Why you should be running internal AI tests

Public broadcasters now ask every producer one extra question: can you prove your AI tools didn’t infringe on someone else’s IP?

That alone is making a lot of indie studios pause before starting to test any workflows with scraped-data models. And fair enough - nobody wants to get into problems with their clients.

Clean models are here and more are coming

Two recent datapoints worth tracking:

👉 F-lite — launched this spring and trained only on Freepik’s 80 million non-AI, non-editorial images. It is also open-source, so you run it locally for only the cost of compute, no need to upload your content into the cloud.

👉 Marey (Moonvalley) — announced as a “clean” video model and expected later this year.

Neither solves everything, but both signal where we’re headed: models that buyers can sign off on without a legal footnote.

How I'm running low-stakes tests myself

In the meantime I’ve been running small experiments to get a feel for the workflows.

Example: capture a start- and end-frame of a plastic toy, then let Kling or Pika produce the in-betweens.
​

Nothing from their training images seeped visibly into the animation, yet the model itself is still out of bounds for a project headed to a broadcaster, because its training set fails the aforementioned due-diligence test.

Why mention this?

💡 The stop-motion-style trick above is exactly the kind of use-case a Marey-class model is unlikely to trip over once it’s available.
​
💡 The prompts, camera planning and timing decisions I practise today will transfer straight to those compliant tools tomorrow.

Your turn - want to share any experiences?

So yes - keep your R&D sprints private for now, but keep them going. When the “safe” models are out, you’ll already know how to best make use of them.

So what does your low-risk test bed look like? What have you been told by broadcasters in regards to AI tools in production? I'd love to hear your experiences, just hit reply and share!

Have a great day!

- Nick

Nick Dorra

Say hi 👋 on Linkedin​
🤝 Book a meeting via video or f2f to chat more