SketchUp Diffusion is no longer called Diffusion. Trimble moved it out of Labs and rebranded it AI Render, with a redesigned panel and new controls — reference images, inpainting, negative prompts. What it does is unchanged in principle: it takes your current viewport plus a text prompt and returns a styled image. It is good, sometimes very good, at making a model you have already built look like something. It is not built for the part of the job that happens before the model exists — and it is metered at 10 credits per generation, which is the number that decides whether it can carry a concept phase.
This article is about that one tool: what it is now, what it costs, where its ceiling sits. For the wider process question — where AI fits before, alongside, and after SketchUp across a concept phase — see AI + SketchUp: concept to render workflow. That article prescribes spending a day in AI before you open SketchUp; this one explains why AI Render structurally cannot be the tool for that day, and what can.
What is SketchUp Diffusion now, and what does it actually do?
AI Render runs on SketchUp for Desktop, SketchUp for iPad, and SketchUp for Web, requires an active Go, Pro, or Studio subscription and an internet connection, and works with SketchUp 2024 and later. On Desktop it is a toolbar extension managed through Extension Manager; on iPad and Web it sits in the Expanded Toolset.
The input is your active camera view. You point the model where you want it, pick a style, write a prompt, and generate. The current control set includes:
- Style presets — modern, sketch, watercolor, photoreal, and others.
- Prompt strength, which trades geometry adherence against stylistic freedom.
- Negative prompts, for excluding things the model keeps adding.
- Reference images, to push the output toward a look you already have.
- Inpainting tools — Sketch, Paint, Erase — for adding or removing elements in a region rather than regenerating the whole frame.
- Overlay and save, so the result can be compared against the viewport and exported.
SketchUp documents that the tool is trained in English and performs best with English prompts; it may recognise some non-English keywords, but that is not the supported path.
An independent 2026 field test (Just Necessary, AI Rendering for SketchUp 2026) reports roughly 1024×768 output in 15 to 30 seconds per generation — observed numbers rather than a published spec, but the right order of magnitude. That test is also the source for the strengths and the failure modes cited below.
The useful mental model is this: AI Render is a styling layer over geometry you already have. Everything it is good at follows from that, and so does everything it cannot do.
When should you just use AI Render?
There is a real set of situations where reaching for another tool is pure overhead. Named plainly:
The model is built and you need a mood image tomorrow. No export, no upload, no second application, no second login. You are already in the file. For a single presentable view off an existing model, that convenience is worth more than any feature comparison.
You want a watercolor or sketch board for a schematic review. The same field test that was blunt about geometry drift called the watercolor and sketch presets the best in the field for early-design moodboarding. That is a genuine strength, and it is the exact register a schematic-stage client review wants — atmosphere over precision, drawn over photographic.
You are working on iPad. AI Render ships in the Expanded Toolset there. Very little else in this category runs natively on the same device as your model.
You want to fix one thing in an image you otherwise like. Inpainting on the generated view — erase the parked car, repaint the cladding on one wall — is faster than regenerating and hoping.
If your week looks like any of those, open the panel and stop reading — nothing below argues you need a second tool for that job. The rest of this article is about the jobs outside that envelope.
How many generations do you actually get?
This is the number most reviews skip, and it decides more than any feature list.
As documented by SketchUp in July 2026, AI Render costs 10 credits per generation — every time you produce output, whether from a text prompt or from the Sketch, Paint, or Erase tools. Monthly credit allowances come with your subscription:
| SketchUp plan | Monthly AI credits | AI Render generations |
|---|---|---|
| Free | 0 | 0 |
| Go | 100 | ~10 |
| Pro | 150 | ~15 |
| Studio | 200 | ~20 |
Those credits are shared with the other AI features, not reserved for rendering — Generate Object costs 30 credits per use, and AI Help is free. A separate SketchUp AI subscription adds 1,500 monthly credits if the included allowance is not enough.
So the precise version of “AI Render is free” is: no extra charge beyond the SketchUp subscription you already pay for, metered per generation, and unavailable on free SketchUp. Both halves matter.
Now hold that against what a concept phase consumes. Eight exterior directions, four refinements of the two the client liked, four interior studies, a few material variants — comfortably 20 to 30 generations before anything is resolved, and most of them get thrown away. That is what exploration is: producing work you discard. On Pro, the monthly allowance covers about half of one such pass, and the credits are gone whether the output was useful or not.
This is not a criticism of the pricing. It is a statement about scope: ten to twenty generations a month is a budget for finishing, not for searching.
Why can’t a viewport tool help before the model exists?
Because it renders the active camera view. There has to be something in front of the camera.
