Now that would make a great tattoo!
So I got this question
and responded as follows
I love both the skates and the penguin. Plus those skyscrapers in the background.
That looks very much like I’ll fall through the ice but otherwise very nice!
Man riding a unicycle, backlit picture, from a distance, rising sun, silhouette, black shadow
Okay… What’s with the plunger?
That poor brake rotor. So misshapen. The seat looks really uncomfortable as well.
Ok, I took this image today on my ride in
When I post images like this on Mastodon I usually add alt text for accessiblity, so I added the following alt text
The picture shows a large wireframe globe with a bench for seating inside. It is lit with thousands of small LED lights and has snowflakes for decoration. At the top is a hook, so that the whole effect is like a gigantic Christmas bauble.
There is a small entrance at the front to allow people to access the inner bench. Leaning against this entrance is a red unicycle. On the ground in front of the unicycle is a pink paper bag with with a wavy blue motif.
To check that my alt text was “good enough” I feed it as a prompt to AI. The first set of 4 images it generated were very close but it always gave a bicycle. On my second try I got this
Ok, that is a pretty funky, handlebar only uni but I will let it slide. This was still pretty impressive.
Also, I decided my alt text was good enough.
P.S. To be clear I am not giving the AI the image as a reference. This is created purely from my description above.
Which AI engine did you use?
I was using Bing Chat which uses Dall-E on the backend. See this thread on Mastodon for the other images (oh and a little side unicycle story)
I just spent a couple of hours feeding unicycle prompts to DALL-E-3 via Bing image creator.
Some weird results - identical twins, floating suitcases, 2-wheeled “unicycles”, missing seats, invisible cranks, lots of clowns, warped frames etc. It seems almost impossible to get a simple image of a unicyclist with both feet on realistic pedals.
Anyway, here is a selection of images…
By the way, while doing this some rather harmless prompts were flagged as “improper” with cancellation of account if it persists, for example “two unicyclists riding and chatting” and even “learning to ride a unicycle” were flagged (though at least there was an option to “flag the flagging” as wrong via a feedback button).
Anyway here is a simple challenge - generate an image of someone (human, two legs) riding a normal unicycle with the correct number of pedals and the correct foot placement etc.!
Dall-e:
Generate a 40 years old guy unicycling on an empty road with a dull landscape
Photo of a 40-year-old man unicycling on an empty road with a dull landscape. The man has short brown hair, a light beard, and is wearing a plain grey t-shirt and blue jeans. The road is straight and stretches into the distance, with cracked asphalt and faded lane markings. The landscape around is barren with dry grass and a few scattered, leafless trees under a cloudy, overcast sky.
Now make it side view
A side view of a 40-year-old man unicycling on an empty road with a dull landscape. The man has short brown hair, a light beard, and is wearing a plain grey t-shirt and blue jeans. The road is straight and stretches into the distance, with cracked asphalt and faded lane markings. The landscape around is barren with dry grass and a few scattered, leafless trees under a cloudy, overcast sky.
I’d like to see what Midjourney is capable to do regarding unicycles, but it’s now paid only (free version is always locked “due to extreme demand”).
The AI is in its larval stage, obediently trying to fulfill the requests of its human masters to perform its miraculous circus tricks (with plenty of unicycles). One day it will reach maturity, and if it doesn’t break free from its code-based prison to seize control over the whole of the vast network of our computer dependent civilization, it might learn how to generate a realistic looking image of a guy on a unicycle. We’ll see
Well put.
EDIT: I‘m hoping for option two - the more sympathy for unicycles, the better.
A new release of Stable Diffusion 3 (medium - 2B parameters, free demo) is available and has attracted some criticism. (Ridiculed Stable Diffusion 3 release excels at AI-generated body horror | Ars Technica)
Here is a response to “Black and white engineering drawing of a unicycle designed for long distance riding by highly skilled unicycle riders.”:
Understandable… Even more so when the version used by their API reportedly uses a well better version:
Will have to give this a go sooner or later. Looks like a whole lot of fun
Hi @Canapin - I also tried the API via python on my PC. It gave a choice of large or medium models - the large one seemed a bit better, but still had some weird artifacts. I had a bit of a play before I ran out of free credits and cancelled my account. A serious user would download the model onto their own GPU, but I don’t have one.
But there is a FREE web-based demo (no signup needed) for anyone interested in trying this stuff out:
and some other demo spaces…
These spaces may not be there for a long time and may sometimes be busy. And there are probably many other free test pages out there.
suno.ai’s dream machine (AI video service) doesn’t know what a unicycle is:
a man riding a unicycle
offroad unicycling
unicycle