Hope this isn't considered off topic too much I used the Emery prompt "Japanese maple with twisted stems and weeping habit in autumn colors with sun in background" and added different artist or style Watercolor Photorealistic Style of A Warhol Style of C Monet Style of Seurat It is fascinating, and impressive and scary too
@AlainK You type in a prompt and the AI generates an series of 4 images For the water color I prompt was "Water color painting of Japanese maple with twisted stems and weeping habit in autumn colors with sun in background" For the others I used something else for the bold text above and the rest remained the same
I think it's a fascinating piece of technology. I called it "art" but I'm far from sure I'd really consider it as such. On the other hand, if you showed me the water color, and said "my cousin painted it" I'm sure I'd think "very nice", without perhaps feeling particularly moved by it. I do intend to play more with it, but of course time is limited. I do find it interesting that one is permitted to use the results commercially. It might be pretty good at coming up with logos etc, though I haven't tried that yet. I did make another pass with the phrase: "gnarled maple with exposed roots and varied fall colors on a rock outcropping, surrounded by sheer gorges and sparse pines and fir". My thought was to try and come up with something bonsai-like, but it seems to want, not unreasonably, to put the pines on the rock. I tried to modify this to be more exact, but it made the "gnarled tree" more clearly a pine: Finally I made the grammar better (clearer subject) and was able to get it to produce the maple on the rock. But, I suppose because it couldn't find many pictures of craggy maples with exposed roots, these are really quite poor compared to the others. "a gnarled maple with partially exposed roots and varied fall colors clinging to a rock outcropping, surrounded by sheer gorges with sparse pines, fir and other maples growing" I think this probably gives some insight into how the algorithm works: it trawls the net for images that might correspond to the request, probably using some existing image search technology as a start. If there are a lot of images (like golden retrievers, or weeping habit Japanese maples) it makes a good go of it. If the subject is narrower, perhaps not so much? More time-wasting "research" needed, lol!
“photorealistic image of a maple bonsai with a broken pot and a man wearing a suit & top hat who is crying while looking at the bonsai.” For @AlainK ;)
Something between Magritte, and Dali, but ... That one is not a "collage", like the ones in the 1930s : William Vallavanis is great, but this Ulmus penjing made him look small. I hope he won't mind my posting this photo here : In Vietnam too, there are tiny gardeners taking care of great bonsai trees :-) Sometimes, a figure to enhance the beauty of the tree, tell a story. Chinese "penjing".
Over coffee this morning I tried uploading a picture of A. circinatum, I wanted to see if the algorithm is capable of aging the tree and putting it in a different location. It isn't, this picture upload is more of a compositing operation, as in you can drop a picture of yourself into a different background, I guess. I then wanted to see if it would be able to find specific species, and it seems to confirm that the initial training set is based on an image trawl. I began with A. cappadocicum, not rare, but not omnipresent either: "Acer cappadocicum growing in summer an a mountainous Caucasian landscape" Not too bad unless you consider it got the species wrong, it appears to be A. buergerianum. In the 4 images, the other 3 were (maybe) negundo and a couple of scrub oaks. Then I tried a more commonly pictured maple, which provides a lot of images to train the algorithm: "Acer japonicum in autumn with a mountainous Hokaido landscape". Including the typo in the name "Hokkaido" which I didn't notice until just now. These are quite passable I think, and the maple is recognizable, which is a start. I wonder if the model continues to train, based on which images users download; those might be presumed to be largely "successes." P.S. that Ulmus bonsai is insane!!
Yeah... Chinese penjing are usually much bigger than Japanese bonsai, but this one is a record breaker.
"A 700 year old Japanese maple in autumn" Mostly they look less than 200 years old to me, but not too shabby for a baby Skynet!
@maf I guess I didn't aim for old enough: I tried 200 year old, but they looked maybe 20, lol. Clearly size, age extrapolations are not well developed. We may be a few steps short of Skynet, yet. Anyway I thought it would be interesting to see how it produced maples in an artificial or urban setting. "a 10 meter tall upright Japanese Maple planted in a modern plate glass atrium, surrounded by dwarf conifers and flowering azaleas, with a flat water feature to the side, in summer, lit with shafts of filtered sunlight coming through the glass from the left at 35 degrees, surrounded by a modern city which can be seen reflected in the plate glass" Some of the reflections are interesting: how about the second image, the reflection in the water along the bottom edge, on the right hand side; what the heck is that, a window washer on the building behind the eye point?
It is amazing how well it does - the algorithm seems to drop at least one or two of the parameters in each picture but from such a long list it is inpressive how much it retains and still manages to look realistic!
LOL, I'm not sure how to make that one, I give up! Made a couple of tries but things got abstract quickly: japanese maple exploding from a bowling ball, in a bowl which floats in a sea surrounded by clouds I fear our young AI is already taking nuclear inspiration... Let's hope not a sign of the times.
I am impressed emery Here is the prompt " “a 500 year old Japanese Maple planted a pot that looks like half of a coconut, surrounded by a translucent bubble, Floating on the surface of a neon yellow semi-transparent lake with volcanos in the background that are erupting” I wanted a maple floating in a bubble but that wasn't the correct way to get one
