Before the visual interfaces, we had command lines.
All commands had to be typed in manually. There was no need for a mouse cursor. Then came the visual interface we all know and love.
While visual interfaces are ubiquitous, those aren’t the most natural ways for humans to interact with software. Do you know what is? – Speech.
In the case of Midjourney, it’s written speech. Early mainstream AI products are based on the dream of no UI (or technically Low-UI). GPT-2, the ChatGPT earliest version (without the chat), was released in 2019.
But it was not until ChatGPT’s magical experience and Midjourney-like text2image that AI became mainstream. It’s UX that brought AI from obscurity and pushed it into the spotlight.
Let’s dive deeper into this, and use Midjourney as our primary example.
One Plugin for All Workshop Needs
26 templates; Get stakeholder buy-in; Become a better UX'er.
Get AI PluginLearnability
Learnability is one of the main bottlenecks to product adoption. Especially if a product is powerful. In the case of ChatGPT and Midjourney, the learning time was reduced to virtually 0. Well... we all know how chats work.
To put that into perspective: GPT-3 with its slightly more complex interface was out there 33 months (!) before ChatGPT. It completely flew under the radar.
Stable Diffusion, despite being revolutionary, received similar adoptions.
Here’s what we can learn from this:
Don’t clutter your interface if you seek adoption;
Don’t reinvent UX... the value of your product should be somewhere else;
Provide / show value ASAP;
Discoverability
I’m sure you come across social media posts like these:
The fact that similar posts perform super well could only mean one thing. People don’t really know what to do with AI. Sure, it's cool, but what can people use it for? – Unless you working in a tech industry or in one adjacent to it, it isn't obvious.
If users can’t go beyond superficial features to grasp the real value, they will never convert into paying customers. A great product guides users to become power users asap. Think of it as a buyer journey:
You start simple and work your way toward complexity. That’s how you avoid as much churn as possible.
Here’s what we can learn from this:
Novice users rarely convert into paying customers.
Empower them in their journey
Make sure they understand how to take advantage of your product and become more effective.
Engagement
Using Midjourney for the first time is exciting and overwhelming due to its social nature through Discord. The constant flow of generated images can inspire new ideas. However, it’s very distracting.
Take a second to compare both experiences as a first-time user. Which one is more engaging?
Here’s what we can learn from this:
Include meaningful social aspect;
Learning from others can be an empowering experience. Leading to the Aha moment;
Be careful of consequences though and plan everything around the social aspect in advance (like moderation, etc.).
Shortcuts
A good way to create an effective experience is to use defaults, shortcuts, and presets. The newly launched Adobe Firefly does that very well.
Other than prompts, you can quickly add presets to your prompts.
Helping users find shortcuts makes their life easier. Notion does that very well, too:
Github Copilot does this by providing autocomplete suggestions in the background:
It’s unfortunate that Midjourney and ChatGPT haven’t yet introduced an easy way to use prompt presets. In my personal experience, ~60% of my ChatGPT & Midjourney queries are similar enough to have them saved as a preset.
Here’s what we can learn from this:
To improve user experience, provide shortcuts and pre-sets;
That especially applies to actions that are likely to be repeated over and over.
UX Writing
Many AI products use technical terms or abbreviations with no additional info, and often even the explanations don’t help much. Just look at the example below.
You can’t have users utilize your product fully if they can’t get through the jargon.
Here’s what we can learn from this:
UX is not reduced to a UX designer role only;
Invest in UX writing;
Test your copy with users.
Low-UI shortcomings
While Low-UI is technically a step forward in terms of usability, complexity is not necessarily a bad thing. Especially when it comes to effectiveness.
Visual cues, feedback, and digital or physical objects help us in our cognitive process. And that’s perceivable when working with the first mass AI products. That’s why many “AI” products out there are merely interfaces on top of ChatGPT and co.
Another example of interface usefulness is DALLE2. Despite having a worse output quality than Midjourney. DALLE 2 has inpainting and outpainting which allow you to put new objects into existing images. – For certain tasks, It's endlessly more convenient than tinkering with Midjourney.
Privacy
Privacy is a big topic when it comes to AI products. But let’s focus on the privacy of what YOU generate. Generating images on a public discord channel can be intimidating (and also a messy experience). And very few people know they can integrate Midjourney into their own discord channel for FREE.
Images generated on private servers are still visible to other users on midjourney.com except for Pro Plan when activating stealth mode.
Customization
Imagine using Midjourney for generating blog thumbnails. Or ChatGPT to help with writing articles. You’d think it was possible to save the tone of voice and illustration style as a preset. – It isn’t. You have to save a part of the prompt elsewhere to use later.
Additionally, ChatGPT quickly "forgets" much of the context anyway. So using the tool for the same task for the 100th time is the same as the first with next to no increase in efficiency.
Tips
Power users tend to customize their interfaces;
Despite their low numbers, these users are willing to pay higher than average and have a high engagement, and can be turned into ambassadors;
Make their lives easier and their workflows more efficient.
Set expectations
At this point, we’ve all had ChatGPT spew nonsense. Unless you check yourself, there’s no way you can trust the output 100%.
Additionally, you can make an argument that AI mirrors the biases of the data it was trained on.
Therefore, you need to make sure you set the right expectations and inform users about the shortcomings. Another limitation of chat-based tools like ChatGPT is their context.
Spoiler: ChatGPT chats are a lie!
ChatGPT context memory is 4096 tokens (around 3k words) and GPT-4 is 8192. It means despite following the same conversation thread, everything after that amount of tokens (word) is forgotten...! Coming back to Notion:
Bottom line
The advent of Midjourney and other mainstream AI tools signifies a shift toward a speech-driven interface. Essentially, you can have a conversation with your software.
The success of Midjourney and other tools teaches us several lessons on improving product adoption, such as reducing learning time, guiding users to become power users, including meaningful social aspects, etc.
By implementing these practical insights, you can optimize your product's design and functionality, increase engagement and conversions, and turn your power users into loyal brand ambassadors.
