Can AI Art Create Beautiful Typography and Fonts
Вставка
- Опубліковано 28 вер 2024
- I wanted to use AI Art to create some fonts and typography for my website. AI Art Text isn't something that normally works. AI Fonts often fail because ai typography just doesn't seem to produce ai text well. Unlike ChatGPT, Midjourney art is not a text model. The images it creates dont see the difference between letters.
I attempted several strategies to create proper letters, and finally figured it out! A combination of ai image blend, as well as text prompts with weights, is able to produce the desired ai art text!
#ai #art #website
This channel also has the backing of Editor X:
www.editorx.co...
You can find our Community Discord below:
www.mergewebde...
This video is fascinating! Your experimentation with prompting AI to generate the letter T and achieving depth and styling is impressive. Boost App Social has been a valuable tool for me in creating visually appealing typography and fonts. Its unique library of fonts and AI-powered features make the process so much easier and more creative.
Boost App Social seems like an intriguing solution! How does it specifically contribute to creating beautiful typography and fonts using AI?
Just watching the first five minutes of you trying to get Midjourney to give you a stylized letter "T". Of the four beautiful suggestions, I feel you should have selected one, upscaled it and then used the dislike option to let Midjourney know how its doing. Nobody seems to use Midjourney right. Its an artist that happily creates images all day long, yet nobody bothers to tell it how it is doing. Create this, create that, then people get frustrated when it doesn't give you the right thing. it's not a mind-reading device. If it gets something right, please take a moment to rate the images and I can assure you that it will give you what you really want.
How does one 'rate' the progress? Negative prompts?
@@bobparsonsartist564 once the images are individually upscaled, the option to rate an image becomes available. Love it, like it, loathe it, makes you want to vomit. Its all about feedback and from one artist to another, you'll agree that feedback from the client is important and helps to shape a work of art. The rating system allows Midjourney and especially the bot you are assigned to, to learn what is correct about your prompt. I would've rated the letter "T's" in the video and run it through 10 - 20 times. By the end of that you would have perfect "T's" with all the foliage. I created my XNY picture using exactly the same process on Midjourney. I've rated around 15,000 images so far and I've received around 50 - 60 hours of fast time in the process.
I think here comes ControlNet in Stable Diffusion into play. It gives you much more controls over the Process.
It's easy and this is how I do my characters. So you're going to ask it to generate that by putting quotation marks around the letter or that you want. Then you're going to like the one that you like and maybe it's not even what you want but it's closed what you want so upscale that and then you're going to either heart but I use the thumbs up sign to get closer to what I want. Then you're going to right click that picture and DM the results. This is going to send a message to the bot.. n
grab the seed..
Repost with prompt add styles ect ar --seed and enter do this about 20 times
However and here's a big however never ever ever ever ever ever ever ever ever ever reroll re-rolling after doing this is literally like taking a big eraser sometimes it will remove all that work you did trust me you know how I know because I did it
I added a heavy weight to the letter and it generates the letter for me each time. Prompt= a stylized letter S::9 At least you'll have the letter and can then play with the style from there.
How to get this interface? Throught chat gpt or jasper or dall e 2
ill suggest this prompt... try like this.. you'll be get better results by this
/imagine prompt:Create a jungle-themed typography design for the letter "U". Keywords: lush, exotic, tropical, wild, adventurous. Colors: green, yellow, brown. Style: bold and organic. Typeface: handwritten or serif.
I would have used text weights to emphasise what was important in the prompts (7) etc to let it know how important the “T” is. Also, if you had got the seed number for the p you liked (by adding the envelope ✉️ emoji to the images it gave you), then you could have tried using the weighted T with that seed number. Final thought is have you ever tried asking it to produce you an alphabet? Again, adding the seed number from the previous preferred P would help
I think the problem you're facing here is that you're trying to make Midjourney approach this as typography, because YOU are thinking of it from the perspective of typography. Inspired by this video, I typed this prompt and immediately got 4 very clear letter "S"s - "/imagine prompt: calligraphy letter S, German Gothic Fraktur style, vector letter S logo, white background"
Approaching it as a "logo" in Midjourney gives much more reliable results with a single letter, and then YOU can approach/use it as typography afterwards. Sorry you had to go through so much of a hassle to get it to spit out something you liked; that original letter P you settled on was gorgeous!
I found that I had some success with the prompt "T is tree," it seemed that add the "is for" gave better results than trying to force it. Not 100% but some success
Oh i might haev to try that!
It is almost as though mid journey has programmed this feature on purpose so as not to have all the graphic designers going crazy, and complaining that their jobs will be lost through this program.
It's likely because they didn't train it off of a bunch of typography. Or images that were labeled with the letters in it. Or single letters
It’s almost as if midjourney uses art without consent...
