To begin using the platform:
- Access the Flow Studio website from your browser.
- Sign in or create an account.
- Once inside, the main dashboard will display the available models.
From this dashboard, you can access tools such as:
- Text to Image
- Image to 3D
- Text to 3D
In this workflow, we will primarily use Text to 3D and Image to 3D to generate three-dimensional assets that will later be used to populate a scene within a traditional ArchViz pipeline using 3ds Max and Arnold.
Asset generation interface in Flow Studio
Once inside Flow Studio, the model generation environment is organized into several sections that facilitate the creation and iteration of 3D assets.
Generator type
In the upper-left corner are the different generator types available, including:
- Text to Image
- Image to 3D
- Text to 3D
In this workflow, Text to 3D and Image to 3D were used primarily to generate the objects that would later become part of the scene.
Prompt input dialog
Below the generator selector is the prompt field, where you describe the object you want to generate. Here, you can specify characteristics such as:
- Object type
- Style
- Approximate materials
- Intended use within a scene
This field provides control over the type of asset generated and makes it easier to explore different variations.
Model settings
It is also possible to define additional parameters before generating the model:
- Model name
- Generation seed for reproducing results
- AI model used
These parameters enable more controlled iterations during the generation process.
Generated model iterations
In the central area of the interface, generated variations of the asset are displayed, allowing you to quickly compare different options before deciding which one to export or use within the pipeline.
This iterative process is especially useful for rapidly generating multiple variations of the same object.
Generated asset library
On the right side is the generated model library, where all assets created during the session are stored.
From here, you can:
- Review previous models
- Reuse previously generated assets
- Organize created elements
This makes it possible to quickly build a small library of props that can later be integrated into more complex scenes.
Credits and generation cost
At the top of the interface, the number of available credits is displayed. The Generate button also shows:
- Credit cost
- Estimated generation time
This makes it easy to evaluate the computational cost of each iteration.
Model generation
Once the prompt and model parameters have been defined, the process is initiated using the Generate Mesh button, which produces asset variations within a few minutes.
These models can then be exported and used within traditional 3D production pipelines, such as the workflow explored in this article using 3ds Max and Arnold.
Initial asset generation with Flow Studio
The process began by using Flow Studio to quickly generate three-dimensional assets that would later be used to furnish and decorate an interior scene.
Rather than generating a complete AI-rendered image, the goal was to create props and environmental elements that could be integrated into an architectural visualization environment.
To accomplish this, two tools within the platform were primarily used:
- Text to 3D
- Image to 3D
These tools allow users to generate 3D objects from text descriptions or reference images, making them particularly useful for rapidly creating elements such as:
- Furniture
- Decorative objects
- Environmental props
During this stage, several asset variations were generated that could later be incorporated into the final scene.
Initial Geometry Generation with Wonder3D
Once some assets had been generated in Flow Studio, the next step was to use Wonder3D to obtain a more structured three-dimensional approximation of certain objects.
Wonder3D generates 3D models from images or descriptive prompts, producing an initial geometry that can later be refined within modeling software.
This makes it possible to obtain:
- Base geometry
- Volume approximations
- A three-dimensional reference that can later be refined
At this stage, some adjustments are still required within modeling software, but the process significantly accelerates the initial creation of assets, especially for secondary elements within a scene.
How to structure prompts for asset generation
One of the most important aspects of working with AI-generated models is prompt structure. A well-defined prompt helps the model better understand the type of object you want to create.
In general, it is recommended to structure prompts using four main components:
- Object type
Clearly describe the object you want to generate.
Example:
modern lounge chair
- Geometric characteristics
Specify aspects related to the object's form or structure.
Example:
curved armrests, wooden legs, minimalist proportions
- Materials or textures
Although textures can be adjusted later, including this information helps the model generate more coherent surfaces.
Example:
fabric upholstery, light oak wood legs
- Style or context
This component helps guide the design toward a specific visual style.
Example:
scandinavian design, minimalist interior furniture
Example of a structured prompt
A complete prompt might look like this:
modern scandinavian lounge chair, curved armrests, minimalist proportions, wooden legs, fabric upholstery, light oak wood, interior furniture asset, clean geometry
This type of prompt provides enough information for the model to understand:
- What object to generate
- What form it should have
- What approximate materials to use
- What visual style to follow
The result is typically a coherent initial geometry that can later be integrated into a 3D production pipeline.
Scene construction in 3ds Max
Refining and exporting assets in Flow Studio
Once an asset has been generated in Flow Studio, the platform provides several additional tools that allow the model to be refined before exporting it to a production pipeline.
These tools appear at the bottom of the interface and allow quick adjustments to the generated model.
Refinement tools
Flow Studio includes several options that help prepare an asset for export.
Retexture
The Retexture option allows new texture variations to be generated using AI.
