Following on from my recent FHIR (Fast Healthcare Interoperability Resources) certification, I’ve created a short GenAI based video which gives a quick intro regarding FHIR’s role in digitalhealth.
This is a handy, non-technical way to get an understanding of what it is and why it matters.
This was also my first time using GenAI to create videos, and it’s been a fascinating experience:
– The first thing that became apparent was that getting the AI prompt right as part of the “Text-To-Video” process is not as easy as one might think. Adding the right description to sound as well as visuals is very important, and a term that may seem subtle in the prompt can send the generated video in a different direction.
– Given the stochastic (random) nature of GenAI, I found that a text prompt could generate a pretty different video if I submitted that same prompt again. And while the majority of the time the generated audio was fine, occasionally it was gibberish.
– To get content that I was happy with, I needed about 100 attempts to generate the video content that I thought worked. With Text-To-Video, from the time you submit a prompt, to the time the video is ready, can take time. The type of AI model can influence this, and anything between 5 and 10 minutes is not unusual.
– Adding text to the video via Text-To-Video is a no-no. The models just can’t handle that at present, so more often than not the text displayed on the screen is gibberish. If you look closely at this video, you’ll see some of that in action in the small text.
– Given how lifelike some of the content and audio is, I made a point of adding a label in the bottom right stating that the content is GenAI generated – just so there are no doubts!
– All in all it was a very enjoyable AI experimentation, and the final version of the video is something that people who are new to FHIR can use as a basis for understanding that concept. I hope to create more content in the near future, so do stay tuned!