Why we are building an AI design tool that skips the prompt

 

And suddenly, every piece of software is a chatbot. The frenzy created in 2023 by the apparent success of ChatGPT had a lot of influence on many aspects of tech companies, and user interfaces couldn’t be left behind. Conversational user interfaces (CUI) or chatbots are now the modus operandi of every company that has artificial intelligence at the core of its products. But have chatbots prevailed due to their superior functionality or are they just a trend?

Undoubtedly, they have many benefits. On the customer side, the learning curve is very small to start with (while difficult to master). There are no screens, buttons, and flows to navigate or learn. There lies an empty screen with a single input field. The chances of missing it and clicking somewhere else are microscopically small. This arrangement also suits product development by solving a well-known problem of Graphical User Interfaces (GUI), namely “where should we place this new feature?” Since there is only so much real estate on a screen, you eventually run out of space to place buttons for new features. You then end up hiding them inside sub-menus three clicks deep, making the discovery of each feature a brainteaser. This is among the reasons why complex software, like Adobe Photoshop, takes so many years to master:

Therefore, chatbots are a good starting point for directing users to the appropriate feature, something now known as on-the-fly or Generative UI. They also help companies get a view inside people’s minds as customers simply tell you what they are trying to do in plain language. Contrast this with traditional SaaS where you need annoying onboarding questions and endless user sessions to infer what the end goal is.

So if chatbots are everywhere, it means they are the superior option for all use cases right?.. right?

Well, like everything in business (and life), they come with significant trade-offs. First of all, great software products make it super clear how they should be used. The only indication that chatbots give is that you must insert characters inside a text input. Especially new or novice users have no clues about what the capabilities are or which set of words can yield the best results. Each user must learn by himself how to use the product by constant trial and error. This has now spawned a new discipline, prompt engineering.

For some tasks, chatbots can feel as intimidating as staring at a blank canvas. Consider this from Andrej Karpathy:

"In particular, in terms of the UIUX, "chatting" with LLMs is a bit like issuing commands to a computer console in the 1980s. Text is the raw/favored data representation for computers (and LLMs), but it is not the favored format for people, especially at the input. People actually dislike reading text - it is slow and effortful. Instead, people love to consume information visually and spatially and this is why the GUI has been invented in traditional computing."

People prefer visual information and it’s often faster to click a button than to type a long text command, especially when the task at hand is complex.

 

Prompting and graphic design

We believe graphic design is one of those complex tasks.

Graphic design artifacts like posters are a synthesis of many elements, namely typography, imagery, decorative ornaments, color theory, brand identity and spatial layout, combined in a way that’s both beautiful and effective at communication.

When the average user, with no design or prompt skills, is faced with a chatbot, he defers to the simplest instruction possible. Since AI models are predictive machines, they default to the most likely output, the one they have seen the most during their training. That's why all models have a default design style and why you keep seeing the same flyer everywhere. Ten companies asking a professional designer for a modern flyer get ten widely different aesthetics and art directions. Ten companies asking a model get pretty similar designs with Inter fonts and beige backgrounds. We tried it:

Repetitive and mediocre content, or AI slop, is in large part a consequence of the omnipresence of chatbots. Unique or professionally looking results can only be achieved via long and detailed prompts describing every element of the design. This requires advanced prompt and design skills and the average user has neither. Plus, long prompts undermine the whole purpose of using AI to save time.

As per Amelia Wattenberger, “natural language is great at rough direction: teleport me to the right neighborhood. But once ChatGPT has responded, how do I get it to take me to the right house?” So how can we help a user get to his ideal “modern” design as fast as possible?

 

Replacing prompts with visual references

At Grafista, we are building an AI graphic designer where amateurs get professional-looking results without design skills. Instead of a chatbot, you start by adding your content inside a modern text editor (or generate text if you don’t have it yet). Starting with your content is essential as designs should be built around what you have to say. Emphasis is shifted depending on what the primary information is or whether there are photos. This flow feels more natural and mimics the process of outsourcing to a designer which starts with your goal and your message.

For defining how the design should look, prompts are replaced with Blueprints which are essentially visual references. Grafista features a wide variety of design directions and art styles like Art Deco and and Futurism. 

 

 

If you choose a Blueprint, its layout, fonts, colors and unique characteristics are applied to your content automatically. Blueprints are flexible enough so that the aesthetic of a simple A4 flyer can be easily used to create a 10-page presentation with long-form text. Combine a Blueprint with your Brand Kit and you get a result that feels yours.

 

 

Giving users the option to visualize upfront what their final design will look like is a great way to help them arrive at their ideal destination fast, without having to generate 10 options. And since graphic design is an iterative process, every element of the generated design stays fully editable so users remain in full control.

Try it yourself for free at grafista.io