Sooooo..... Very interesting experiment. I am interested in knowing if anyone else tried the AI contest or have done AI image work at all. I downloaded the free software StarryAI and gave it a shot. I don't think it will be for me but it was kind of fun to play with.
There's three ways you can make an image. I'll give you my thoughts here but I'm very interested in other people's because this was just me goofing around making some stuff up. All methods begin with giving the software a prompt or a few words or a phrase that it will use to form the subject of the picture.
1) after giving it the prompt you can just say create image and see what you get. My outcomes were bizzaar as I'll show below.
2) you can give it the prompt, and then select other artists or styles you like and it will use them to inform its design of your prompt. You can pick multiple artists or just one. This was kind of fun but often you pick artists that don't really go together well, but it's still what I enjoy doing most.
3) The third method is to upload your own image, then give it the prompt and you can even still pick artists but it will develop an image based on your image.
The problem with all of this is that the AI program doesn't have any theory or I don't really know what it knows. It just kind of makes some crazy images and then you don't really have the ability to change them all that much. Even though there's an 'evolve' function which allows you to make some more suggestions, you can't make specific changes. I found the process very frustrating because it's like you're telling another artist a brief phrase and then they do something and give it to you and it's supposed to represent your work and it totally doesn't. I don't think this is something I would really get into I'm. I'm just not techy enough.
One thing I will say if you go and look at the things in the contest, is there are some freaking amazing amazing amazing images in there which I have no idea how they did but they all are very sterile. They look computer-generated to me as there's no human input of emotion in the images. I found that to be something throughout all of the images I saw when I was investigating the AI system. Any thoughts about that?
I put some images below just to show the kind of stuff that it makes.
Again, if anybody's done this before or downloads the software which is free and gives it a try, I think it would be interesting for our experimental side.
The prompts for the images below (in order) are:
Bridge from nature to city at sunset
City in decay
Children in the forest at night
Green eyes crying
Bridge night moon
Ship in a storm
Self portrait (so this is Mr A?I 😄)
Ship in a storm
Green eyes crying
She is sad
I just got into this very thing last Thursday and have lost some sleep over it. haha. I use MidJourney and it seems to produce some quality images... Without proper structure the results can still feel "computer-blobby" if that is a word?
It also works on using your own photos, then AI builds new scenes or portraits. Also supports motion and motion blur... So Im also including a nice rendering of a lady dancing with a slow shutter.
I did enter the AI contest but I have better work since "those days", what was that four days ago? hah! But a lot of fun. I'm hooked, what can I say. It would be interesting to have a computer AI group; this stuff is not going away, ever. MidJourney is having trouble keeping the servers up there is so much demand and users submitting prompts.
Apparently there are different groups you are automatically added to; and I was just added to the 2500 image group. 🙃😁
... and because MidJourney is technically in beta, they request we add dates when the images were rendered. So Left to right: Aug 9th, Aug 6th, Aug 8th.
They also have a nice gallery site that automatically loads up all the images you rendered... Fun part is seeing everyone else's work (real time and in the gallery).
On the subject of being sterile, I completely understand what you are saying. There are tons of options for emotions, feelings, motion like dancing... but people do not use them much (yet).
Kind of like how the emotionless-blank-stares selfies was big for a few years? Same kind of thing. People will get more creative over time with their keywords. 🙂
..... and welcome back Ruth! 😁
That was fun! But probably nothing I will get overly excited and fling monkey droppings over.
Here are a couple before and after.
I think this process is bizzaar. It makes me feel old! I actually told my husband that this would be something "young people would love". How hilarious is that!?! :)
This are fun images! The second one is cool.
Thanks for chiming in too.
Hi Ruth! Always interesting to read your thoughtful remarks & observations. I agree with your words about sterility, and I often find an unappealing flatness in AI imagery. It calls to mind fantasy comics. All this may merely reflect the beta state of AI as a whole at the moment.
I can't muster enough interest to pursue this personally. I am slightly curious, but wonder what the aim of the exercise is. Like robotics, it can be amazing, but the appeal soon pales unless it serves some useful function. To me (and I know this is an idiosyncratic position, and don't mean to denigrate anyone who feels otherwise), it's an idea in search of a raison d'être, if we're talking about art and creativity.
Having said all that, your seventh ("self-portrait") image does NOT have that sterility or flatness so much, if at all, and for me is without doubt the outstanding image of the contest. Well done to you both!
I never heard of this, Ruth! Thank you for sharing. I look forward to giving it a try.
