Aside from flooding my LinkedIn with Studio Ghibli images of everything (and sparking a deep discussion on the impact of AI on creative endeavours), we were pretty impressed with ChatGPT’s new image generation feature (read about our first attempt).
We have been having a deeper explore to see just how useful the new capabilities are for us.
Unblurring a Photo
On Tuesday SW*AID (South West and Wales Women in AI and Data) ran a lovely Bring Your Own Learning event, and I took one photo. Of course it was blurred.
My terrible attempt at event photography
L: Can you un blur this image
ChatGPT's attempt at unblurring my terrible photo
This weirded me out a little as the picture looks like a real photo - but this isn’t what these women look like. The hair is right-ish in most cases but the facial features are not. It also completely removed one person from the photo. Another odd choice it turned my blurred sunglasses into a pair of normal glasses.
L: Can you make a cartoonish version of the original?
ChatGPT's cartoonish representation of the SW*AID BYOL session
I do really like this - it more captures the women and the vibe of the room.
One thought that occurred to me was about permission - I had verbal permission to use this image on social media, etc. but do we need to be asking for separate permission for using image of people in AI models? This felt like it didn’t matter as it was blurred, but that’s just my opinion.
Editing a Canva Image
We use Canva to create our LinkedIn images - they are pretty similar each time with just an edit of text and images. But as we use ChatGPT to proofread our blogs, it would save time to have an image created there as well (especially if ChatGPT was able to just update it as well).
The original LinkedIn post image - made by me in Canva
L: can you change the text from Spain's Proposed Bill to "UK Looks into AI Act"
L: can you change the text from Spain's Proposed Bill to "UK Looks into AI Regulation" and "Cracking down on deepfakes" to "Following in the EU AI Act's footsteps"
L: you've added a bit twice
The progression of ChatGPT edited LinkedIn post images
Some things we noticed:
- ChatGPT was able to edit one detail at a time - any more and it seemed to get confused and make mistakes. While this isn’t an issue, it feels like a waste of computing power (although it did ask if I wanted the image to be regenerated when I mentioned the mistake).
- When trying to do this on the train, the spotty internet (which doesn’t impact chatting to ChatGPT that much) meant the images kept not being created.
Creating An Infographic
As ChatGPT can now put words in images properly, I’m excited to test out the infographic capabilities as we’d love to share more of these on our LinkedIn (as they are great for summarising topics).
I decided to check what context I should be providing to ChatGPT to get the best infographic.
L: i want you to create an infographic - what information do you need from me?
It gave me a long list which I replied to (but made a point of being a bit vague as I’m intrigued)
Here is all the information I provide in my prompt:
1. Topic and Purpose
- What is the infographic about? Different AI Tools and when to use them
- What’s the main point or message you want to get across? There are lots out there - use the one you have access to (and it works better for different tasks)
- Who is the intended audience (e.g. general public, businesses, policymakers)? People on LinkedIn (we have a range of tech, marketing, accounting, educators) etc. following us)
2. Content Please provide:
- Key facts or data points (stats, quotes, figures, dates, etc.) ChatGPT released in Nov 2023 also Copilot, Claude, Perplexity, Deepseek
- Sections or categories you want to include (e.g. problem → solution, timeline, benefits, comparison)
- Any headings or subheadings you want me to use
- Any sources if data needs to be cited
3. Style & Branding
- Do you want it to match your brand style (colours, fonts, tone)? Brand style in our brand colours background - #ffffff font color - #192b4d #7a28cb #41e2ba #5cc8ff #f9b9f2 #f7dd72 fonts - kolletif or roboto we use triangles and dots a lot
- Would you prefer it to look professional, playful, minimal, bold, etc.? professional but a little playful
4. Format
- Where will you use this infographic? (LinkedIn, a blog, a presentation, print, etc.) LinkedIn
- Preferred size or orientation? (Portrait, square, A4, etc.) For LinkedIn
It actually replied with ideas and suggestions for the layout (which is an improvement rather than just jumping straight in). If you’re interested:
ChatGPT: Perfect — thanks for all that! Here's a quick content plan for the infographic before I design it visually. Let me know if you'd like to change or add anything:
📌 Infographic Title
"Which AI Tool Should You Use?"