That sounds obvious, and it is, but it is also the whole limitation. The point of AI concept exploration is to answer “what should I model?” — and a tool that requires a model to produce output cannot answer that question. You can rough in massing blocks and generate from those, and people do; but at that point you are modelling in order to prompt, which inverts the time saving you were after.
This is the clean line between the two articles on this blog. AI + SketchUp: concept to render workflow argues for spending a day generating directions before you open SketchUp. AI Render cannot serve that day — not because it is weak, but because it is a viewport tool and there is no viewport yet. It serves weeks five and six of that workflow, not weeks one and two.
If the distinction between the two phases is unfamiliar, AI rendering vs concept design sets out why they are different problems with different tools, and what AI concept design actually means covers what the earlier phase produces.
What happens to a project that spans exterior, plan, and interior?
AI Render works one view at a time, and it does not know that two views belong to the same project.
There are no plans. AI Render styles a raster of a viewport. SketchUp generates plans from the model natively; dedicated concept tools generate schematic plans from a brief. AI Render does neither. Plan exploration has to live somewhere else in your process.
Style continuity is manual. Nothing carries from one generation to the next except what you retype. If the facade came out with board-formed concrete and warm oak, the interior render will match only to the degree your prompt reproduces it — and the field test’s observation that even explicitly painted SketchUp materials (asphalt versus grass) come back unreliably suggests the model is not reading your material assignments as instructions. Reference images help, but you are re-supplying context on every generation.
That per-generation re-setup is the tax, and it compounds across a project. We have written about why separating exterior, plan, and interior into different tools breaks a project and about the mechanics of keeping AI designs consistent across a whole project. The short version: consistency you have to re-establish by hand is consistency you will eventually lose.
What happens to the alternatives you did not pick?
They mostly disappear.
The panel keeps recent output within a session, but there is no structure recording that variant C grew out of variant A, or that the four material studies were all branches off the same massing. Two weeks later, when the client says “what was that second one, with the deeper overhang,” you are reconstructing it from a prompt you no longer have — at 10 credits a try.
Exploration only works when discarded directions stay reachable. Otherwise every generation is a commitment, and people generate conservatively to protect their budget — exactly the wrong instinct at concept stage.
What does a dedicated concept tool do differently?
Nuit is built for the phase AI Render cannot enter: the design does not exist yet, and it has to end up coherent across several kinds of output.
It starts from a brief, not a viewport. Describe the site, programme, and constraints and get exterior directions back with no geometry involved. Reference images are optional inputs, not prerequisites.
Modes belong to one project. Exterior, floor plan, and interiors are parts of the same project, and images you save as project references feed subsequent generations automatically — so material and style decisions carry from facade to plan to room without being retyped.
Every image is a branch point. The exploration tree is preserved. Branch from the version you liked three days ago and both directions survive; the lineage shows what grew from what.
Three models, chosen per job rather than ranked. Light is faster and costs half what Nuit N or Nuit F do per image, which is what you want for a wide first pass where most of the output gets discarded. Nuit N and Nuit F behave differently and give a different character of result — neither is the better one, and which you prefer on a given project is a matter of taste. We did not build these models; we selected them for the jobs the tool has and put them behind a picker.
And the honest inverse: Nuit is not a SketchUp plugin. It does not read your model, it does not render your viewport, and it does not lock geometry. If your input is a finished model, AI Render is closer to your file than we are.
Which tool fits which job?
| The job | Better fit | Why |
|---|---|---|
| Style a view you have already modelled | SketchUp AI Render | No export, no second tool; works on the active viewport |
| Watercolor or sketch board for a schematic review | SketchUp AI Render | Preset quality independently rated best in field for early-design moodboarding |
| Fix one element in an image you otherwise like | SketchUp AI Render | Inpainting works directly on the generated view |
| Working on iPad without a second login | SketchUp AI Render | Ships in the Expanded Toolset |
| Decide what to model in the first place | Dedicated concept tool | Viewport generation needs geometry that does not exist yet |
| Schematic floor plans at concept stage | Dedicated concept tool | AI Render has no plan output |
| Keep exterior, plan, and interior coherent | Dedicated concept tool | AI Render is per-view, with no project-wide references |
| Compare six directions and keep the five you rejected | Dedicated concept tool | Branching lineage vs. a flat session history |
| Run 40–60 generations in one concept week | Dedicated concept tool | 10 credits per generation caps Pro at ~15 per month |
| Strict line-for-line geometry lock | Neither | AI Render reinterprets; concept tools do not lock geometry either — use a render engine or a geometry-locking render tool |
That last row matters. Neither category solves it, and any article that awards it to one side is selling you something.
How do the two fit together on a real project?
