A current Chaos comparison says a freelance architectural visualizer typically charges $500 to $3,000 or more per image. It then argues that even a $199 monthly Veras tier costs a fraction of outsourced visualization for a firm producing many renders. The arithmetic is obvious. The comparison is not.
A software subscription provides access, a credit allowance, and a set of controls. A commissioned image can include scene preparation, material decisions, lighting, entourage, revisions, coordination, color work, and delivery by a responsible person. Put the sticker prices side by side and all of that work disappears from one column.
Price the same unit
The Chaos six-tool guide is right that monthly subscription prices can mislead. It is also the source of the $500 to $3,000 freelance range and the $199 monthly comparison. We did not independently verify that broad freelance range, which will vary by market, scope, quality, rights, schedule, and revision terms.
The first correction is to define the unit. “One render” could mean a rough mood study, an approved exterior hero, six coordinated interior views, or a camera-matched revision after the model changes. The unit should be an accepted deliverable with written requirements. If the office needs one 4K exterior that preserves every opening, includes two revision rounds, and is approved for a planning presentation, price exactly that.
Then define acceptance before generation. Record geometry that cannot move, required material zones, output size, number of views, deadline, revision allowance, and permitted post-production. An attractive image that changes the facade is not a cheaper version of the required image. It is a different product.
Compare accepted deliverable with accepted deliverable, not software access with somebody else's finished work.
The in-house equation
Start with the portion of subscription and credit cost attributable to the project. Add operator time for setup, prompting, generation, selection, and repair. Add coordination time for extracting the right view and checking changes against the model. Add specialist time for Photoshop, compositing, typography, or color when required. Finally, add the cost of rejected attempts and schedule risk.
A compact office equation is:
In-house cost = access + consumed credits + staff hours + review hours + repair hours + rework allowance.
Use loaded hourly rates, not salary divided by working hours. The loaded rate should reflect the practice's actual cost basis or internal billing method. If that figure is sensitive, keep the worksheet internal and publish only the resulting range.
Do not divide the monthly plan by every file downloaded. Divide total monthly tool and labor cost by accepted deliverables at the required standard. Fifty experiments and four approved client images mean the relevant denominator is four.
| Cost line | Evidence | Common omission |
|---|---|---|
| Access | Seat, billing term, tax, add-ons | Annual commitment presented as a monthly fee |
| Consumption | Credits used by all attempts and operations | Counting only the selected generation |
| Operator | Setup through export, by role | Treating architect time as free |
| Review | Model check, redline, client preparation | Calling visual inspection automatic |
| Repair | Masks, compositing, geometry correction | Excluding work outside the AI interface |
| Rework | Repeat after design or model change | Pricing only the first approved camera |
Credits are operations, not images
Chaos's own support documentation shows why a seat price is not an output price. Its August 2026 operation table assigns different charges by engine, resolution, enhancement type, and video duration. Nano Banana 2 images are listed from 5 to 20 credits depending on resolution. Nano Banana Pro 4K is listed at 35. AI Enhancer operations use 10 or 20. Video charges can apply per second. Chaos notes that it may adjust credit pricing.
A single approved still may therefore consume credits for several generations, an object correction, and a full enhancement. The exact combination belongs in the job record. A video is even less suited to “subscription divided by outputs” because duration and resolution change the consumption directly.
Allowance ownership matters too. The Chaos named and shared pool guidance says named credits are reserved for an assigned user, while purchased packages and floating-license credits enter a shared pool that invited users can consume. Admins can set user limits. A busy team needs to price contention and governance, not just total credits printed on a plan.
The outsourced equation
Ask a visualizer for a scoped quote and make the inclusions visible. Record briefing time inside the office, file cleanup, vendor coordination, included revisions, rush fees, licensing, extra views, and internal review. Add the cost of changes beyond the agreed scope.
The equivalent formula is:
Outsourced cost = quoted fee + internal briefing + internal review + extras + change allowance.
Do not assume outsourcing removes staff time. It moves production labor but still requires a good model, a clear brief, decisions, and approvals. Conversely, do not treat the visualizer's margin as waste. It pays for judgment, production capacity, equipment, quality control, and responsibility for delivery.
Run one paid pilot
Take a recent project with known requirements and select one representative view. Give the in-house AI route and an external visualizer the same brief, source package, deadline, and acceptance checks. Do not ask either route for speculative unpaid work.
Track elapsed time and active staff time separately. Record every AI attempt and every requested external revision. At the deadline, classify the outcome as accepted, repairable, or rejected. If repairable, finish it and count those hours. Repeat once after a small model change to expose revision cost.
The result will not settle every project. It will give the studio its own evidence. A small office making frequent concept images may find that AI pays quickly. A team buying a few high-stakes marketing images may find that specialist production remains cheaper once senior staff hours and correction risk appear.
Keep quality bands separate
Do not average concept studies and final campaign images into one cost per render. Create at least three bands: internal option, client-review visual, and public final. Give each band its own acceptance rule. A rough study can tolerate invented planting and minor material drift. A client-review frame may need the correct massing, openings, and principal materials. A public final may require close coordination, high resolution, controlled rights, and detailed retouching.
Report the cost and acceptance rate inside each band. If AI produces twenty cheap internal options but only one acceptable public image, that is still useful capacity. It does not support a claim about final-image replacement. If an external visualizer is expensive for quick options but reliable for finals, the office can route work accordingly.
Also record who performed the AI work. A principal, junior designer, and visualization specialist carry different opportunity costs. The cheapest route on paper can be the most expensive if it consumes the one person needed to finish the drawing set.
Our take: buy capacity, commission judgment
AI access and visualizer labor are not enemies. They occupy different lines in a production plan. Use software to expand iteration where the office can operate and check it efficiently. Commission a specialist when the deliverable needs concentrated craft, predictable coordination, or production capacity the project team does not have.
There is nothing wrong with a subscription saving money. Prove the saving with the accepted files, the time sheet, and the credit log.
If architect hours are missing, the bargain is imaginary.
Editorial basis: the 15 September 2026 ArchiGen AI intel sweep, the current Chaos six-tool comparison, and Chaos support pages for operation costs and credit pools. All product and price figures are attributed to those sources and access date. This article proposes a cost method and does not claim hands-on testing or independently validate freelance market rates. ArchiGen AI carries no sponsored placements.