This wasn't too hard a task in Stable Diffusion with the new ControlNet models. But even when I feed those results into MidJourney to show it exactly what I want, the letters morph into some kind of weird Cyrillic-looking letter. If MJ ever implements something like ControlNet, people's minds are going to be blown.
What needs to be done is new models need to be trained specifically on typography, with deliberately constructed data sets, rather than web scraping. It would be an expensive project. I can't do it on my puny nvidia 1060GTX card.
SD & midjourney are the way they are because a bulk of what they are trained on is portraiture and fine art.
We need a wide range of models trained more specifically on graphic arts, or architecture, or vehicles or whatever is currently "weak" in SD & Midjourney.
But control net can be utilized to get things closer to any intended result, including typography.
I have been using control net to get things to conform to a 3D cube rendered in blender from all angles as a way to get greater control over camera angle.
Play with the values, and you can tease it to be close to but not completely the input image in control net.
I love control net, it is a game changer.
Midjouney, I don't do Discord, and I don't like the idea of paying to generate per image. So, I'm not into Midjourney. You can have it.
A1111/SD all the way here!
But still, we need a more diverse set of models trained on different things. Too many portraits right now. Art is so much more than portraits. Glad some are exploring the typography capabilities of this stuff.
I was pleased when you finally started spelling beautiful properly.
This illustrates well the frustration of writing a "prompt" when there is no available documentation on how to choose its syntax, notation, grouping and keywords. AI does great things, but simple things are still difficult for it.
This is more a MJ problem than an AI art problem. Because it has no direct way to compose a scene or direct the AI its very hard to be specific. In fact the more specific you want to be the worse MJ is as a tool. Its amazing at producing concepts and themes with distinct styles, but it sucks at specific subjects within those frameworks. Nobody outside of the MJ knows what sauce they add on in the background but its suspected it heavily adds stylisation to whatever your prompt is to guarantee artistic good looking results. This messes with targeted prompts. Where as Stable Diffusion has multiple ways to force subjects and direct scenes now, a simple Img2Img could work here, as could controlNet (which allows full posing of subjects and would work for a letter like the video) SD though requires much more work and knowledge to get good looking artistic results on a par with MJ (though it can be done)
Indeed, typography in general was one type of prompting that was missing, sure there are lots of cyberpunk prompts, but what about things in web design!
I think this year we will see lots of amazing AI tools that will make our lives easier. Those who can use these AI tools smartly will be the gainers. Recently I created amazing images with Blue Willow
The free tools I have used up to now, Blue Willow seems to be the best. The outcome is fantastic and art is more realistic to me. The great news is now anyone can add BW bot in their discord server
Very insightful. Could You have reuploaded the final T to expand the gradient background to add text either to the left or the right? It was a really good looking background to use in a full section.
If there is any image that you like you just save it to your hard drive and then re-upload it and then paste the link
What about defining the typeface like _Garamond,_ _Caslon_ or classifications like _Grotesque_ or _Antiqua_ or even _with_ or _without serifs._ Would that not add needed context?
Thats a good idea!
I would have transformed the P typo in photoshop to a letter that looks like a T then put that in midjourney. Yeah the tranformation wouldnt look great but hopefully mid journey could have make it good.
This video Is the most helpful explanation I've seen so far. Thanks!✨✨
Glad it was helpful!
I used “Calligraphy” in prompts, can get better results, but probably not suitable for generating modern style typography
I was worried I would get curvy letters!
Why not use image to image in Stable Diffusion? Canny use of denoising will eventually give you what you are after. Midjourney is great, but you have limited control over crafting your images compared to Stable Diffusion and its many extensions.
And you can run it locally, for free.
Not sure if I missed it in the video but did you try asking to give you the complete alphabet instead?
This is great work, but it shouldn't be this difficult to pull a stinking letter. Hoping the next version of MJ would be addressing this.
A designer would've done it in under 5 minutes.
Have you tried describing what the letter t looks like?
Question; what if you try midjourney to create a list of letters in a single sheet?
You need to list and tell it to exclude A,B,C,D,.. except the letter T. That's what I had to do in another software.
I'm wondering if you could use playgroundai and play with the sliders to get what you're looking for :)
Could you ask it to do each letter in the alphabet in the same image?
can you use it for a word or a phrase?
Adobe firefly has made this video redundant!
Dude when it confuses T with P, there's a high chance it confuses some other letter with T. Maybe try that too?
"H8TCH!"...lol
Never heard “H”pronounced that way. Interesting.
Did you try "T"?
Thank you , think I am not as good like you I need the Word but I didnt get it right
How does Midjourney treat prompt typos?
you should tell us how to write a prompt for weighting particular parts. you spend the video showing all your mistakes and then speed past the solution with no example... why????? what do you write to prioritize/ weight aspects of the prompt? Im gonna go search for the answer, but would be nice if it was here, since the video is about this... no?
