Exciting, but alarming in terms of copyright.
It's important to know how Midjourney handles data. MJ's terms of use state, among other things, the following:
The midjourney /describe feature, or generally uploaded images for use in MJ, are brilliant from a copyright perspective for MJ. They allow MJ to collect a dataset of royalty-free images for ethical model training before making a legal decision about their current models (which are precariously based on "fair use").
Midjourney can train its models with your images without you ever knowing!
The most frightening thing: anyone can upload any photo, regardless of a copyright. And MJ says thank you without the copyright holder ever knowing.
Nevertheless, I tested the feature because I was particularly interested to find out if it works better than GPT-4. And yes, it does.
With the command /describe you can now upload an image and get four text prompts that try to describe the image. Afterwards you can regenerate it with MJ.
📸 However, in my test with an unprocessed cell phone photo of Zermatt, I found that the results were unusable. You can see the best result (with V4) here, while the other three prompts were totally useless. At least the feature works in contrast to GPT-4.
🤖 Not surprisingly, on the other hand, the analysis and regeneration of AI images works quite well.
Here on an example I created in PicsArt for this test. I generated the flower with AI:
And here are the results (with V4) of the four MJ prompts proposed above based on the image analysis:
AI is exciting and I like to experiment with it. But at the same time, I have big concerns. AI compositions are based on a lot of training data with the goal of meeting the tastes of most people. They simply can't beat the value of a hand-created composition based on personal vision and inspiration.
Sometimes it's the unconventional creations that are the most captivating.
Try creating a minimalist creation in a more abstract composition using AI. I'm almost sure you'll fail, as it won't suit the tastes of the majority, which is what the AI has been trained to do.
I think the value of handmade artwork (including its tactile qualities) and craftsmanship in general will increase.
However, what I find quite unfortunate is that now with AI, even a person with no talent (I call it talentless) and no passion can use AI to make a fairly appealing creation, as important aspects such as color harmonies and appealing image composition are taken care of by the AI.
What are your thoughts on this?