A comparison table puts an annual-plan monthly equivalent beside a student subscription, a per-render estimate, and “contact for pricing.” The cells all contain prices. They are not the same unit.
Today’s sweep surfaced a Chaos comparison of six AI rendering options. Its table is useful because it names BIM integration, engine, starting price, best use, and a limitation. It also exposes a recurring procurement problem: the price column mixes different buying conditions. Chaos publishes Veras, one of the compared products, so the page is vendor material rather than independent research. That does not make the mixed-unit problem less important.
Before ranking cost, a studio needs one scenario, one period, and one definition of usable output.
A starting price is a doorway
The smallest advertised number may apply only to annual billing, education, one user, low resolution, limited credits, or a narrow feature set. Another product may charge per image. A third may require a sales quote. Sorting those numbers from low to high creates arithmetic without comparison.
Record the condition attached to every number. Is the plan monthly or annual? Is payment due up front? Is tax included? Who qualifies? How many seats are covered? Which engines, resolutions, export rights, privacy controls, and commercial uses are included? Do unused credits expire? What happens after the allowance?
A missing answer is not zero cost. Mark it unknown and obtain it before a purchase decision.
| Advertised unit | Hidden question | Common unit |
|---|---|---|
| Per month, annual plan | What cash commitment is due? | Annual studio spend |
| Student plan | Does the practice qualify? | Commercial seat price |
| Per render | What counts as one charge? | Cost per accepted image |
| Credit bundle | Which actions consume credits? | Credits per accepted image |
| Contact sales | Seats, term, and support? | Quoted annual total |
Write the studio scenario first
Use a scenario that resembles an actual month. For example: three operators, four active projects, twelve source views, six candidates per view, two revision rounds, and twelve accepted images. State the required resolution, host applications, commercial rights, privacy condition, and deadline.
The numbers are not universal. They are a denominator. Every product now faces the same workload. If a candidate cannot meet one required condition, mark it ineligible rather than pretending a lower price compensates for the missing capability.
Include trial behavior. A free allowance can answer whether the workflow fits, but it should not reduce the projected recurring cost unless the same allowance renews under the intended commercial plan. Likewise, a discount that expires after the first term belongs in a separate introductory column.
Count attempts, not final images
A client receives twelve images, but the studio may generate seventy-two candidates and twenty-four revisions to get there. Cost per delivered image therefore depends on the keeper rate and revision behavior, not only the vendor’s nominal render charge.
During a short trial, record every billed or credit-consuming action: first generation, variation, edit, upscale, video, download, and failed job. Note whether the same action uses a different amount at higher resolution or with a different engine. Then calculate:
Direct cost per accepted image = total product charge for the trial divided by accepted images.
Do not report a projected figure as measured fact. If the free trial masks billing, label the result an estimate and show the assumed rate. If the service does not disclose a needed rate, leave the cell open.
A price becomes comparable only after the workload is identical.
Add seats, hosts, and companion software
A plugin may require software the firm already owns. Another team may need a new host license. A web tool may avoid that dependency but add upload, download, and file-management work. The comparison should show both cases.
List the operator seats needed at the busiest point in the month, not the total staff count and not a single trial login. Check whether seats are named, floating, concurrent, or transferable. If reviewers need accounts to comment or approve, include them. If a shared style, project space, or admin control requires a higher tier, price that tier.
Then list required companion products and add-ons. Do not assign their full cost to AI rendering if the studio already uses them for other work. Do record the incremental upgrade or new seat caused by this adoption. Procurement improves when sunk cost and new cost are kept separate.
Put labor beside the subscription
A lower license price can be erased by repeated exports, uploads, repair work, and review. Time one complete route from current design file to accepted output. Include setup, prompt entry, masking, queue supervision, selection, correction, verification, naming, and storage.
Use active staff minutes separately from unattended compute time. The distinction matters. A fifteen-minute queue that needs no attention is different from fifteen minutes of masking. Multiply active time by the studio’s internal planning rate if a money total helps, but keep the raw minutes visible.
Do the same for onboarding and maintenance. A local node system can avoid recurring image charges while requiring model management and troubleshooting. A hosted service can reduce setup while adding recurring seats and data review. Neither structure is automatically cheaper.
Make uncertainty visible
Attach a retrieval date and source to every price. Product plans change, and a roundup can become stale before the software does. First-party pricing and terms pages should control the procurement sheet, even when a comparison article introduced the candidate.
Use three confidence labels. Verified means the condition appears on a current first-party page or written quote. Estimated means the studio derived it from a stated assumption. Unknown means a required detail is missing. Never let an unknown cell silently become the winning low price.
Run a sensitivity check on the uncertain variables. What happens if keeper rate falls from one in six to one in ten? What if two more seats are needed? What if high-resolution exports consume more credits? A tool that wins only under its most favorable assumption is not the cheapest option. It is the most assumption-sensitive option.
A procurement sheet that can survive review
For each candidate, retain the source URL, retrieval date, eligible plan, billing term, cash commitment, seats, included allowance, overage rule, required add-ons, test attempts, accepted images, active minutes, and unresolved questions. Put technical fit and rights gates before the cost ranking.
Chaos is right that workflow fit, control, speed, price, and learning effort all matter. Its mixed starting-price formats also demonstrate why a published table is the beginning of buying work. The studio must replace each attractive entry number with its own commercial case.
Do not ask which cell is smallest. Ask what the same twelve approved images cost.
Sources and method
This field guide responds to the Chaos comparison of six AI architectural rendering tools surfaced on 25 September 2026. The source table mixes annual-plan monthly pricing, student pricing, per-render pricing, and undisclosed pricing, and Chaos identifies five evaluation criteria. Chaos sells Veras, one of the products it evaluates. This article does not independently verify the listed competitors’ prices or report hands-on testing. Buyers should check current first-party pricing, terms, and written quotes.