Gausium Phantas In-Situ
Personal
A set of AI-generated in-situ product shots of a robotic floor cleaner across three real-world environments.
The Brief
After following AI image and video generation throughout 2024–2025, I wanted to test first-hand what it was capable of. Having spent five years in the B2B sector, I knew in-situ product shots were among the most sought-after assets — so that became the first test. I chose the Gausium Phantas, a popular robotic floor cleaner from an industry I knew well, and set out to generate speculative website-ready imagery. This was a personal project; the rights remain with their respective owners.
The Approach
I broke the task into three stages: first, generate a high-quality background image that reads like a professional stock photo. Second, create a custom LoRA for the Phantas product so the details, scale, and colours would be accurate. Third, combine the background with the LoRA to produce a convincing in-situ image.
The Execution
Using a small tool I built called Screen Prompt, I generated image prompts from stock photos of locations I liked — these became the basis for the backgrounds. I then used InvokeAI with FLUX1-dev for image generation, iterating 3–4 times per image to get a solid base before correcting details like text and hallucinations. For the product LoRA, I downloaded Phantas product images in various lighting conditions and angles, then used Flux GYM on Runpod to train the LoRA. Once trained, I returned to InvokeAI locally and combined the background prompt with the Phantas LoRA to place the product convincingly in each scene.
The Result
Three final images placing the Phantas in an office lobby, airport lounge, and school hallway — with a supermarket scene as a bonus. The results demonstrated that AI-generated in-situ product photography can achieve quality comparable to a traditional production shoot in a fraction of the time and cost. What would have taken a full crew, location fees, and post-production could be produced in a single session.
Stills
