The chocolate frog card that comes to life
In Harry Potter, the famous wizards on the chocolate frog cards are alive: they step out of the frame, look at you, then walk away.
As a child I thought this was one of the best ideas in the film. Why do we expect our avatar to just stand there all day? When I was looking for an intro picture for our site, the same thought came to me: why shouldn’t the profile picture move? Why have a plain profile picture like every other website, when we can do more than that?
The profile video you can also see on the About page
The video was made by Grok from a single photo of mine and a short description. The picture comes to life: I greet the visitor, look at you, then walk out, and the empty room remains. The same video plays on our website. It plays once; the green dot next to the name plate shows that I’m “in”, and it turns grey once I’ve walked out.
How was it made?
- The ideaA moving portrait, like the chocolate frog card. You need an idea that fits in one sentence, otherwise the AI guesses. A good prompt is essential, because we don’t want extra things to end up in the video, or something to happen that we didn’t intend. Saving tokens is getting more and more important, and will be even more important in the future.
- Starting imageOne good-quality, well-lit photo. But it’s not a big problem if you have 4–5 photos from which the AI can work out roughly who will be in the picture. We can add detail with text.
- Description (prompt)What should happen, in what order, in what mood. A short, specific description works best.
- Picking the bestWe generate several versions and choose the best. The first attempt is rarely the best.
One place for many video models
Higgsfield is a platform where you can reach several AI image and video models from one place. We use it when we want to animate an image, add camera movement to a scene, or try the same idea with several models.
There is no need to jump between tools, and the results are easy to compare.
3D scanning and eye tracking
Video generation is only one direction. We are also experimenting with building a 3D model from the camera image, and having the computer track your eyes and head movement. The screen then works like a window: as you move, the 3D object is seen from a different side.
Where we are now
Two recordings from our 3D scanning and motion tracking experiments.
What to watch out for if you generate video too
- Rules: only use a face and voice you have the right to use. Your own is fine; someone else’s only with their consent.
- Labelling: if it matters to viewers, say that the video was made with AI.
- Short videos: clips of a few seconds are the most stable; in long scenes details drift more often. This depends on the model used, and some models can already generate quite long videos. I can safely say it comes down to how much money we spend on a given video.
- Post-production: for the best result you usually need a bit of editing and sound.
Frequently asked questions
How much does an AI video cost?
It depends on many things: from the data and sample videos available, to how long a video we want to create. How refined and how high-quality we want the video to be is also a factor that can’t be left out.
How long does it take to make?
A simpler video takes 30 minutes without post-production.
Do you need expertise?
I don’t think any special expertise is needed. It’s worth knowing the different models and what is worth combining them with.
How many attempts does it take to get a good video?
You should always plan for 3–4 attempts. Often we don’t regenerate because the result is bad, but because we start to see what else is possible, and we want more and more.
What went wrong, what did the AI do unexpectedly?
I’d split this in two. The first is my intro video: in the prompt I asked for a Harry Potter example and got a magical-world background with sparkles. It didn’t look bad, but it wasn’t what I expected. I designed the prompt wrongly. The other is scanning, where a prompt isn’t necessarily enough and preparation is needed. I mean the visual assets you want to show: it doesn’t hurt to first generate the object you want to display with a 3D modelling AI.
A moving profile video from one photo
From one photo and a few sentences you can now make a moving profile video and build it into your website. Grok and Higgsfield are a good starting point, and 3D scanning and eye tracking show where this can go next.
We teach it
Within our training we teach people of any age the basics of video generation, in the area where they want to use the knowledge.
ChatGPT, Claude or Grok? My own experience →
Get in touch and we’ll show you what can be made from your photo or product.