«Б» - очень красивое получилось 👏
Might be easier using Playground AI, Midjourney is so fiddly
Try this prompt /imagine the 20th letter of the alphabet T with a floral design --v 6.0 --s 250
wow so much learn i want build is network like you sold before to client can you help me on this project ? of course learm from the best
Lol😂 TP would be a good substitute for the 💩emoji (but the 💩 has a smile so idk)
this video is really good ！
Your imagine prompt needed weights
Good idea, I only recently learned about weights, I should of tried that!
Can dalle 2 design like this
Maybe try the word ‘shape’ instead of the word Letter 🇦🇺❤️
Adobe firefly is the perfect tool for this.
Still in beta.
Starts at 12:15 you are welcome
AI needs to "borrow" more text graphics into the system, maybe people should start making them to feed the AI and pay the AI instead of the creator which inspired the prompts.
I would've manually removed the background in Photoshop.
You just need to shout at the bot loudly. Threaten to turn off the Internet. It should then oblige and give you the letter T.
Oh that could of worked! I’ve seen people do this for chatgpt haha
I lost the will to live.
It appears midjourney ai suffers from "Letter T" phobia, you made it nervous, really nervous
😆😆😆😆😆
Honestly, by the time you get "sort of" what you want, a good designer would've been done with the work.
See this kid, this is what happens when you don't read the manual on the website
you can create anything by midjourney but you have to learn at first how to command.
So basically, people are going to flood the internet with a.i junk for 36 days of type 😬
Ridiculous. Isn't it easier to create a pattern and stamp it out with the font that you want? Not very I in this case of AI.
How complicated...
pick a different letter and maybe it will give you a T? hhaha
Ha wow just amazing how ‘Intelligent’ AI MidJourney is and can never understand simple letters and text
This is the problem with AI art in general, it’s not original. It needs to copy data/pixels and if there are only a few relevant examples, it can’t “mix and create” another version that is supposedly “original”. If all the data AI has is three images, it can never go beyond creating anything beyond the mix of the three images. AI can improve life, imagine immediate access to medical care and information that may be superior than what one human doctor is capable of providing. But AI can also destroy, steal and manipulate.
I’ve yet to see an artist in this century (or the one before it) do something that I’ve never seen another artist do in some form. Everyone takes inspiration from various sources and blends it all together to make something new. It’s incredibly rare to see something totally and completely new that no one ever imagined before.
Seems just like human nature!
I would’ve been livid 🫠
I have over 14 years photoshopping experience so I would have saved myself a lot of frustration just in putting a bunch of different T in the style that I'm going for.
Or I would have a bunch of different pictures of the foliage different letters and I would have blended them
. There are plenty of different options
There are so many different tips and tricks I would have turned my prompt into a prefer option so that way I could do it just over and over simple if it's me and this is what I do if I had this time if not then I'll probably just blend it and play with it. Let's just say that I don't have experience for Photoshop what I would have done is give it what I'm looking for I probably would have uploaded a photograph I'm sort of what I wanted but stay I don't know how to do that okay so here's what I would do I would start with my prompt and I will make sure that I described it exactly how I wanted it and in my prompt I would put the name..letters whatever I would go " Renee" or [Renee] I would add a weight onto it so it was actually more weight on that then it would the scrollwork behind it remember for your subject to have more weight it has to have a higher numbers than anywhere else in your prompt then I would reroll this until I had something close to what I wanted. Even if it's a little off maybe I just like the scrollwork behind it or something about that image this is my starting spot. So then I'm going to upgrade that one picture for maybe three of those pictures. Then I'm going to like the one that I like the most. I'm going to set up prefer option to make it actually easier for me to just it's a shortcut the name Renee and I'm going to make the value whatever I say the prompt is like Renee and red with swirls behind it and roses. It doesn't matter because then I'm going to right click my image and I'm going to DM the results this will send a private message to the bot..
Then I'm going to grab my seed. Then I'm going to go back to my mid Journey server and I'm going to paste my exact prompt. Which like I said before I'd already program so all I have to do is whatever code I put in for it which would be like dash --renee . And then I have my style my aspect ratio and at the end I add that seed that I copied before. And I'll keep doing this about 20 times until I get the one that I like. And that doesn't matter even if I get all four pictures and I hate one I'm going to like the one that is closest to what I want and then once it comes on I'm going to tell the bot yeah this is gross I don't like it and then I'm going to DM the results go back to the bot grab the seed includes all the information that I have given so far... Even if you do not like the pictures that you upgraded do not do not do not reroll please do not re-roll it's literally like you haven't done anything yet. So unless you want to waste hours and hours and hours and hours and hours and hours of doing this all you have to do is once it generates a picture and you upgraded and you still hate it say thumbs down DM the results grab the seed and go right back but make sure that you use your your exact prompt... but I did find out something if you happen to reroll and you're like s*** if you happened to use your last seed with that prompt again you might be able to get your results. Unfortunately you have to do this you have to keep telling the but hey I like this I hate this I like this I hate this because you're literally telling it hey look when I put these codes in that's what I want
Oh! You’re saying Colon? Not Column.