The handoff is clean because the tools sit on opposite sides of the model. Concept first, without geometry — brief to directions, plans and interiors in the same material language, rejected options kept in the tree. Then model the chosen direction in SketchUp properly. Then AI Render, from inside the file, on the views you actually need to present.
Used this way the credit allowance stops being a constraint, because the exploration that would have burned it happened before the model existed. Fifteen generations a month is a comfortable budget for finishing; it was never a budget for searching.
The honest limits on our side
Nuit’s output is concept-grade. Schematic plans, not dimensioned construction drawings; atmospheric renders, not photometrically accurate ones. We do not read SketchUp files, we do not render your viewport, and we do not offer geometry lock — if you need a render that preserves a finished model line-for-line, neither we nor AI Render is the tool, and you should look at a render engine or a geometry-locking render tool.
There is also a real cost to adding a tool: another login, another place your work lives, another thing to learn. If your concept phase is short and mostly happens in your head before you model, that cost may not be worth paying — and plenty of architects work that way.
Conclusion
SketchUp Diffusion — AI Render, now — is a good tool wearing a small envelope. It styles a view you have already modelled, it ships where your model already is, its sketch and watercolor presets are genuinely strong for early boards, and it costs nothing beyond a subscription you already have. For finishing work that is a good deal, and no comparison table changes it.
The envelope closes in three places: it needs geometry, so it cannot help you decide what to build; it works per view, so a project spanning exterior, plan, and interior stays coherent only by hand; and at 10 credits a generation it budgets for images you keep, not images you throw away. Concept design is mostly the second kind.
Decide with one question: does the thing you want to see already exist in your model? If yes, AI Render is where to do it. If no, that is a different tool’s job.
Frequently Asked Questions
Is SketchUp Diffusion still called Diffusion?
No. Trimble moved the feature out of SketchUp Labs and rebranded it as SketchUp AI Render. It is the same lineage of tool — generate an image from your current viewport plus a text prompt — with a redesigned interface and added controls including reference images, inpainting, and negative prompts. Most search traffic and most forum posts still call it Diffusion, so the two names refer to the same thing.
How many AI Render generations do I get per month?
Each AI Render generation costs 10 AI credits, and monthly credit allowances are tied to your SketchUp subscription: 0 credits on free SketchUp, 100 on Go, 150 on Pro, and 200 on Studio (as documented by SketchUp in July 2026). That works out to roughly 10 generations a month on Go, 15 on Pro, and 20 on Studio — and those credits are shared with other AI features, such as Generate Object at 30 credits each. A separate SketchUp AI subscription adds 1,500 monthly credits if you need more.
Is SketchUp AI Render free?
It is included at no extra charge with an active SketchUp Go, Pro, or Studio subscription, but it is metered — every generation draws 10 credits from your monthly allowance. Free SketchUp accounts get zero AI credits, so AI Render is not usable there. Calling it free is accurate only in the sense that you do not pay a separate line item for it.
Can SketchUp AI Render generate floor plans?
No. AI Render styles the active viewport of a model you have already built — it is an image layer over existing geometry, not a plan generator. SketchUp itself produces plans from the model, and dedicated AI concept tools generate schematic plans from a brief, but AI Render does neither. If plan exploration is part of your concept phase, it has to happen somewhere else.
Does AI Render preserve my SketchUp geometry exactly?
Not reliably. It is a prompt-influenced reinterpretation of the viewport rather than a geometry-locked render. An independent 2026 field test by Just Necessary found that in four of twelve generations the tool invented a different building, front stairs turned into a ramp in three, and second-floor windows collapsed into solid wall in two. Wide establishing shots hold up better than close-ups. For line-for-line geometry fidelity you need a render engine or a geometry-locking render tool, not AI Render.
When is SketchUp AI Render enough on its own?
When the model already exists and you want it to look like something — a mood image for tomorrow’s review, a watercolor or sketch board for a schematic presentation, a quick material or time-of-day variant, or a small inpainted fix to an image you otherwise like. If that describes your use, AI Render is the right answer: no export, no second tool, no second login, and it runs on Desktop, Web, and iPad.
Do I need a separate AI tool if I already have SketchUp?
Only if your work starts before the model does. AI Render needs geometry to render, so it cannot help you decide what to model, generate schematic plans, keep exterior, plan, and interior coherent across one project, or preserve the alternatives you explored and rejected. Those are the jobs a dedicated concept tool exists for. If you mostly style models you have already built, adding a tool is unnecessary overhead.
Try Nuit free — 10 generations, no card required. Run one brief through exterior, plan, and interior to see whether project-level consistency changes how your concept phase runs. Start your project →