This is useful when:
- Exploring alternative materials
- The initially generated texture is unsuitable
- The asset needs to be adapted to a different visual style
For this workflow, most automatically generated textures were retained in order to evaluate how usable they are within a traditional ArchViz pipeline.
Remesh
The Remesh tool regenerates the model's mesh to improve polygon distribution.
This can help:
- Optimize geometry
- Reduce mesh artifacts
- Improve topology before export
In some cases, this makes integration into modeling software such as 3ds Max easier.
Rig
The Rig option generates a basic rigging structure for the model.
This tool is primarily intended for animation pipelines and was not required for this workflow. However, it may be useful when generated assets are intended for animation or character-production projects.
Exporting the model
Once the asset has been generated and refined, Flow Studio allows it to be exported directly from the interface.
The system supports downloading models in several common formats used in 3D pipelines.
Available export formats
USD (Universal Scene Description)
USD was originally developed by Pixar and is designed to handle complex 3D scenes and collaborative production pipelines.
Advantages include:
- Efficient management of complex scenes
- Support for hierarchies and scene composition
- Compatibility with modern production pipelines
It is widely used in film production, animation, and advanced VFX workflows.
OBJ (Wavefront Object)
OBJ is one of the most widely supported formats in the 3D software ecosystem.
Key features include:
- Broad compatibility with 3D software
- Support for geometry, UVs, and basic materials
- Easy import into tools such as 3ds Max, Maya, Blender, and Cinema 4D
For this workflow, OBJ was used primarily because it allows generated assets to be quickly integrated into 3ds Max for continued scene construction and rendering.
STL (Stereolithography)
STL is primarily intended for 3D printing.
This format contains only triangulated geometry information and does not include:
- Materials
- Textures
- UVs
For this reason, while useful for digital fabrication and prototyping, it is generally not the ideal format for visualization or rendering pipelines.
Once exported, the model can be integrated into the 3ds Max production pipeline, where final adjustments to scale, scene organization, architectural materials, and lighting are completed before final rendering with Arnold.
Geometry optimization in 3ds Max
When importing assets into 3ds Max, it is highly recommended to perform an initial geometry optimization process.
AI-generated models often contain very high polygon densities because generation algorithms prioritize visual fidelity. While this can be beneficial for capturing detail, it can also result in unnecessarily heavy meshes within the scene.
To better manage these models, a useful tool is the ProOptimizer modifier in 3ds Max.
This modifier reduces the number of polygons while maintaining the object's overall shape, making it easier to integrate assets into more complex scenes without significantly affecting visual quality.
Benefits of using ProOptimizer include:
- Significant reduction in polygon count
- Preservation of the model's overall shape
- Retention of UV coordinates
- Improved performance in scenes containing multiple assets
In many cases, geometry can be reduced by 30% to 70% without any noticeable loss in final render quality.
This is especially useful when working with multiple AI-generated assets, helping maintain a lighter and more efficient production scene.
Materials and textures
One interesting aspect of this experiment was that most assets generated with Flow Studio retained their original textures.
Rather than completely rebuilding the materials, the decision was made to keep the automatically generated textures in order to evaluate how usable they are within a traditional visualization pipeline.
The only textures and models created from scratch were those associated with the architectural shell of the space, including:
- Walls
- Floors
- Ceilings
- Windows
- Moldings
For these elements, PBR textures were used, primarily sourced from Poly Haven (completely free and an excellent online resource), and applied within Arnold's material editor.
During this stage, basic adjustments were made, including:
- Texture scale
- Normal maps
- Roughness values
- Surface micro-detail
This helped maintain visual consistency between the architectural elements and the AI-generated assets.
Lighting and final rendering with Arnold
Finally, the scene was lit using Arnold lights, primarily simulating natural light entering through the window.
Several lighting tests were conducted to achieve an appropriate balance between:
- Physical realism
- Clear readability of the space
- Visual contrast
The lighting was combined with exposure adjustments in the render settings to reinforce a sense of depth and naturalism within the space.
The final result was a 4K interior render generated entirely in 3ds Max and Arnold, using the previously generated assets as environmental elements within the scene.
Conclusion
Several interesting observations emerged during this experiment of integrating AI-generated assets into a traditional architectural visualization (ArchViz) pipeline:
- AI-generated assets can serve effectively as a starting point for scene dressing and environmental setup.
- In many cases, automatically generated textures are usable without major modifications. However, they should still be reviewed carefully and adjusted when necessary, particularly for final-production projects.
- Most of the work within the pipeline still occurs in traditional stages such as scene organization, lighting, architectural materials, and composition.
- AI is particularly useful for accelerating the creation of secondary props and environmental elements, reducing the time required to build a scene from scratch.
- In this sense, AI tools do not replace the traditional ArchViz pipeline; rather, they act as accelerators during the early stages of the creative process.
I would be interested to hear how others are integrating AI tools into their own ArchViz or visualization pipelines. Have you tried something similar within your production workflows? Please share your experience in the comments.
2 Comments
