Does Adobe Firefly Generate Believable Equestrian Graphics?
If you’ve spent time on social media browsing horse-related posts, you’ve probably seen horse images generated by artificial intelligence. A few of these are believable, but there’s usually a tell-tale sign that something isn’t right—like an extra leg or oddly floating tack. Some of these “photographs” are downright ludicrous.
Most people posting this material attempt to pass it off as real, and they collect likes and comments by saying something like, “Help me name my beautiful boy!” (You know who you are.)
I’ve had so much fun looking at these silly AI-generated images, that I decided it was time to play around a little myself.
I entered simple horse-related prompts into Adobe Firefly. Then, I sat back to enjoy the results! There was far too much to include in one post, but I’ve included some highlights. Each image below is paired with the actual prompt I entered to generate it outlined in quotation marks. I do not advise reading on with a full bladder. Here’s what I learned:
1. Generative AI models don’t understand equestrian disciplines or topics yet.
When given straightforward, everyday prompts that most horse people understand, Firefly shot back some pretty hilarious interpretations. The program took many of the prompts too literally and added plenty of humorous head gear throughout:
2. AI struggles with equine anatomy.
Many of the images added limbs, turned hooves in the wrong direction, or distorted equine subjects in disturbing ways:
3. AI still has a lot to learn about generating people.
My images returned some odd-looking horses, but I was surprised by how much trouble it had with people. I generated around a hundred images, and overall, there were just as many issues with the humans as with the horses. It especially struggled with fingers holding reins, feet in stirrups, and facial expressions.
Firefly seemed to have trouble portraying emotion, despite claims that Firefly Image 2 Model is better at rendering facial features than the previous version.
4. Equine action shots are difficult for AI to generate.
AI might be able to create a horse standing still, but check out these action prompts:
3. AI has trouble including multiple horses or multiple species.
When I prompted the AI model to generate images that included multiple horses, cows, or dogs, things got especially interesting:
How Many Believable Horse Images Did Firefly Generate?
You might be wondering—are these images typical results? Or, are they the cream of an otherwise-realistic crop that I’ve gathered for comedic value?
Maybe I was extra lucky, but the images I’ve included here were pretty representative of what Firefly generated overall. For each prompt I entered, I received four different interpretations. At most, only one in twenty images interpreted a prompt well enough to pass as real.
While some generative AI models may interpret these sorts of prompts better than others, the only thing that Firefly did well consistently was generate serenely smiling humans in generic riding attire sitting on or standing next to decent-looking horses. If I entered a prompt that asked for anything more specific, things got funky fast.
Next time you’re having a bad day, I recommend checking out Firefly and having a little fun! You won’t regret it.
If you have stories to share about how a real horse has made a difference in your life, please get in touch!
