LogiBlend AI

LogiBlend AI is a headless generative engine powered by Google Gemini 3.0. It accepts text, voice, or sketches, and generates professional 3D assets in real-time—without you ever opening the software.

Comment

Inspiration

The inspiration for LogiBlend AI came directly from the challenge brief: "Alleviating Cognitive Overload."

Blender is the industry standard for 3D creation, but it is notoriously intimidating. For a generalist creator, the creative flow is constantly broken by the need to memorize over 400 keyboard shortcuts, navigate dense sub-menus, and manage complex node graphs. This friction turns digital creation into a technical chore.

We asked ourselves: "What if we could remove the monitor from the equation?"

Inspired by Logitech’s mission to make the digital world more tangible, we wanted to transform the MX Creative Console into a physical architect’s tool—allowing users to build, iterate, and feel their 3D designs without ever opening the complex Blender UI.

What it does

LogiBlend AI is a "Headless" generative design engine. It creates a seamless bridge between the MX Creative Console and a background instance of Blender.

  1. Multimodal Generation: Users can input ideas via Text, Audio (Voice), or Images (Sketches). The system analyzes the input using Google Gemini 3.0 and generates the corresponding 3D geometry in real-time.
  2. Headless Architecture: The plugin runs Blender in the background as a server. The user doesn't need to interact with the software's UI to model, texture, or export assets.
  3. Tactile Control: The logic is exposed entirely through the Logitech Actions SDK. This allows the user to trigger complex generations via the OLED keys and perform precise adjustments (like rotation, scaling, or iteration scrubbing) using the physical Dial.

How we built it

We architected a solution that decouples the Interface (Logitech) from the Logic (Blender):

  1. The Backend (Python & Flask): We built a local web server that imports Blender as a Python module (import bpy). This server handles the heavy lifting—receiving prompts, managing the 3D scene, and executing geometry scripts.
  2. The Brain (Gemini 3.0 Pro): We integrated Google’s latest multimodal model to interpret user inputs. We engineered a system prompt that forces the AI to output valid, executable Blender Python API code based on visual or audio context.
  3. The Bridge (Logitech Actions SDK for Node.js): We utilized the Alpha Node.js SDK to create a universal plugin. This plugin acts as the remote control, sending asynchronous fetch requests from the Logitech software to our local Python server.
  4. The Interface: We configured the manifest.json to support the MX Creative Console (Keypad and Dialpad) and the Actions Ring, ensuring a seamless connection to the hardware ecosystem.

Challenges we ran into

  • The Alpha Frontier: Using the Node.js SDK (Alpha) presented initial configuration hurdles, specifically regarding the "Remote plugin failed to start" errors and Windows permission issues with symbolic links. We had to deep-dive into the LoupedeckSettings.ini to enable Developer Mode and debug the WebSocket handshake manually.
  • Headless Stability: Running Blender as a library inside a Flask thread required careful management of the Python Global Interpreter Lock (GIL) to prevent the server from hanging during complex renders.
  • Cross-Platform Architecture: We had to design the Node.js plugin to be "dumb" (just a messenger) so that the complex logic stays in Python, ensuring that the plugin remains lightweight and portable between Windows and Mac in the future.

Accomplishments that we're proud of

  • A Truly Headless Engine: We successfully generated 3D models from a simple API call without the Blender GUI ever appearing on screen.
  • Multimodal Success: We implemented a workflow where a user can upload a napkin sketch and get a 3D house model back, purely through code.
  • SDK Mastery: We successfully navigated the new Node.js Actions SDK, getting our plugin to link, load, and communicate with our backend environment (the "Green Dot" status) ahead of receiving the physical hardware.

What we learned

  • The Power of "Universal Plugins": We learned that by using the HasNoApplication capability in the manifest, we can create tools that empower users regardless of what software is currently in focus on their desktop.
  • Tangible Computing: We gained a deeper appreciation for how hardware inputs (dials vs. keys) dictate software architecture. A dial requires a continuous stream of events, whereas a key is a discrete trigger—this shaped how we built our API endpoints.

What's next for LogiBlend AI

We have the Brain and the Bridge; now we need the Body.

  1. Hardware Integration (March): Once we receive the MX Creative Console, we will map our API endpoints to the physical dial for "blind" iteration (scrubbing through AI variations physically).
  2. Haptic Feedback: We plan to implement the SDK's haptic features to give users a physical "thud" or "vibration" when a render is complete or when an object snaps to the grid.
  3. The Grand Finale: We aim to bring this fully functional, tactile "No-UI" workflow to Switzerland, demonstrating that the future of 3D design isn't about more menus—it's about better hardware interaction.

Currently, the Python engine runs locally to ensure zero-latency haptic feedback and data privacy. In the future, we plan to offer a 'Cloud Rendering' option for users on lower-end devices, where the heavy processing is offloaded to a remote server while maintaining the local control interface.

Built With

Share this project:

Updates