Last week I applied to be a beta tester of a new AI art project called MindsEye (@midjourney on twitter which is slightly confusing). They ended up releasing a public link so I hopped on and started experimenting. It's a text to image program that creates some really incredible results and requires no knowledge of code, you just click a few buttons and it creates an applet interface for you!
- Correction, thanks to @kaliyuga and @speakerjohnash I realize I am in fact using mindseye and not midjourney, I thought they were the same thing, midjourney is still beta and only in a discord server ATM. I guess I guess it was crashing for me because I reached daily limits for google collab or something!
OK back to the art:
One of the prompts I gave it was : BOB ROSS JELLO
I notice that it seems to interpret prompts pretty accurately and does textures well!
I did two iterations and these are how they came out:
I prefer the one on the left as it came out quite painterly (and has extra eyes) but they are both pretty cool! The one on the right has that old school jello mold style.
I was able to play around with it for a few hours before the app started losing connection, I later found out that since they released the link publicly the servers got smashed so they are already looking to go to a pay to use model which sucks but it's understandable :) I am not sure if I will pay to use it or not, I can see a lot of applications for using these elements in my collages but we shall see!
I was able to launch the app today but it quit after 7% and I got a pretty blurry image, so it is still working but obv can't keep up with demand.
I can't get into their server or find out what's going on since the public invite link has expired, I am guessing they don't want more people in there atm.
Maybe someone else has used it too and can give some insight? @kaliyuga maybe?
To follow on twitter, personal account, Alien Community account <3
Hey Julia!
I love your work : )
I've done a few blog recently on how I approach using midjourney and other diffusion based models:
https://peakd.com/hive-158694/@speakerjohnash/head-in-the-clouds
https://peakd.com/hive-158694/@speakerjohnash/how-to-paint-a-mouth-with-ai
Awesome will check it out! I kind of got it going again but I am doing so with the expectation that it may not finish lol
What might be helpful is that I believe mindseye and midjourney are actually different. Mindseye is an interface to VQGAN and clip guided diffusion. Midjourney is on a discord server. Mindseye is a lot slower.
Ohhhh I see 🤣 I can't get in the discord server lol that explains things
*I guess I am still on the beta test list for midjourney so maybe I'll get an invite
Have you applied for the beta? I sent you the link on twitter.
Here's the first output of "bob ross jello"
I love its appreciation for the happy little trees :)
I like that one only has his hair 🤣
Heya! Yeah, speakerjohnash and I are both beta testers for Midjourney, and a lot of the things I've been posting lately are made using that. I saw what speakerjohnash said about midjourney vs mindseye being different and he's right--I'm 99% that midjourney is going to be in private beta for a while while they figure out the pricing tiers, although I have seen on the discord that they hope to offer grants and things for artists who can't afford the service.
I really hope you get into the beta--I know the list is absurdly long right now, but if you dm me some prompts you'd like to try I can run them and send them back to you?
ETA: I can also point you towards a bunch of other AI resources and I'm always thrilled to talk about pros/cons of the various methods, prompt formatting, weights, etc. And I love your Bob Ross jellos!!
Yeah I figured out why I was so confused, saw that link for mindseye and thought it was from them for some reason, then got an email the next day saying the servers were getting over run, thought it made sense that it went public, lol anyways! I actually wouldn't be opposed to paying for it if it isn't too expensive, I guess I need to decide if I would use it in my art or not or if it's worth it just to play around 😀
If you have any other low tech ones I'd love to hear about them. Do you have google collab pro? I wondered about the processing power, my comp definitely can't run any of this stuff locally...
I have three accounts running colab pro, which is 20$ cheaper than running pro+ on a single account, so definitely worth it, I'd say! Have you had a poke around my colab library? It's my pinned post, and I updated it pretty recently, so things should mostly be in-date still :)
I think I did but didn't have enough time to explore or try anything out, will def look again when I have more time!
