Integrating AI into the Creative Workflow

Our Experiment

At Four Nine Digital, we value design, technology, and innovation. Besides client projects, our team frequently explores new tools and pushes boundaries. We wanted to explore ways to integrate AI into our creative process. Our experimental project: a custom vehicle wrap to bring the CyberTruck to the Dune universe. We will walk you through our experience and discuss some of the challenges we faced along the way.

Disclaimer: Ironically, in the Dune Novels written by Frank Herbert, “thinking machines” were banned after a catastrophic war between humans and AI. The purpose of this blog post is not to focus on AI’s impact on our lives, the job market, and careers. We don’t know how society will adapt and regulate these tools yet. That’s a discussion for another day.

Preparing the ground

A typical step in visual projects is to collect reference images to help us set the tone and get a sense of the look and feel we’ll pursue. Here is our mood board for the project with Dune and CyberTruck-related images.

Which tools did we explore?

Before starting the fun part, initial research was necessary to define which AI tools to use. We decided to focus on those that do not require additional costs. We already have the Adobe Creative Cloud subscription, so Adobe Firefly was an easy choice. The other two were Dall-E3 (integrated with Microsoft’s CoPilot) and OpenArt AI, which offer many credits and have similar functionality.

Exploring generative AI tools

Step 1

We started with a basic prompt, exploring the three tools, to understand how they work. You’ll see what we got from each one.

The Dune movie brings inspiration from Middle Eastern culture. This was our initial prompt: Intricate Middle Eastern patterns, black, shiny reflective mixed with opaque matte.

Step 2

For another set of explorations, we used something more specific to the movie to see what kinds of designs we could get from the tools: Dune movie inspired minimalist artwork, flat design, black and white shapes.

Step 3

We used different ways to describe our request while getting familiar with the tools: Line artwork based on the Dune movie.

Step 4

This time, we focused on a different aspect: Dune inspired creative design, based on the spice effect on people’s eyes, with blue details and dreamlike vibes.

It was exciting to create images quickly, but we still wanted to test other ways AI could bring a creative solution. Based on the results we generated, we reflected on it and tried another perspective. As an example, one of the flying vehicles (the ornithopter) is inspired by the dragonfly.

Step 5

Drawing parallels to the Ornithopter, we brainstormed desert animals that could be used to inspire the vehicle’s design. Check below how AI explored this creative challenge by combining an animal with a vehicle.

Through this experience, we learned the value of clearly defining the ideas you want to work on in the early stages. AI allowed us to quickly explore various concepts with minimal time and effort. However, it’s easy to get carried away, diving deeper into exploration and losing track of time as excitement takes over.

Choosing one AI tool

After an exploration of different generative AIs, we gained a basic understanding of how they work, but a more detailed analysis is needed to fully capture the nuances of each tool.

Here are some things we learned: Firefly offered customization features that could be combined with the prompt. OpenArt AI provided a good amount of credits, but additional features like image guidance and upscaling require a subscription. During the exploration phase, we found that DALL·E 3 didn’t allow for fine-tuning the results. As an alternative, users could try other editing capabilities available through Microsoft CoPilot integration.

We found that the most efficient approach was to concentrate on a single tool. Our choice was Adobe Firefly. The other tools are valid options but we decided to focus on one to refine our process and gain depth of knowledge.

An extra motivation for choosing it came from a DesignThinkers presentation in Vancouver this year. Tomasz Opasinski, a Creative Technologist at Adobe, shared his process with Firefly.

An approach with AI

One of the insights we gained from Tomasz’s presentation was to focus on creating design assets for the project rather than attempting to write a “perfect” and highly detailed prompt. By generating and manually assembling smaller parts of the design, you reduce the need to explain every detail as one prompt and avoid relying on the AI to produce one complex final output. You can check this video and this article to understand his process better.

Creating the canvas structure

We wanted a specific design and the possibility of checking how the wrap would look from different angles. So first, we needed to work on the wrap design and then apply it to the CyberTruck in distinct stages.

When working with 3D objects in a 2D environment, it’s common to divide the design into separate images. The wrap design consists of separate panels. Think about the car’s sides: each panel is a surface with a specific height and width.

For this experiment, we investigated two options:

• A: Generating isolated panels in different rounds. Using separate reference images.
• B: Creating one image reference to generate all the assets at once.

We discovered both methods had problems with the consistency of the generated textures across all panels, but option A was better regarding the image size; we didn’t have to divide the final output. It helped reduce the work on upscaling the images to achieve a good resolution.

Refining the process

In Firefly we started to explore ways to get more useful and precise results. In addition to the prompts, we’ve found that uploading reference images can be very helpful. We recommend using one reference image for the composition structure and another for the visual style reference. Both images can be adjusted using sliders to control intensity.

There are also additional options available to experiment with visual effects, colour and tone, lighting, and camera angles.

We found that these adjustments brought us closer to the desired look. However, they may be time-consuming until you become more familiar with the tool. We had to balance the prompts with the sliders while using different references in order to achieve better results.

After more rounds of generating images, we filtered the best ones to proceed with the design experiment. The images below are based on variations of the following prompt: Rendering of metallic matt black shapes set on a white background.

It’s fair to mention that additional tweaks on software like Photoshop will be necessary to reach a precise output. Sometimes, the results can be inconsistent. The process requires combining the AI-generated images with digital editing using more traditional tools.

We used Photoshop to balance the image contrast, colours, adjust light, add the Atreides logo and clean up some unwanted shapes AI can generate when interpreting our prompt.

Implementing the design

We chose a conventional method to apply the design to a CyberTruck, allowing us more control over the results. Another point is that with Firefly, you can create futuristic and cool car designs but not use a specific model due to image rights and other technical capabilities.

Pros and cons

When exploring a new tool, there’s always a learning curve, things we discover and improve. For this project, after testing with generative AIs and going deeper with Firefly, we can list some positive and negative aspects found during the process.

Pros:

• Save time during the initial stages by experimenting with ideas quickly.
• Try more ideas in a short amount of time.
• Discover new creative solutions when exploring with the tool.

Cons:

• Lack of full control over the designs you’re generating.
• Experience rounds of generation that don’t deliver useful results.
• Risk going down a rabbit hole when trying new ideas and refining outputs.

Based on our experience with Adobe Firefly, we recommend uploading your structure and style image references to get more specific results.

A good thing is that you can select between different ratios to generate images, but depending on the resolution you’re looking for, you’ll need to spend extra time upscaling images on software like Photoshop. We still consider the combo Illustrator + Photoshop a powerful choice to get your final designs the desired look and feel.

Here you can see our experiment integrating AI into the creative workflow.

Bonus

Here is one of the alternate designs we created during our process with Adobe Firefly.

Conclusion

We’ve reached a time when artificial intelligence is widely accessible, and these tools are incredibly captivating. Things are progressing rapidly, so it’s important to stay informed about new updates and continue exploring AI.

We’re excited to share the initial phase of our exploration, but the journey of learning and improving with AI is ongoing. We’re eager to learn from your experiences as well. How do you integrate AI into your creative workflow?

Four Nine is a full-service Digital Agency located in Vancouver, BC. Learn more about us by visiting our website or contact us at info@fournine.digital