Generative AI · Model training case study
AI Trainer Image Generation
A model-training experiment that used a small set of trainer photos to generate new product images, colourways and lifestyle scenes.
- Role
- Research & build
- Organisation
- Independent
- Context
- 2023
The work
This project explored whether a model could learn the visual identity of a specific pair of trainers from a small set of source photos, then generate convincing new images from prompts alone. I photographed the original black trainers on a sheet and used that dataset to produce new variations: different colours, different environments, and even images of someone wearing them.
Delivery detail
- Built a small training set from original product photos captured in a simple home setup rather than a studio pipeline.
- Used the trained model to generate new colourways and product variations while preserving the recognisable structure of the trainers.
- Expanded the outputs beyond packshot-style images into lifestyle scenes, including outdoor settings and on-foot imagery.
- Tested how far prompt-driven generation could push consistent product identity from a limited number of reference images.
- Training setup
- Few-shot
- Colour variants
- New
- Scene generation
- Lifestyle
Inside the work
