Generate Variations
There’s a reason I don’t like t-shirts. It was illustrated to me by Open AI’s Dall-E2. I painted out my shirt and asked for it to create variations. It did a great job of creating new styles and I realized that a collared shirt is definitely my cup of tea.
Midjourney did a better job of making a dreamier version of myself:
It even added back some eye glasses like the ones I used to wear:
If I asked for a cool version, it came up with the following:
Now, these definitely don’t look like me. There might be a slight variation but what it does teach me is what the training material might be tagged for “handsome” and “cool”. Sure, this dataset might be biased but it taught me what might be classified by the taggers:
- A thousand yard stare
- Head tilted with chin out
- Wide open collar
- Longer flared hair
- Losing some weight
I could see the dangers of someone with an impressionable self image relying on this technology to influence their sense of beauty. Worse, filters might obscure what they’re experiencing about themselves and others around them.
It might not be the Metaverse we get lost in but unending imagery of things that appear better than reality. We may need to wait for the pendulum to swing back for this technology to balance into a useful space.
