missingno.eth (@canine) • Hey
Deeply interested in web 3. Aspiring to create. Love to read. Eth maxi, ENS superfan. (Pokemon ENS 👀)
Chemical engineer, biggest coffee nerd you know.
Publications
- Scientists have discovered a pill that helps end class warfare by the rich on the rest
- When lens posts start getting shared onto Twitter, it's over
- Second attempt to show this AI experiment on lens.
I ran classic paintings through Midjourney to describe them, then ran that description back through /imagine and Dall-E.
First- Mona Lisa
The result of response 1 was very solid. This painting was famous enough it should have been an easy test. Impressive.
Response 4 was way off, but had an interesting interpretation.
Dall-E had a certain look dedicated to Mona Lisa, which I would not call accurate but specific.
- No clue why only half of my posts are showing up
- Test 2
- Test 1
- In summary, depending on the database, art can be replicated very closely, but even feeding the /describe function back into /imagine works will never be perfect.
A photo is not worth 1000 words. A photo is worth so, so much more than we can describe with language. If we want to create a specific feeling, there is no other way to do so than by the hands of a master.
- Tenth- Impression, Sunrise by Monet
The quintessential impressionist painting, I totally expected this to be replicated and even improved upon.
I was surprised it got the shape exact the same in a few images. It still never got close to the feeling of the original art.
Dall-E makes some impressive impressionist works.
- Ninth-Morning in a Pine Forest by Shishkin and Savitsky
AI couldn't compete with this realism and depth of focus. It didn't recognize the artists either. Humans-1 AI-9999999
- Eighth- Les Demoiselles D'avignon by Picasso
First cubist attempt
Results were surprisingly close but missed the style by just a little bit
Dall-E redesign is remarkably simple, almost takes the image down to its essence.
- Seventh- The Great Wave Off Kanagawa by Hokusai
Two had the perfect description with the exact title, result has perfect foreground but different backgrounds
Four surprisingly had the same effect without mentioning the painting title. Each variation has comic book illustration vibes.
Dall-E killed it here, I loved every single output much better than Midjourney.
- Sixth- The Death of Socrates by Jacques-Louis David
The results here were probably the most impressive of all. The painting was obviously known, but the artist was not, which didn't really matter. The off the wall comments really stood out. You can can even see a health monitor in one of the depictions.
The final results were very human by Midjourney. In contrast, Dall-E made humans that will certainly smother you with the skin of your loved ones.
- Fifth-Broadway Boogie Woogie by Piet Mondrian
I expected this one to be pretty difficult to tell apart from the results. This time the first round was pretty disappointing while the fourth had very cool outputs. This style has minimal accuracy.
Ignore Dall-E completely if you intend to make something with straight lines or consistency. I used the description from image 3 attached in Dall-E and had amazingly different results.
- Fourth- Pandemonium by John Martin
Compared to the last few, much less famous. Obvious by the responses that Martin does not quite have a style recognized by the AI.
What's interesting with these versions is the color scheme was very accurate due to its striking visual.
Dall-E again feels like a paintbrush taken over by a computer. It seems to prefer pastels over all else.
- Third- Starry Night by Van Gogh
I'm unsure what some of the outputs mean from /describe at this point. "1688" isn't much of a description. This one was famous enough to get pretty close with version one.
Version four was completely off the mark but again made very interesting derivatives.
Dall-E feels like it's actually replicating using brush strokes.
- Second- The Persistence of Memory by Dali
Accuracy was completely lost on this one.
The first prompt had some crazy cool outputs, but man I really need to know what mushroomcore and pigeoncore are, because describe keeps plugging those terms in.
The fourth prompt was off as well, but made some of the coolest Dali arts.
I'm satisfied that the Dall-E interpretation made some very nice recreations of its namesake, accuracy be damned.
- A series on using Midjourney /describe function to recreate famous art. In each I have the original, two Midjourney descriptions fed back into the /imagine function (creating the quad images), and Dall-E recreation of the Midjourney description.
First up- the Mona Lisa
The result of response 1 was very solid. This painting was famous enough it should have been an easy test. Impressively close.
Response 4 (water descriptions) was way off, but had an interesting interpretation.
The Dall-E variations had a very specific style to them distinctive from Midjourney.
- Ok, I completed an AI recreation using some of the most famous artistic compositions in history, and tried to replicate them using Midjourney and Dall-E 2. I'll be posting the series here. For the time being, a link to the thread: https://twitter.com/missingno_eth/status/1644147420517879808?t=bdRZTLBRhH6uyxcKMyMYeA&s=19
- Is there an easy way to make a thread here?