetched it, no figures came back. Third-party review sites report a free Basic tier including a starting allocation of what LookX calls X-points, an individual plan around $20 a month with a seven-day trial, an annual plan near $199, and a team plan around $269 a month. We are repeating those as reported, not confirmed, because none of the sites quoting them is the vendor and they are the kind of numbers that go stale quietly.
This is a small thing that tells you something. A tool asking a practice for a recurring commitment should be able to show a number without an account. When it cannot, the evaluation gets slower and the comparison against a seat of something conventional gets harder to put in front of whoever signs.
The reviews are not reviews
Worth flagging, because a reader searching this tool will hit the same pages we did.
The AEC-focused review we pulled describes LookX as built by architects and great for rendering, lists automation, web-based access and collaboration as advantages, maps it to concept, schematic and design development phases, and carries a testimonial saying it transformed how a team approaches visualization. It records no weaknesses. It shows no output. It quotes no price.
That is a directory entry with a compliment attached. It is not evidence about whether the images hold up. Several of the other top results are the same shape. So the honest position on output quality today is that we do not know, and neither does anyone publishing about it in the pages that rank.
The Revit gap
SketchUp and Rhino cover concept and early schematic for a lot of studios, and if that is where you live, the plugin list is fine. LookX is positioned at exactly that stage and the integrations match the positioning.
But if your model of record is Revit, this tool sits outside your BIM-first stack and every image involves an export. That is not fatal, plenty of good work happens through a PNG round trip, but it is a real cost we have measured before against native integration, and it compounds on revision-heavy projects. A tool you have to feed manually gets used for the pitch and abandoned by design development.
How we would test it
If this makes your shortlist, the free tier is enough to answer the only questions that matter, in about an afternoon.
| Test | What it settles |
|---|---|
| Push a real SketchUp massing through the plugin, not a demo file. | Whether it respects your geometry or redesigns it. |
| Generate the same view six times on one prompt. | Variance. High variance means you will burn credits hunting. |
| Train a model on ten of your own finished visuals. | The headline feature, and whether ten is enough or it wants a hundred. |
| Generate three views of one scheme against that model. | Set consistency, which is where these tools usually fail. |
| Count what the afternoon cost in points. | Whether the monthly figure is the real figure. |
That last row is the one people skip and regret. Our trial credit plan covers pacing it, and the benchmark protocol covers keeping the comparison fair against whatever you already run.
Our take
No verdict, because we have not put a project through it and we are not going to pretend otherwise on the strength of a feature list.
What we will say is that custom style training is the right feature to be building, and it is a better answer to the sameness problem than another round of preset packs. If it works at a realistic corpus size, ten or twenty images rather than hundreds, LookX is a legitimate candidate for concept and schematic work in a SketchUp or Rhino office. If it needs a corpus most practices cannot assemble, it is one more generator.
Two things to watch before you commit. Find out what it costs from LookX rather than from a review site. And run the six-generations test early, because a tool that cannot repeat itself cannot hold a style no matter what you trained it on.
The feature list is promising and the evidence base is thin. Go make your own.
Written from the 22 July 2026 intel sweep. Feature claims are as published on lookx.ai and captured on 22 July 2026 and may change without notice. Pricing figures are reported by third-party review sites, were not confirmed on the vendor's own pricing page at time of writing, and should be verified before purchase. This is a first look based on published material, not a tested review; ArchiGen AI has no commercial relationship with LookX and carries no sponsored placements.