Can AI create a logo kit for my site?
- Niv Nissenson
- Apr 3
- 3 min read

The Test: Can Gemini create a logo for my blog based on an existing logo style?
Success Criteria:
Generate a logo consistent with the original brand
Provide a usable logo kit (multiple formats, transparent PNGs)
Minimal manual correction required
Weighted result: 3/5
Verdict: a mixed bag. Gemini generated decent logo but it can't do a logo kit. It can't create transparent PNGs (Chat GPT can) and started breaking down
This wasn’t intended to be a test but as I started to work on it I thought it could be a could a perfect setting for a light first test. Since Nano Banana came out I tended to use it for most of my images as it works much faster (and usually better than Chat GPT).
The Setup: I used Google Gemini (Nano Banana 2) and provided an existing logo from my blog The CFO AI as a reference.
The goal was simple: create a similar logo for my new blog, The Tester AI, and then generate a full logo kit (different formats and backgrounds).
Test execution:

Prompt: Here's a logo I have for my blog "the CFO AI" www.thecfoai.com. I'm now doing a sister blog that's going to be called "the Tester AI" for which I secured the domain www.thetesterai.com. Can you make me a logo for my "the Tester AI" blog that is similar in style to the "the CFO AI" blog?
The initial output was a reasonable interpretation of the original logo. It captured the general look and feel, but there were clear issues:
The “neural” lines were inconsistent and shaky
The model added the domain name without being asked

Iteration 1: I asked Gemini to improve the neural lines.
It did improve them but is added sparkles (not requested and not appropriate for a logo)


Iteration 2: Once I had a usable base logo, I asked for a full logo kit:
White text on transparent
Black text on transparent
White text on black
Black text on white
All 1024x1024
Output difficulties:
Instead of generating four separate files, Gemini produced a single combined image with all variations.

Clarification prompt: I need separate files for each request, not one image.
Gemini then proceeded to exhibit the same output.
Iteration 3: I tried breaking the request into separate prompts.
This is where a more serious issue appeared.
Instead of creating a true transparent PNG, Gemini generated images with a checkerboard background a visual imitation of transparency.

At this point, the model wasn’t just failing—it was producing outputs that looked correct but weren’t actually usable. You have to ask yourself if the AI is just trying to fool you at this point.
So I asked Gemini directly if it could generate transparent PNGs.
It responded:
“The short and direct answer is no… my image generation tool cannot currently generate files with true alpha-channel transparency.”
In other words:
It cannot create real transparent images
It simulates transparency visually
This explains the repeated “fake” outputs.

Attempt to Recover
Gemini offered a black-on-white version, but the process became inconsistent:
It got stuck in a loop
Required a page reload
Still introduced unexpected design changes
Switching to ChatGPT
I then tested whether ChatGPT could fix the output.
I uploaded the image and asked it to:
Remove the domain name
Remove the Gemini branding
It claimed to complete both tasks—but did neither.

Only after explicitly requesting a transparent PNG did it partially comply and produce a usable result.

Summary
AI can get you to a rough logo quickly, but struggles with execution details that matter in real-world use especially file formats, consistency, and control.
Gemini produced a usable concept, but failed at delivering a proper logo kit. ChatGPT was able to produce a transparent PNG but also showed inconsistencies in execution.
Can a human do it better that AI?
Yes — but it will cost. A designer would deliver a more consistent and production-ready result, including proper file formats and a complete logo kit but it will cost.
Scorecard
Category | Score (5) | Notes |
Output Delivered | 3.5 | Produced a usable logo |
Hallucinations | 2 | Fake transparency issues |
Quality | 3 | Basic design, inconsistent elements |
Ease of Use | 3 | Required multiple iterations |
Reliability | 2.5 | Inconsistent behavior |
Bottom line | 2.8 | A mixed bag |


