3D Generated User Interface Icons

• 

  - an icon of a cymbidium orchid in light blue metallic iridescent material, 3D render isometric perspective on dark background

• 

  - an icon of a cattleya orchid in light blue metallic iridescent material, 3D render isometric perspective on dark background

• 

  - an icon of an orchid in light blue metallic iridescent material, 3D render isometric perspective on dark background

• 

  - small game piece autogenerated on magespace and Scenic.gg

Inspiration

Having continually tried many TTI (Text to Image) machine learning models, I decided that the best for this type of Icon Generation was DALL-E-2.

What it does

Takes text descriptions and generates a 3D Icon image that can be used in mobile and web applications.

How we built it

I can describe the 'evolution' of my text prompts but can take no credit for Openai Machine Learning Foundational TTI Models and their construction.

Challenges we ran into

Surprisingly, most modern TTI ML models are quite picky about the 'descriptors' and syntax that are input. This required an extensive amount of textural modification, but when it is finally 'tuned' it can produce reliably consistent 3D iconography. We were also able to produce extremely accurate 'museum quality' sculpted bronze statues that would take a human decades of formal art academy education to produce.

Accomplishments that we're proud of

Applying for and being accepted to the following beta ML TTI projects in order to test this. MidJourney (beta) Discord Served DALL-E-mini HugginFace Served RuDALLE Google Colab and HugginFace Served DALL-E-2 Openai (beta) served

What we learned

After having tried 'MidJourney', DALL-E-mini', 'RuDALLE' ML TTI models, I found that the Openai (beta) DALL-E-2 was the most advanced and successful Machine Learning model for this particular application.

What's next for 3D Generated User Interface Icons

I hope to incorporate a series of these UI Icons into mobile games and webpages in animated forms using the 'RIVE' platform. The new 3D mesh aspect should enable an easyer splineing/bone process for animating the Icons during UI activations.

Built With

• dall-e-2
• machine-learning
• midjourney
• text-to-image
• tti