Hey gang—Strato here. In this video I walk through how I turned a simple Home Assistant dashboard idea into a live, interactive, isometric-style 3D floorplan using ChatGPT (with DALL·E for image generation) and a little bit of Photoshop. The whole thing started with inspiration from a Home Assistant subreddit post and the wireframe aesthetic that Madelena helped pioneer on the design side of Home Assistant. Most people tackle this with CAD or full 3D modeling tools, but I wanted to see how close I could get with just an LLM, good inputs, and some spare time. I break down my process step-by-step: how I take the right photos (angle matters a lot), how I prompt ChatGPT to generate clean isometric renders, and why I don’t rely on the model for every “state” variation. In practice, getting pixel-perfect consistency between images (lights on/off, blinds open/closed, etc.) is where Photoshop and layers save the day—because the LLM loves to “help” by moving walls and furniture, and that breaks immersion. Finally, I show how I wire it all into Home Assistant using an image helper and Jinja2 templating so a single picture card swaps to the correct image based on entity states. It’s design-forward, tidy, and still very much in the spirit of a privacy-first smart home dashboard.