Archive for December 13th, 2022

  |  
2022/12/13

Hemingway, by Midjourney

I now have like 190 images in the folder that Windows uses to pick desktop backgrounds from; building on the twenty-odd that I posted here the other day. They are fun! But I’m not going to post any more right now; right now, I’m going to post some images comparing the various Midjourney engines (which they have generously kept all of online). I’m going to use the prompt “Hemingway’s Paris in the rain”, because why not! We can do other prompts some other time.

For most of these (all but “test” and “testp” I think), it produced four images, and I chose one to make bigger. Otherwise (except as noted) these are all just one-shots on that prompt. I’m going to paste them in more or less full-size, and let WordPress do what it will. Click on an image might or might not bring up a larger version or something who knows.

Here is the quite historical v1:

A rather vague but definitely rainy image of Hemingway's Paris in the rain. There is a tall black tower to the left that may be inspired by the Eiffel Tower, but resembles it only vaguely.

Here, similarly, is v2:

Another vague and rainy, perhaps slightly less streaky, image of Hemingway's Paris in the rain. A possible bit of Eiffel Tower inspired tower shows over the buildings to the right.

I rather like both of these; they are impressionistic, which I like, and I suspect it’s mostly because that’s the best they can do in rendering things.

Here is “hd”, which may be the same thing as v1 or v2 I’m not sure; this particular image is more strongly monochrome and sort of vintage-looking photo-wise:

A somewhat blurry and rainy of an old city square with some people in it, some with umbrellas. Could be Hemingway's Paris; no towers evident.

Now v3, which is pretty much when I started using Midjourney; it’s interesting how impressionistic this is, given that we know v3 can also do rather more realistic stuff (all of this, for instance, was v3):

A rather impressionistic drawing, perhaps in charcoal, with a somewhat Eiffelish tower to the left. Definitely rain, likely Paris.

Between v3 and v4, we had this engine, lyrically named “test” (I used the additional “creative” flag, because why wouldn’t one?); one is getting a bit more photographic here:

A slightly less vague still image of Paris in the rain, black and white, umbrellas, and so on.

and here is the “testp” variant of “test”; the “p” is supposed to be for “photographic”; I used the “creative” flag here also. It’s not notably more photographic than “test” in this case; maybe it’s the rain:

Another rainy city street, monochrome, a few cars, shiny impressionistic pavement, townhouses.

Now brace yourself a bit :) because here is the first version of v4 (technically in terms of the current switches it’s “v 4” and “style 4a”):

A soft-edge realistic painting of a Paris street in the rain, in muted but glowing colors. A few people walking in the distance are vague but convincing shapes. The Eiffel Tower is visible in the distance.

Yeah, that’s quite a difference. We have colors, we have lanterns casting light, we have very definite chairs and awnings and things. But now, the current v4 (“style 4b” which is I think currently the v4 default):

A rather realistic painting of vintage Paris in the rain; a couple of old-style cards on the street, their headlights and the lights of the shops reflecting in the wet pavement. Shopfronts and awnings, people in identifiable clothing. There are words on a couple of the shopfronts, but they are unintelligible: something like PHASESILN for instance.

Yeah, that’s gotten rather realistic, hasn’t it? It’s even trying to spell out the signs on shopfronts, even if it hasn’t really mastered language. But those cars are extremely car-like and detailed compared to anything earlier.

Can this currently-fanciest engine give us something a bit more like the atmosphere of the older ones, if we want that? Basically yes, if we ask for it. Here is the latest v4 again, with “impressionistic” added to the prompt:

Yet another wet rainy city street scene, again in full convincing muted color, but more impressionistic than the last. Again we have people (and hats) and umbrellas and shopfronts, but no attempt at individual letters on signs.

I rather like that! And “monochrome” would make it monochrome, and so on.

It’s perhaps interesting that the more recent engines were less insistent that pictures of Paris include the Eiffel Tower. Possibly just the random number generator, given how tiny our sample is here, but possibly significant in some way.

So there we are, nine probably rather enormous pictures of Hemingway’s Paris in the rain, as conceived by various stages of development of the Midjourney AI, and with only very minimal human fiddling around (picking the prompt and the one to feature from each set of four, having the idea to compare the versions in the first place, and like that) by me.

Comments welcome as always, or just enjoy the bits. :)

Posted in ai, weather | Leave a Comment »

Tags: , , , ,
  |