Still watching the video but the beginning is exactly my experience when I tried this before in MJ and I spent forever trying to get specific letters and almost never did. I had a lot better luck doing this in Stable Diffusion where you have a lot more control, but then you lose some of the artistic nature :(
Yep definitely a challenge!
Midjourney has masked text so it cannot be recognised when they use signed/copyrighted works - you will never get a word or letter that you ask for
To get a letter 'T', try using an "!" I have an expression that I use and it always keeps making the "!" into a "T". YMMV, but figured it worth a mention.
"T" font, with particles in the background --ar 16:9 --v 5 - gave me a T ALONE inside of the first four previews. Try that
nice vid but you've been spelling "beautiful" incorrectly
This is why this doesn't qualify as real AI...more like a fancy search..
Midjourney is realy bad with letters I have spend hours and he never made a I for so . Hang in there.kind regards
Ask him all the letters of the alphabet... He will finish by creating a 'T' 🤓
you are good my word is BAIA midjourney refuses to write the I van BAIA 🤣
Thou shall NOT steal. Ban AI or spend eternity in hell.
I hooe it will never. Dont also ruin that part
Adobe Firefly!!
I need a T
Best I can do is P
@Codex Community Well I think I found an answer although I tried it only on two letters A and B, if you want specific letter just need to use negative prompt and exclude all other letters, so for letter A something like this (abstract letter A --no letter B, C, D, E, F, G, H, I, J, K, L, M, N, O, P, Q, R, S, T, U, V, W, X, Y, Z) Note:upon further inspection it seems that for some reason it only works for some letters like A,B,V...tired to hack it but it didnt go well,tried many combinations with and without words like letters and with weights but no sucess
“T” - quotes lolz. 😂
No
F$#% AI
Hey Codex Community, I just watched your video and I must say that it was really informative and well-made. I loved your videos. I was wondering if I could help you edit your highly engaging shorts videos for you.
Man, thought Ai was supposed to make life easier lol
I could have put plants around with the letter T in photoshop manually way faster
Think maybe use mid journey to get a reference, so for example, you got a nice looking F on the first go and just copy that style into photoshop and don’t get lost into tedious midjourney
This is more painful than teaching a toddler how to write letters.
They should use exclamation marks to indicate weight or emphasis. i.e. "make a f*****g letter T for God's sake!!!!!!!!!!!!!!"
You: "I'd like a T..."
AI: "But... maybe instead have a P? P is my favorite letter. What do you say, nice P, no?"
😅
You do not have to add a white background for a jpg. Just export it as jpg, the background will be added automatically.
Stable Diffusion > Midjourney, simply because we need more people using AI then being used by AI.
My technique to get the letter I want. "A big and beautiful 'T' letter capitalized." followed with attributes
@codex Midjoruney doesn't get it because you address it as if it was ChatGPT and could write.
Congratulations Mr you have earned the most Letter T teller in a single youtube video 👏
It seems like you should just train stable diffusion. It's not hard. A lot easier than this
I believe this is partially to help avoid copyright issues
Is it possible to make lowercase letters in midjourney?
Sooo no
This is exactly why Adobe will kill it when they integrate generative art into Photoshop.
Yeah that will be amazing!
They did that a while ago, don't know where you've been, and this is still here.
Did you just chose one letter lolol fake it till you make it
I could of taken the easy part and just said P, but T was the most difficult to produce!
Why spend all of this time asking AI to generate what you want? Just dedicate 60 or so hours to learning the basic functions of a graphic design program like canva, or Corel Draw and have the ability to create your own graphics with a real skill you'll have for a lifetime?
This is clown town, just shorten your video to 1 min and tell us how to do it NOT how to not do it. Stop wasting peoples time
What is wrong with a non-native English speaker doing a slow video at his own pace? At least he had the balls to do it. Plus he is attempting to show his entire process here. You don’t have to watch it, there are millions of other things to click on in UA-cam.
Oh I see your channel doesn’t contain any contact, a real creative dude! 😴
An Australian English speaker I meant to say 🙂
Better yet, just spend all of this time learning a free graphic design program. Even better than that buy a solid affordable program like Corel Paintshop Pro and develop a real skill.
@@daveking-sandbox9263 hmm, I don't think so. I'm Aussie and it sounds like he's perhaps European who's spent some time in Australia.
I know this is an older video (in terms of AI tech progression), but I'm wondering that creating letters will be a significantly easier now that version 6 is out. Seems to be working well for others.