I volunteer for a local charity and they’ve asked me to start sharing images on their social channels. I don’t have a stock library so how can I utilize AI to create images quickly for specific events?
The creation of images to share on social media channels like Instagram, Facebook, X (née Twitter), and Bluesky is an excellent use case for AI-powered images. You’re still in control of the details and specifications of the image, but you don’t have to be a graphics expert or artist to be able to succeed – and quickly! If your local non-profit is like most, it also has no budget to hire graphic artists and it’s risky to just use a Google search to find candidate images and republish them without permission.
A few months ago, AI image generation required you to jump through a number of hoops but things have gotten a lot easier with Microsoft’s slick integration of Microsoft Copilot into the Edge web browser and Windows itself. A click and you’re talking to ChatGPT 3.5 and, though it’s not obvious, OpenAI’s Dall-E image generation system too. There are other AI image generation tools and we’ll peek at a few later, but for now, Copilot is going to be our consulting graphic artist.
Shortcuts: Microsoft Copilot | Ideogram.AI | Midjourney | Adobe Firefly | Conclusions
Since I don’t know anything about your charity I am going to take advantage of the fact that today, Feb 16, is National Tartar Sauce Day. I mean, who doesn’t love a bit of tartar sauce with their fish sticks? 🙂
WORKING WITH MICROSOFT COPILOT
The real secret to getting great results from AI tools is to have detailed prompts, but sometimes you want to work your way into that by offering something fairly vague and seeing what the system will create. That’s how I’m going to start. In Edge I click on the Copilot button on the top right, then type in:
Definitely vague. Note that the key phrase to have Microsoft Copilot produce images is “create a picture“. That tells it to utilize Dall-E rather than answer with text, a typical ChatGPT interaction. It thinks for a while and produces this grid of four possible images:
Pretty cool looking, actually, though the lower right one’s a bit weird with the deep fried fish with the head still visible! Not my idea of a great lunch, I have to admit. 😆
One thing I really like about Copilot is that it offers suggestions to refine the image:
I like “Make it look like an advertisement for tartar sauce” though “Can you add some text?” is intriguing too (particularly since AI is so darn bad with text). I’ll click on the first and here’s what it produces:
What’s curious is that it also generated four more images to consider:
A close look reveals that it changed its own prompt along the way (“advertisement for an ad…”). While almost all of these have worse spelling than my cat running across the keyboard, the top right image of the second grid is actually pretty good:
Not exactly sure about the middle right image (fish sticks in the style of a hotdog on a bun with ketchup?) and the tartar sauce is looking a bit more like chipotle mayo, but still, it’s a fun, lively image, and the main text has no misspellings, making it a prime candidate for this project.
I like to experiment with other AI image creation tools when I’m producing content, even knowing that they all have lots of challenges with words, so I’m going to try a few more…
CAN IDEOGRAM.AI DO BETTER?
One of my go-to tools for images that include text is Ideogram.ai. Usually, it nails it. I’m also going to demonstrate a significantly more detailed prompt to see if I can get just what I’m visualizing after having explored various image ideas:
I really like the ability to change image aspect ratio here, but how will it do with this fairly detailed prompt?
Curious but I’m not sure any of them are 1950s style print advertisements stylistically. The spelling’s pretty close but not quite right on two of them but… the other two are good, if not exactly what I described. I decide to tweak the prompts to say “include deep fried fish and chips and beer”, but it seems to then completely mess up the spelling:
This isn’t going in a good direction, so it’s time to move to the next tool…
ROUND THREE: MIDJOURNEYAI.AI
Midjourney is considered one of the best AI image creation tools at this point but that’s mostly for photorealistic images, which we aren’t trying to create. Our prompt is now:
“create an advertisement for national tartar sauce day with the text “happy national tartar sauce day” and a catchy slogan in the style of nostalgic 1950s print advertisements that appeared in magazines of the era. Include deep fried fish and chips and beer.”
Here’s what it delivers:
The design of both are great but the fish and bread both look weird and unappetizing and, most of all, the spelling is atrocious! Also, um, a beer can of “Tarttar”? Not exactly the concept we’re seeking.
ONE MORE TRY: ADOBE FIREFLY
Another favorite of mine is Adobe’s Firefly AI software, part of its Creative Suite. Same prompt again, and this time the results are, well, interesting:
You can see that it gets a bit confused about the idea of both beer and tartar sauce in the same image, offering up the disgusting prospect of a beer with a tartar sauce topping, as if it’s sculpted foam. Uh, no. The spelling is downright bizarre too, with none of them even slightly close to the desired.
CONCLUSION: COPILOT FOR THE WIN
There’s no other conclusion possible here than that for this particular task, Microsoft Copilot was the winner. While none of the images created are perfect, the ones created by Copilot were some of the most interesting and when we lucked out and the words were spelled correctly, we were good to go. Now, let’s grab some fish and chips and enjoy our National Tartar Sauce Day celebration!
Pro Tip: I’ve been writing about AI for a while now. Please check out my AI and ChatGPT Help Area for more tutorials and help articles while you’re visiting!