For FYI and thought provoking: Here are a bunch of images all created with the prompt "sailboat on the ocean at sunset". After doing this for each available style (there are 20 - I spared us all of them here) I could see what I liked and then make combinations of styles which helped get a final image that was pleasing to me. The images here on this post are the two combination outcomes that I liked. In the reply are a few examples of the individual style examples. All just for info and curiosity.
Single style example
More.
It can get addicting Ruth, be careful!
So true! I don't think it will last though. I am enjoying fooling around with it but am not feeling inspired. I think I will break out the tank and do some ink work. Now THAT is addicting!
Very true! It's a nice escape as it's just too hot for me to get out and get into the forest for wildlife... Soon enough it will be Autumn and Mushroom season! Can't wait. 🙂📷🍄
Did anybody submit an AI for the Contest?
I know you, Joe and I did. Both of you have top images in my mind!
I have looked since I posted on the final day. Time to take a peak.
Doesn't make sense.
Voting closed prior to Submission closing; therefore, you can Submit and have 0 votes.
The contests are always weird. I only do them for my own desire to try new things. I even go through and vote everyone a 5 just to counter balance the nasties who vote everyone a one. I realize that means more pointless voting but at least it is on the side of good!
Anyway - even if they don't have the video ready, I hope they post a new topic soon. Apparently I'm a glutton for punishment!
PS - I wish I could have put this in. I think she actually seems sad. I was more interested in showing emotion than beauty.
Late to the party as I needed a bit of background before commenting. I have been finding it difficult to see where AI fits into a photography category, but as FStoppers have taken it on board then I feel they have set the tone.
I do see AI being accepted as a creative tool that allows creative enhancement of uploaded images (of the photographer's creation), but I still struggle to see how images creative purely from text prompts fit in.
How do others feel? As a group I think it would be good to experiment with this new medium and learn from one another. I think I would personally like to see a description /detail attached to posts (eg original image if applicable, prompts used etc).
I suspect that there will be a MidJourney/AI group appearing soon on FS (if not already) as this seems to be gathering traction
Congrats to Joe BTW for having his work reviewed in the FS critique.
Just as an experiment myself, this is the result of a quick prompt (Geisha_girl_beatiful_dance_swirl_realistic_painterlly) which produced surprisingly good results (note..... no photography needed.....)
Thank you Alan! Big update to MidJourney last evening with the activation of --beta at the end of the prompt. The results are very impressive to say the least.
Thanks Joe. I see that MJ is limited to using an image URL for inspiration. It can't build on top of any image due to concerns over public content.
Ive found it dose pretty good with images as references, even two images. You have to weight them fairly heavy though. An image with a "--iw 10000" weight is almost identical to the reference image. while an image weight of < 100 will have slight references to the provided image. The new model was very good but was only for 24hrs, hopefully it will be back shortly.
That's odd - the documentation indicates a default iw of .25, with examples stopping at --iw 5.
I'll have to try again
Its a little confusing, there are "weights" and there are "image weights".. You can weight any prompt.. all the numbers are added up then divided as a percentage. Images have their own weight, a tag like --iw which has it's own ranges and set numbers. Kind of like how --style works. I'll find my reference here in a sec...
Where are you getting your information from Joe? I can't see the --beta parameter listed.
I did try the Beta Upscale Redo and that has a significant impact on portraits at least (fixes some of the obvious facial flaws).
I don't see any MidJourney (or Dall-E/AI) groups available on FS - as you are far ahead of others perhaps you may wish to start one so we can all trade ideas/work and learn from one another?
apparently --beta was for only 24hrs. --upbeta (or beta redo) upscales the image sufficiently but at a cost.. usually details. Not great for portraits but good for landscapes (as noted in the announcements).
https://github.com/willwulfken/MidJourney-Styles-and-Keywords-Reference
Scroll down to Styles... Then when in an area scroll down again, and right before the categories click on either Landscapes or Just the Style... the default view is Spheres and... well.. they just do not help me. (so use Landscape or Just Style).
Once that is selected go into a category and it will show what all the prompts/keywords do. From just looking at the Discord Channels and the art people are producing, they are barely touching any of the keywords/Prompts available.
Things seem to be changing almost daily. hah. I know I saw the weights explained (like above) but can't find it now. Maybe I'm just going insane. :)
Now it's 1 through 5
For fixing faces check out this video, it works wonderfully. Though --beta has fantastic faces but it's not available for the moment.
https://www.youtube.com/watch?v=RXWpz7UBYM4