AI how-to: Create an original logo in under 10 minutes
Step-by-step guide on iterating with ChatGPT and DALL-E 3 to create logos
Last week, OpenAI opened up DALL-E 3, its premier image generation model, to all Plus and Enterprise ChatGPT users. While image generation models in the past have been an amusing play toy, this new model marries a new powerful image model with the GPT-4’s ability to craft fine tuned prompts that best suit it. Combined, the two now enable ChatGPT users to generate a wide variety of highly detailed original images in various styles.
Among those, logos. Something that seemed almost impossible with past image generation models is now fun and easy with GPT-4 and DALL-E 3 in ChatGPT. By just having a conversation with ChatGPT, anyone can put together a useable logo in 10 minutes or less.
The Pitch Meeting with GPT-4
Because of how intimately the prompts between the two models work together, it’s worth having GPT-4 generate a starting prompt before engaging with DALL-E 3. Remember, GPT-4 acts as the mediator, translating your vague ideas into precise prompts that DALL-E 3 can interpret.
Let’s say I’m working on starting a new coffee shop, and need a digital logo for our website (that I’ll eventually want to put on my brick and mortar store as well). My coffee shop’s name is “Azalea Brew”, so I know that I vaguely want some azalea and coffee theming.
I’ll start by throwing my embryonic idea at GPT-4—something like "Give me a prompt optimized for DALL-E 3 to generate a logo for a coffee shop with floral, azalea theming."
GPT-4 whipped out a more elaborate, fine-tuned prompt. It might say, "Generate a logo for a coffee shop that incorporates floral and azalea elements. The design should evoke a warm and inviting atmosphere while highlighting the unique azalea theme. Include a steaming coffee cup as the central focus, surrounded by azalea flowers and leaves. Opt for a color palette that harmonizes well with deep coffee browns and vibrant azalea pinks." This prompt carries substantial weight, loaded with nuanced instructions that can set DALL-E 3 exactly on the path we want.
Finetuning DALL-E 3’s First Iteration
Armed with your enhanced prompt, you're ready for the main event: a virtual logo design meeting with DALL-E 3. Initially, being specific about what you definitely want in your logo is key. Add some of the below keywords to the end of your GPT-4 generated prompt to ensure things are generated properly.
Vector: This keyword ensures that the design is scalable, maintaining quality at any size.
[Business Type] Logo: This descriptor gives context, nudging DALL-E 3 towards a design oriented for your particular use case. (For our example, we’ll go with “Coffee Shop Logo”)
Iconic: This prompts DALL-E 3 to aim for a design with memorable elements.
Minimal: This word signals a streamlined, uncluttered look.
Monochrome Black and White: Optional. This will ensure that your logo is only generated as black and white, allowing you to customize your colors to whatever you want on your own and encouraging a design that works in all sizes and situations.
With these added, here’s how our initial instruction looks: "Generate a logo for a coffee shop that incorporates floral and azalea elements. The design should evoke a warm and inviting atmosphere while highlighting the unique azalea theme. Include a steaming coffee cup as the central focus, surrounded by azalea flowers and leaves. Opt for a color palette that harmonizes well with deep coffee browns and vibrant azalea pinks. Vector, Coffee Shop Logo, Iconic, Minimal."
Here’s what we’re starting with:
As you can see, already DALL-E 3 has created some incredible useable logos. However, these are a bit too detailed for what I want. Let’s refine it to our specific tastes and ensure we get something we really want.
Iterating Further with DALL-E 3
While all of the initial generations were too detailed, I did particularly like how the fourth one looks. ChatGPT makes reiterating on this image extremely simple. Just converse with it as you normally would, providing details on what you’d like refined and being sure to reference the particular image you want worked on.
From here, you can continue in a feedback loop to iterate and fine-tune as much as desired. Use specific keywords to adjust elements like symmetry, line thickness, or geometric shapes. I’m a big fan of the upper right design, so I’m going to go ahead and download it by hovering over the image and clicking the download button that appears in the upper left.
If desired, you can attempt to get DALL-E 3 to add text to your design. This is typically hit or miss, with frequent misspellings and a tendency to change the logo design a bit in the process, so I would generally recommend adding text to the logo on your own after the fact. As you can see, attempting to do so in this case gives us less than desirable results:
If you find yourself wanting to go back a couple steps, you can grab the prompt that GPT-4 generated to create a specific image (it creates a new prompt for every single image!) by clicking on any of the image generated.
Copy-and-paste this prompt into a new ChatGPT DALL-E 3 session to start back from that particular design. They won’t be exactly the same (that’s AI for you), but they will be extremely similar and should help you restart at a particular point in the iterations.
Getting an Editable Vector Image
While ChatGPT is fantastic at designing vector-style logos… they’re not actually in a vector format. For a working vector, we’ll need to take our design over to a site like Vectorizer.ai. This online tool will convert your bitmap image into a crisp vector file instantaneously, which you can then import into design software for final touch-ups.
Here's a quick guide on how to use Vectorizer.ai:
Upload your generated logo to Vectorizer.ai.
Choose from various settings to customize the conversion, such as "curves" for smoother lines.
Download the vector file, usually available in SVG, EPS, or PDF formats.
And that’s it! Within around 10 minutes we’ve gone from ideation to creation with a wholly original vectorized logo that can be easily used online and offline to represent my new coffee shop. The combination of GPT-4 and DALL-E 3 show us just how powerful and useful the next wave of multi-modal generative models will be — and this is just the beginning.