This week I visited neural.love, a multi-purpose AI site with a focus on images, video and sound (i.e. not chatGPT). The suite includes image, video and sound enhancement, as well as multiple creation tools. I took a spin with its Avatar creation tool.
The fundamental principle of image generation with AI is that sample size = accuracy. In other words if you ask MidJourney for a funny image of Joe Biden or Donald Trump the results are scarily accurate. Why? Because the AI has millions of photographs to "learn" from.
So, in order to get good avatar results, you need to upload a LOT of photos. Ideally at least 20 and up to 100 with this tool. The recommendation is to include different angles and facial expressions. I uploaded about 40 and fired up the AI. With AI tools like chatGPT, we get used to millisecond-fast response times. But with this kind of intense image gen work, like MidJourney, the engine behind it, the neural.love AI needs up to an hour to "learn" from the uploaded images.
The results range from breathtaking to downright weird! With underlying prompts ranging from anime to military, oil-painting to superhero, your 50 avatar images provide quite the variety. Here are just a few:
The hilarity that ensued sharing these with my family, inevitably resulted in requests for their own set of Avatars. Using roughly 20,40 and 60 images each, we all agreed that the best results (accuracy of features) definitely came from more images uploaded.
Here are a few of the kids' images:
This test was by far the most entertaining, and at $10 for 50 images is worth the "investment".
As it relates to the wider world of Marketing, based on the qualitative study of one family, this seems like a natural for consumer engagement and entertainment while the novelty lasts.