Complete guide: How to use AI in your archviz pipeline

AI in architectural visualization transforms the traditional 3D pipeline by automating labor-intensive tasks such as asset population, material generation, and final rendering. In this guide, Chaos CTO, Vladimir Koylazov, shares easy and practical use cases on how professionals can use AI across the full pipeline.

Key takeaways:

• Drastic efficiency gains: AI rendering enables the generation of multiple design variations in minutes, bypassing the hours of manual lighting and material setup required in traditional workflows.
• Enhanced realism with controlled editing: Tools like the AI Enhancer refine difficult elements like people and vegetation while preserving the original architectural geometry.
• Model-free conceptualization: Architects can leverage AI to generate high-quality visuals from simple hand-drawn sketches and LLM-generated prompts before any CAD work begins

Table of Contents:

• Accelerating conceptual design with Chaos Veras
• Advanced AI archviz workflows: LLMs, control, consistency & alternate views
• Populating spaces: Interiors and vegetation
• Refining visuals with AI enhancers and upscaling
• Future outlook: From 2D AI renders back to 3D scenes
• FAQs

Traditionally, converting an architectural sketch into a photorealistic render has been no small feat.

The old way requires you to build a complex model, spend hours adding entourage like vegetation and people, apply materials, and configure lighting, only to wait for the render to finish before starting post-production. And that is before the inevitable cycle of revisions begins!

The rise of generative AI is changing this process, bringing massive efficiency gains to architects, designers, and archviz artists.

In this guide, Vladimir Koylazov, CTO at Chaos, walks you through several pragmatic ways to implement AI at every stage of your pipeline, helping you go from a rough sketch to a final image in a matter of minutes.

Accelerating conceptual design with Veras

Chaos Veras serves as the primary bridge between BIM software and generative AI. It integrates directly into the Chaos tab in the Revit ribbon and Forma, as well as other modeling tools like SketchUp, and is also available as a standalone web app. This guide will focus on the Revit workflow.

From 3D model to photorealistic iteration

To generate a render, you can open your project in Revit, adjust your view, and launch Veras.

Imagine a scene with very basic shapes, no materials assigned, and no lighting.

By entering a descriptive prompt such as "Traditional English houses, red bricks, autumn" the AI interprets the geometry of your model and applies lighting, materials, and atmosphere instantly. This allows for rapid exploration of different architectural styles or seasonal contexts without manual re-texturing.

Turning sketches into reality

One of the most powerful features of AI is the ability to generate visuals from a simple hand-drawn sketch.

By describing the sketch's intent (e.g., "Two-story building, gym on the bottom, offices on the second floor, stairs on the left"), the AI generates a plausible, high-quality image that demonstrates the potential of the space before any CAD work begins.

Additionally, you can adjust the previously generated render and explore different designs, views, seasons, and more by prompting Veras with the desired changes.

Advanced AI archviz workflows: LLMs, control, consistency & alternate views

If you want to go one step beyond the basic use case and make the experience more controllable and consistent, there is a more advanced approach. You can actually integrate Large Language Models (LLMs), such as ChatGPT or Gemini, directly into your creative workflow to serve as a bridge between your idea and the final render.

What does this advanced process look like?

• Prompt engineering via LLM: Instead of writing a simple sentence, you ask the LLM to describe exactly what the viewport image should look like in great detail based on your architectural requirements.

• Mood board integration: You can provide a reference mood board image to the AI. This doesn't mean the AI will copy it directly; instead, it serves as a visual guide for lighting, color palette, and atmosphere.

• The execution: When you combine your CAD layout, the detailed LLM prompt, and the mood board in Veras, the results are much more predictable. The AI respects the basic structural layout of your model while adopting the specific style and objects from your reference imagery.

Generating alternative views and macro shots

One of the most powerful aspects of this workflow is achieving consistency across different shots. Even though you are working with generative AI, using a detailed, structured prompt will allow you to: 

• Maintain consistency: Generate multiple views of the same project that feel like they belong to the same building.
• Macro views: You can take the same base setup but adjust the prompt to specify a macro view. This allows the AI to focus on high-detail portrait shots of specific corners or furniture pieces, emulating the way a professional photographer would walk through a physical space.

Achieve granular control with JSON

For the ultimate level of control, you can use JSON-formatted prompts generated by an LLM.

Because JSON is highly structured, you can go in and edit specific lines.

For example, if you have a full room render but want to change just one element, you can identify the table section of the code, change it to a "marble coffee table," and use Veras to update only that portion of the image.

This gives you surgical precision over the final visual without re-rendering the entire scene from scratch.

Populating spaces: Interiors and vegetation

AI is exceptionally efficient at populating empty architectural environments, which traditionally, is a time-consuming asset management task.

• Instant interiors: You can take a screen grab of an empty office from Revit and use Veras to populate it with furniture automatically. You no longer need to spend hours sourcing individual 3D assets.
• Geographic specificity: You can test vegetation specific to different geographies simply by changing the text prompt. This allows you to generate versions of your image featuring plants native to Cyprus or the Dominican Republic in seconds.

Refining visuals with AI enhancers and upscaling

AI is not only for conceptualization; it is also a powerful post-production tool for final renders.

• AI Enhancer: This algorithm automatically detects elements that are traditionally difficult to render realistically, such as people and vegetation. The Chaos AI Enhancer can refine these assets while keeping the actual architectural design exactly the same.
• Targeted adjustments: You can use AI to modify specific demographics in a render, such as selecting a person's face to make them older or younger to better fit a project's target audience.
• AI Upscaler: To save massive amounts of render time, you can cast your original render at a lower resolution in Enscape or V-Ray and use the AI Upscaler to increase the resolution by up to 4x while adding sharp detail.

💡 Read more: The complete guide to AI in Chaos products

Future outlook: From 2D AI renders back to 3D scenes

The next frontier for Chaos involves moving beyond 2D image generation and returning to the 3D environment. Since we are already working with 3D scenes, the goal is to use AI to manipulate full 3D environments directly, enabling:

• Real-time navigation and VR: Interactively exploring AI-generated environments.
• Simulation and analysis: Using AI to assist with thermal comfort or daylight analysis through modules like "Impact".
• 3D generation: Generating 3D elements like lighting fixtures, air ducts, or complex landscaping based on the time of year.

FAQs

How does AI rendering differ from traditional rendering?

Traditional rendering requires manual setup of every material, light source, and 3D asset, often taking days. AI rendering uses text prompts and existing geometry to generate photorealistic results in minutes, significantly accelerating the ideation phase.

Can I use AI if I don’t have a 3D model yet?

Yes. Tools like Veras allow you to upload a simple sketch and provide a text description to generate a photorealistic version of your idea. This is ideal for early-stage conceptualization.

Will AI change my architectural design?

When using the AI Enhancer on a final render, the algorithm is designed to improve the realism of entourage elements like people and plants while keeping the actual architectural geometry exactly the same.

How does AI help with render times?

AI upscaling allows you to render your scene at a lower resolution, which is much faster. You then use the AI to upscale the image by 4x, adding detail in post-production and saving hours of render time.