IMesher

Inspiration

Turning memories into reality shouldn't require a degree in 3D modeling. We wanted to empower anyone—technical or not—to transform a photo of themselves or their pet into a high-fidelity digital character and a 3D-printable file in seconds.

What it does

IMesher automates the complex pipeline of 3D asset creation. -Segment: Instantly isolates the subject from the background. -Stylize: Reimagines the subject as a unique digital character. -Enhance: Injects 3D depth cues and shadows to create a "2.5D Proxy Blueprint." -Generate: Feeds this enhanced visual into a shape generator to produce a clean, ready-to-use 3D mesh.

How we built it

We leveraged the Gemini 2.5 Flash and Gemini 3 APIs to act as our primary engine: -Preprocessing: Gemini 2.5 Flash normalizes noisy photos into clean illustrations and performs semantic segmentation. -Geometric Analysis: Gemini 3 analyzes the object’s anatomy to understand its volume and extracts critical feature maps. -Synthesis: The model "hallucinates" lighting and depth cues, creating a structural input that far surpasses standard 2D-to-3D conversion methods.

Challenges we ran into

Navigating API rate limits and prompt variability required a rigorous "pipeline stabilization" phase. By refining our prompts, we ensured Gemini’s outputs remained consistent even with complex, diverse user inputs.

Accomplishments that we're proud of

-Higher Fidelity: Our "Proxy" workflow significantly reduces distortion and improves mesh quality. -Efficiency: By consolidating tasks into Gemini, we reduced the need for multiple heavy models, saving on GPU resources and cutting processing time. -Feature Extraction: Gemini consistently outperformed specialized APIs in recognizing and preserving intricate subject details.

What we learned

Gemini 3 isn't just a chatbot; it’s a powerful alternative to traditional computer vision models. It handled complex geometric reasoning that previously required a custom-trained stack.