Subheading: There are loads out there – use what you have access to (and pick the best one for the job).
🧠 Tools to Include
Each tool could sit in a coloured box with:
Tool name + logo/icon
- One-line summary of its strength
- Best for…
- Released (if relevant)
🗨️ ChatGPT (OpenAI)
Best for: General tasks, text, creative writing, summaries
Strength: Most widely used, great at everyday questions and content
Released: Nov 2023 (GPT-4 Turbo)
🧑💼 Microsoft Copilot
Best for: Word, Excel, PowerPoint tasks; business users
Strength: Works inside Microsoft 365, great with documents and spreadsheets
Note: Access via work account
📚 Claude (Anthropic)
Best for: Reading long documents, thoughtful answers
Strength: Reads and reasons over big files, safer design
Released: 2023
🔍 Perplexity AI
Best for: Research, citations, up-to-date info
Strength: Combines AI with web search
👩🔬 Deepseek
Best for: Technical queries, coding
Strength: Trained with a focus on scientific and code content
New!
💡 Bottom Section
“There’s no best tool — just the best tool for the task.”
Pick one you can access. Try it out. They’re changing fast.
Would you like me to go ahead and create the visual version of this now with your branding (for LinkedIn, portrait or square)?
L: no using emdashes (use hyphens) and British english spelling. Make sure it's in Taught by Humans tone of voice
A ChatGPT generated infographic on AI tools
One thing I didn’t catch was a small mistake that GPT-4-turbo was released in Nov 2023 (it was actually 3.5). And the infographic is actually cut off at the bottom.
L: can you update:
- the deepseek is cut off so maybe just remove that one?
- are you able to use logos?
- there is a small error - it was not GPT-4-turbo was released in Nov 2023 (it was actually 3.5)
ChatGPT: Thanks – great catches! Here’s what I’ll do for the next version:
✅ Updates
Remove Deepseek – it’s a niche tool anyway, so no problem trimming it.
- Use Logos – yes, I can add simple, recognisable logos for each tool to help with visual clarity and brand recognition.
- Fix ChatGPT release info – I’ll update it to:
Released: Nov 2022 (GPT-3.5)
GPT-4 Turbo came later and is available via ChatGPT Plus
Would you like me to:
Keep the same layout but just adjust those bits?
Or slightly rebalance the layout now that we’re down to four tools?
Let me know, and I’ll get the updated version generated.
I do enjoy the back and forth, makes this feel a bit less like a tool randomly doing what it wants. I had to call out another small mistake.
L: "GPT-4 Turbo came later and is available via ChatGPT Plus" don't include this (it's outdated info)
but go with rebalancing
A ChatGPT generated infographic on AI tools (second attempt)
While this is definitely an improvement, the logos for Claude and Perplexity are not right. (and our triangles are a bit off).
If I was to continue, I would provide the logos. Maybe the background as well - like for the Canva post. But this is definitely looking promising for helping us make asset fasters.
Known Limitations
OpenAI themselves have highlighted some known limitations:
- Cropping (which we’ve seen)
- Hallucinations (a general problem across these AI tools)
- High binding problems (asking it to make a poster with lots of knowledge based elements can result in issues)
- Precise graphing (I personally would never trust this type of AI to make a detailed graph as I would still need to check every point, but that’s the data person in me)
- Multilingual text rendering (unsurprising in a world geared towards English)
- Editing precision (happens especially when editing drawings - which we saw last week when editing Dotly)
- Dense information with small text (probably bad practice for what we’re trying to do)
It’s important to be aware of the limitations of any tech tool you’re using. We’re still really impressed with this tool, and will be refining how we use it (and what we use it for). If you have any specific ideas you’d like us to try out, let us know via our contact form or LinkedIn.