I'd recommend starting out with Disco Diffusion; v 5.1 just launched tonight, and it supports video inits, which means you could run any of your videos through it and do some really wild style transfer or glitch stuff :)
here's the official tweet linking to the colab:
~~~ embed:1509312289442340866?s=20&t=KRyV6U1pyMwO0R82GKahQA twitter metadata:Z2FuZGFtdV9tbHx8aHR0cHM6Ly90d2l0dGVyLmNvbS9nYW5kYW11X21sL3N0YXR1cy8xNTA5MzEyMjg5NDQyMzQwODY2fA== ~~~
Bookmarked! Looks like I may need collab pro for this one
Update! Scratch that I got it running, going to have to wait a bit to see if I did it correctly!
Why you did Bob like that Julia? 🤣
I think BOB would have loved it!
Feel free to vote on my twitter poll!
The results are incredibly good. I'm gonna try it. Thanks for sharing.
good luck, it's disconnecting often because the servers are too busy but it works sometimes!
My favorite is the one on the left, it looks almost like a Picasso painting.
I was looking for a free program like this a few weeks ago. I saw some github links I might try when I really have some time. But there doesn't seem to be anything point and click that's free and generates without severe limits 😢
If you find something cool let me know, I don't want to have to do too much tech stuff 😅
I will :) I know about one without too many limits but it's not text to image, just playing with images. It's called Artbreeder. It can be addictive, so be careful, lol
Oh yeah i know that one well, I integrate artbreeder stuff into a lot of my collages!
Hello @juliakponsford. Looks appetizing and quite surreal. Didn't know such things existed. Probably such a generator is a fascinating thing, I hope the prices of such a democratic. I heard about some similar image generator from my friend a couple of years ago, but I don’t remember what it was called :( I remember that he created quite psychedelic interesting things there, maybe you know what this generator is called!?
could have maybe been deep dream? that's been around for a while, you feed it an image and can choose a preset style or add your own image style:
https://deepdreamgenerator.com/
if you want to pay for it in order to get high res images, then there is this https://neuralstyle.art/ - but if you are familiar with Deep Dream and have already a good collection of styles (of your own) on the free plan, you could just switch to a paid plan on Deep Dream. It is by far the longest established, and I've been on it since the very beginning, when all there was was the Deep Style. They came a long way since then.
Hopefully you read my recent post about AI (you did upvote it).
This cannot compare what you can do with https://app.wombo.art/
and here is a hint: on the PC you only get the word prompt, but on the phone you can upload also one of your own images to blend with the style.
But even that is amateur stuff - to really work with AI, I blend it also with Deep Dream and then muck around in Photoshop. So do check my post: APRIL FOOL - a AI Orgy.
BTW, I also tried Night Cafe, but it is by far not as good. Bottom line would be that no AI can come close to being as innovative as a "real" artist.
Yes totally agree that nothing beats the human touch always! I have been playing around with disco diffusion today and you can also load an initial image so I am looking forward to experimenting with what it can put out using my own creations. I am just trying to get get the basics down first to make sure I understand what's possible, it's a pretty fascinating world :D I have wombo and have never found the image upload function??? maybe i need to update my app, I got it confused and thought that was available on the PC version, interesting!
Bingo I had to update the app to use the image upload feature 👌 I only wish the images were larger but it does a damn good job...
Wombo: the style image you can upload has 3 settings, shallow, middle (default) and strong. You have to experiment what works best. I found it best if the image has some strong contrasts and simple color fields, so in future I might make some sketches just for that purpose. For size of the result, I take it to 150dpi in Photoshop and then play with sharpening and contrast. As far as that goes (enlarging), the standard Deep Dream style is easiest to enlarge and still get a good image. Thus taking some Wombo images to Deep Dream gives it a extra dimension. I had also mirrored the results (to double the size) and then cloned in some other stuff (like a collage). The possibilities are endless.