Render engine for architects: Lumion or Corona?

TL;DR

• Lumion Pro runs at US$ 1,149 a year and runs on its own. Corona Solo runs at US$ 414, but it only works inside 3ds Max or Cinema 4D, and the host application dominates the bill (lumion.com and chaos.com, consulted 23 September 2026).
• Project phase is a better criterion than the feature list. A scheme-approval image is due in hours. A publication image is due in days, and only that one pays for the photoreal route.
• Since 27 May 2026 Chaos ships generative AI with every Corona subscription and charges part of the usage in credits (CGChannel, 27 May 2026). Rendering software now carries a variable line that studio budgets never had.

The feature comparison hides the bill

The question usually reaches a studio in the same shape. The practice is closing the year, there is capital budget left, and someone suggests subscribing to a render engine to stop outsourcing imagery. The debate then turns into Lumion's real time against Corona's photorealism, argued feature by feature.

The feature list is the least decisive part. Lumion Pro costs US$ 1,149 a year on a named-user licence, or US$ 1,499 on the floating seat that circulates across the team's machines (lumion.com, consulted 23 September 2026). Corona Solo costs US$ 414 a year (chaos.com, consulted 23 September 2026), which appears to settle the argument in favour of the second.

Except that Corona is not an application. It is an engine that lives inside 3ds Max or Cinema 4D (chaos.com, consulted 23 September 2026), and most residential practices run neither. A 3ds Max annual subscription sits at US$ 168 a month (Autodesk, consulted 23 September 2026), which Super Renders Farm put at US$ 2,010 a year on the standard plan (15 September 2026). Cinema 4D lands lower, around US$ 839 a year billed annually, according to a Drop and Render breakdown of 17 April 2026.

Add it up honestly and entry into photorealism costs two to three times entry into real time. The larger gap never appears on the invoice. It is the person who will operate 3ds Max in a practice whose models come out of SketchUp, Revit or Archicad.

Project phase is the criterion that holds

A practice produces imagery at two moments that have nothing in common.

The first is scheme approval. The client needs to read the site strategy, the massing, the relationship with the ground, the ceiling height, the feeling of walking in. The image is due before the next meeting and will change three times before then. Real time wins here by definition, because the loop between moving the model and seeing the result is the product being sold. Lumion shipped 2026.0 on 24 February 2026, with a rebuilt library of trees and outdoor objects plus fill-in clipping planes, and 2026.1 on 14 July 2026, which added a pattern editor for tile layouts and brick bonds inside the application (Lumion release notes). On 23 March 2026 Lumion View arrived in Archicad, putting a ray-traced viewport inside the modeller.

The second moment is the publication image: the one that goes to ArchDaily, to an award submission, to the portfolio that wins the next commission. It is due in days, it passes through art direction, and it has to survive enlargement. That is where the photoreal route pays for itself.

Which leads to the part that tends to start an argument among partners. The scheme-approval image should not be photoreal. The more finished the image that reaches a client early, the more the conversation drifts towards curtains, joinery colour and armchair models, and the less room is left for massing and site strategy, which are the expensive decisions to reverse later. An image that is too beautiful too early does not speed approval up. It drags the finishes conversation forward into a moment when the design cannot yet support it.

What changed on both sides in 2026

Corona 15 shipped on 27 May 2026 and now includes Veras, Chaos's own generative AI ideation tool, with every subscription. An artist sends a frame straight from the render buffer into Veras and generates material and lighting variants faster than the engine alone would allow. The release also improved Gaussian splat integration, so imported splats now bounce global illumination light onto surrounding geometry, and it cut animation times in 3ds Max by reprocessing only what changes between frames (CGChannel, 27 May 2026).

Alongside it came the change that deserves the attention of whoever signs the invoice. Chaos introduced a hybrid licensing system built on credits (Architosh, 28 May 2026). AI features consume credits, and extra credits are bought separately, in packs listed at US$ 77 and US$ 208 (chaos.com, consulted 23 September 2026). Rendering software stopped being a fixed cost and became a fixed cost with a variable tail that depends on how much the team leans on AI in a given month.

On the real-time side the pressure comes from elsewhere. The first page of search results for architectural rendering is now crowded with AI image generators promising photorealism in thirty seconds. None of them replaces either engine in authored work, but they destroy any commercial argument built on raw speed. Where that line falls in practice is covered in what moves the price of an architectural render, and the disclosure question it opens is covered in the AI listing rules that skip the off-plan rendering.

Three questions before signing anything

How many publication images does the practice produce a year? Under twelve, the sum of a photoreal engine plus its host plus the learning time rarely beats outsourcing those twelve. At forty, it wins comfortably.

Where does the model come from? A practice modelling in SketchUp or Archicad with nobody on 3ds Max buys two problems when it buys one. A practice that already has a 3ds Max artist buys half of one.

Who operates it during delivery weeks? A render engine is not a partner's tool. It needs someone with protected calendar time, and that time is the first thing invaded when drawings run late. With no name against it, the subscription becomes a standing charge for idle software.

When buying is the wrong call

One arrangement solves most mid-sized practices and almost never shows up in the comparison: real time in house for the design phase, photorealism contracted out for the image that will be published. The studio keeps the cheap licence, the short learning curve and control of the phase that changes every day. The image that has to survive enlargement and art direction goes to people who produce those in series.

Buying a photoreal engine is rarely a design decision. It is a portfolio decision, and the portfolio is what brings in the next commission. Treated as a design decision, it reaches the budget with the wrong number and the calendar with the wrong person.

Frequently asked questions

Can Lumion replace Corona in a practice portfolio?

For study work, client presentations and walkthrough video, comfortably. For the still image that will be enlarged, published and judged beside construction photography, the photoreal route still offers a margin of control that real time does not reach. The practical split is by the purpose of each image.

Can Corona be used without 3ds Max?

Yes, inside Cinema 4D. Chaos lists 3ds Max and Cinema 4D as Corona's host applications (chaos.com, consulted 23 September 2026). There is no standalone version, which is why the US$ 414 a year Solo licence does not describe the real cost of adoption.

Does the generative AI in Corona 15 remove the need for the engine?

No. Veras works as an ideation tool, starting from an already rendered frame to generate mood and material variants (CGChannel, 27 May 2026). Geometry, scale and correspondence with what will actually be built still come out of the engine. What changed is that part of that usage now consumes credits bought separately.