On 26 June 2026, I led a workshop at the Render to Reason immersive art festival in Łódź. We met at the Łódzka Specjalna Strefa Ekonomiczna, in the former Fabryka Ludwika Grohmana, as part of the Laboratorium Praktyk Twórczych. For an hour, I talked about the consistency of visual language, meaning how to maintain the same character, style, and narrative continuity in production with AI tools.
At the beginning, I showed a study by Uniwersytet SWPS, for which I generated images. Most of the answers from participants evaluating whether a graphic was AI or a photo were incorrect. Then we moved on to practice:
- A good prompt starts with the character and space, and ends with technical parameters.
- With a weak prompt, you need to determine what is missing: shot type, technique, style, lighting, equipment, or emotion. It is worth gathering this knowledge in a skill that will rebuild the prompt for us.
- It is worth adding a knowledge base to the skill with up-to-date information on specific models, e.g. from their creators' materials and from GitHub.
- LoRA and moodboards help maintain consistency, as do 3D and Gaussian Splatting.
- AI combines well with traditional recordings, e.g. with a green screen.
You can find the festival programme on the Render to Reason website.
