AI products are fraught with basic usability errors, violating decades-old UX findings. Simple fixes will save AI users much pain, but AI companies should also invest in fundamental user research and integrate UX with development to address new issues like managing ideation abundance.
Nah, the interface works brilliantly. Midjourney has a sh*t interface, yet it's the dominant diffusion model. Why is that, you might ask? They are piggybacking on an existing platform, a massive platform such as Discord. The fact that you don't use it doesn't negate the fact that many others do. Yeah, the ranges are weird but almost nobody uses parameters. Yes, the generations can be hard to navigate but you do have a profile on midjourney's site where you can see a gallery. Yes, the ordering of the upscale/variation buttons isn't the most intuitive... in the first 2 minutes, but it's second nature after a couple of rolls. All in all, there will be less and less UX with generative AI. The interface is plain text and just voice in the very near future, and the output is something that requires mininmum tweaking. I'm sad to say this, as a UX professional, but UX will soon become obsolete. ChatGPT can interpret and analyse a 100 page PDF, then output a summary and plot it on a graph. How many clicks and workflows would that require just a couple of years ago - 200? 500? It's just a sentence now.
How in all f**k can you even dignify a product based entirely on theft with a UX review? They used exclusively pirated material and wreak havoc on the image professionals they stole from. It’s like criticizing the UX of a guillotine.
Pretty quick to label things, are you? Stolen is something that is copied 1:1. If your images are in the public domain where everyone is free to observe them and the later incorporate certain things in his own work, as "inspiration", then there's no theft. Gen-AI tools create novel images, yes, they might have "seen" and "trained" (gained inspiration) on your stuff, but it's not your stuff. Adapt. Or become obsolete. If you think original human art will lose its value because of things like MJ, then and only then we have a real problem. It's much the same as with ordering stuff on Fiverr or Upwork for a couple of bucks. A branding agency would charge you thousands of dollars for a corporate idtenity, yet you decide that's overpriced and go pay some Pakistani dude to design you a logo for $10... You get what you pay for.
Midjourney is a service that rents out access to a model. That model is based on quality images and text labels. It wouldn’t exist without those images and it wouldn’t be any good without quality images.
The normal way of sourcing training data is to license it in bulk from an image supplier like Getty, who negotiate rates and remunerate their licensees in turn.
In addition, you can use Public Domain images whose rights have expired. And if you need specialty datasets, you might even commission them with new, hand labeled and curated images. Visit datasetshop.com for an example.
This is no different from how you source images for your client projects. You might use any scrap found on Google Images or Pinterest for placeholder art, but you don’t put that in a sharp client project -- you either license or commission images.
Except Midjourney and Stability. They just took stuff, and use it to compete directly against the licensors and creators of their source data.
Publicly available for humans to view has nothing to do with Public Domain waivers of commercial rights. These are very different concepts.
You need to learn this if you are to have a professional future within design.
I’m currently on a team with a 30-to-one ratio of engineers to UX designers. Perhaps it will help if I share with them how the received wisdom of the past few decades could help us address the “idea abundance” problem using time-tested solutions that are both easy to implement and likely to be very useful. Thanks!
Nah, the interface works brilliantly. Midjourney has a sh*t interface, yet it's the dominant diffusion model. Why is that, you might ask? They are piggybacking on an existing platform, a massive platform such as Discord. The fact that you don't use it doesn't negate the fact that many others do. Yeah, the ranges are weird but almost nobody uses parameters. Yes, the generations can be hard to navigate but you do have a profile on midjourney's site where you can see a gallery. Yes, the ordering of the upscale/variation buttons isn't the most intuitive... in the first 2 minutes, but it's second nature after a couple of rolls. All in all, there will be less and less UX with generative AI. The interface is plain text and just voice in the very near future, and the output is something that requires mininmum tweaking. I'm sad to say this, as a UX professional, but UX will soon become obsolete. ChatGPT can interpret and analyse a 100 page PDF, then output a summary and plot it on a graph. How many clicks and workflows would that require just a couple of years ago - 200? 500? It's just a sentence now.
How in all f**k can you even dignify a product based entirely on theft with a UX review? They used exclusively pirated material and wreak havoc on the image professionals they stole from. It’s like criticizing the UX of a guillotine.
Pretty quick to label things, are you? Stolen is something that is copied 1:1. If your images are in the public domain where everyone is free to observe them and the later incorporate certain things in his own work, as "inspiration", then there's no theft. Gen-AI tools create novel images, yes, they might have "seen" and "trained" (gained inspiration) on your stuff, but it's not your stuff. Adapt. Or become obsolete. If you think original human art will lose its value because of things like MJ, then and only then we have a real problem. It's much the same as with ordering stuff on Fiverr or Upwork for a couple of bucks. A branding agency would charge you thousands of dollars for a corporate idtenity, yet you decide that's overpriced and go pay some Pakistani dude to design you a logo for $10... You get what you pay for.
Midjourney is a service that rents out access to a model. That model is based on quality images and text labels. It wouldn’t exist without those images and it wouldn’t be any good without quality images.
The normal way of sourcing training data is to license it in bulk from an image supplier like Getty, who negotiate rates and remunerate their licensees in turn.
In addition, you can use Public Domain images whose rights have expired. And if you need specialty datasets, you might even commission them with new, hand labeled and curated images. Visit datasetshop.com for an example.
This is no different from how you source images for your client projects. You might use any scrap found on Google Images or Pinterest for placeholder art, but you don’t put that in a sharp client project -- you either license or commission images.
Except Midjourney and Stability. They just took stuff, and use it to compete directly against the licensors and creators of their source data.
Publicly available for humans to view has nothing to do with Public Domain waivers of commercial rights. These are very different concepts.
You need to learn this if you are to have a professional future within design.
I’m currently on a team with a 30-to-one ratio of engineers to UX designers. Perhaps it will help if I share with them how the received wisdom of the past few decades could help us address the “idea abundance” problem using time-tested solutions that are both easy to implement and likely to be very useful. Thanks!