Project: Maestro - Substack Newsletter - Wordmark

I needed a title or banner image which Substack calls a “Wordmark”. The models struggled with this concept on a few levels. I used the following request prompt:

Give me ideas for a wordmark for a Substack newsletter called 
"Project Maestro" about using AI to design and develop a 
consumer application.

NOTE: I had to abandon the prompt above and used the following:

Describe the design aesthetics of a stylized image version of 
newsletter title "Project Maestro". The topic of the newsletter 
is using AI to design and develop a consumer application.

Strategy

Stack multiple AI technologies in three phases:

1. Theme Creation - Brainstorm themes using large language models
2. Prompt Optimization - Use LLM to build efficient text-to-image prompts
3. Image Generation - Create images using generative models

Theme Creation

Models really struggled when I started including words or titles. They also did not understand that I wanted a banner style image with the newsletter title. This required a moderate amount of prompt experimentation to get desired result

Where did they succeed?

• Good suggestions on font and
• Color pallete aligned with logo
• Provided different conceptual themes

Where did they fail?

• Did NOT understand the term “Wordmark”
• Had to experiment with prompt phrases to get desired output

AI Tool Report Card

Notes

1. ChatGPT (A) - Provided font and style options
2. CoPilot (A) - Provided font and style options
3. Claude.Ai (A) - Provided font and style options
4. Bard (C+) - Provided multiple concepts for newsletter names
5. Grok (C) - Fluff and not much substance

Prompt Optimization

Prompt outputs were regurgitated from my input prompt. No real improvement other than abbreviating the commands. Co-Pilot auto-generated images, which was nice.

AI Tool Report Card

Notes

1. CoPilot (A+) - Generated 4 usable options
2. Claude.Ai (C) - Abbreviated my original prompt (no improvement)
3. ChatGPT (C) - Abbreviated my original prompt (no improvement)
4. Bard (C) - Abbreviated my original prompt (no improvement)
5. Grok (F) - Regurgitated my request

Image Generation

Images were square, busy and unusable. Created the banner directly in Microsoft Designer where had more control over image.

Where did they succeed?

• Designer’s image suggestions were useful
  • Text generation was better than Dalle-E
• Using AI to modify images is MUCH more effective than prior tools
  • Background removal, palette shifts, color replacement

Where did they fail?

• Pure prompts generate images were sub-par
• Requires a lot of iterations and prompt tweaking to generate anything remotely usable
• Hallucinations are a problem and lead to design drift
• Dalle-E REALLY struggles with text

AI Tool Report Card

Notes

1. Microsoft Designer (A) - Winning image, had to crop image to 21:4 aspect ratio
2. CoPilot (B-) - Aspect ratio wrong, required edits
3. Dalle-E (C) - Unusable images, text misspelled
4. Stable Diffusion (F) - Weird shapes and colors (was that a bowl of lettuce?)

Previous Article | Next Article