AI-integrated production pipelines

To use AI in a professional workflow you still need extensive knowledge, both of film making and of generative AI. Without judgement at the steering wheel, the product will look generic and random.

I like to think of how artists reacted when the camera was invented. Faced with a threat to their livelihood, they responded with a creative explosion, inventing movements like impressionism, surrealism, and modernism. They created what the camera could not.
Generative AI is now stripping away technical limitations in much the same way, forcing us to go beyond the bland, generic slop that is everywhere. Although it raises questions about my own creativity, I do think the friction is essential.

AI-Showreel

This is a sample of what I am working with at the moment. Experimenting, failing and learning something new for every prompt.

In order to create a working AI-workflow, that actually saves time, the focus should be on accomplishing a prompting strategy that is replicable and scalable and to force your model into doing what it’s been trained not to do.
But first you need to know what’s desirable and not. And also when not to use AI at all.

Misstakes will be done …
… make sure it’s not yours
  • Replicability – can someone else get the same results, using your workflow?
  • Scalability – can the workflow be adapted to other situations?
  • Consistency – can you recognize and avoid inconsistent output?
  • Originality – can you de-tune your model to avoid genericisms?
  • Taste and experience – can you really tell the difference between slop and art?
  • Rapid iteration: Brainstorming and testing ideas and concepts brings clarity and direction into the project which will pay off in the end, both artistically and economically.
  • Good hallucinations: Gen AI is a frustrating partner to work with a lot of times, but that just give me a reason to formulate my ideas in a much clearer way. I need to know what the hell I’m even trying to create. So does my team and my clients.

An efficient workflow will contain a mix of legacy technique, fundamental understanding and AI-powered tools. Optimizing the workflow will continue to be a moving target but with a modular approach to tools and methods you can implement generative AI where the gains are high and risks are low. Storyboarding, backgrounds, color concepts and other asset generation. Then a new tool comes along and you can replace that procedure as well. I’m all in on change, I want to be part of the new movements and do what AI can not.