Are AI generated 3D models any good?

Introduction

Exploring AI in 3D Model Generation: An Evaluation

In recent weeks, I've been diving into the world of AI, particularly focusing on the latest advancements in 3D model generation. While generative AI isn't new, with tools like ChatGPT being widely used, the progress in 3D model generation is relatively new and intriguing. This article explores various AI platforms for 3D model generation and evaluates their performance.

Testing Meshy

One platform that caught my attention was Meshy. Here's an example of a test I conducted with Meshy. I used the prompt: "A low poly rabbit sitting on a small barrel holding up a blank sign. The rabbit is smiling." Despite a typo in my prompt, I was curious to see the results.

Results from Meshy

Result 1: A rabbit sitting on a small barrel, but no sign.
Result 2: Slight improvement, but still no sign.
Result 3: A rabbit on a barrel with a blank sign, but not holding it up.

The resolution wasn't the best, so I used the "refine" feature to enhance it. Despite the advancements in image generation, 3D modelling has a lot of catching up to do.

Experimenting with Tripio

Next, I tested Tripio with a new prompt: "A pig riding a motorcycle." The options included realistic, cartoon, sculpture, and low poly art styles. Here are the initial results:

Result: Varied results, with some looking more like a "motorcycle pig" than a "pig riding a motorcycle."
Refined Result: Improved realism, with some models even looking like they could be 3D printed despite minor flaws like missing handles or weird shapes.

Hugging Face and Instant Mesh

Hugging Face's Instant Mesh offers a different approach by converting 2D images into 3D models. For example, I uploaded a motorcycle image, and within seconds, it generated a 3D model by removing the background and imagining the unseen parts.

Analysis

The generated 3D model was impressive given the quick processing time, but it wasn't as high-resolution as other methods. This approach, however, shows promise and could become even better with continued development.

Photogrammetry and AI

Lastly, I explored using photogrammetry combined with AI for enhanced 3D asset creation. Websites like Maker World and services provided by Bambu Labs use photogrammetry to scan real-world objects and then refine the results with AI. This method requires access to real-world data and images but offers high-quality results when combined with AI.

Conclusion

The advancements in AI-generated 3D models are promising but still have room for improvement. Tools like Meshy, Tripio, and services from Hugging Face and Maker World present exciting possibilities for the future.

Keywords

AI
3D Model Generation
Meshy
Tripio
Hugging Face
Instant Mesh
Photogrammetry
Bambu Labs

FAQ

Q: What is the most promising development in AI-generated 3D models? A: The refinement capabilities and speed improvements in platforms like Meshy and Tripio, as well as the new image-to-3D model techniques from Hugging Face, show significant promise.

Q: Are AI-generated 3D models ready for professional use? A: While AI-generated models are improving, they may still require some manual refinement for professional use. However, the technology is rapidly advancing.

Q: How does photogrammetry benefit from AI? A: AI methods can enhance photogrammetry results by speeding up the process and adding details that may not be captured in the original images, thus creating more realistic 3D models.

Q: What are some challenges with AI 3D model generation? A: Current challenges include resolution quality, accurate representation of specified details, and sometimes incomplete or unrealistic parts in the models.