Control Text in AI Images
HTML-код
- Опубликовано: 7 июн 2024
- With some tricks we can add text to your images with Stable diffusion's ControlNet. This ControlNet workflow will help you add any text to almost any image. If the text is too small, remember to inpaint in a higher resolution.
Support me on Patreon to get access to unique perks! / sebastiankamph
Chat with me in our community discord: / discord
Control Lights in Stable Diffusion
• Control Light in AI Im...
LIVE Pose in Stable Diffusion
• LIVE Pose in Stable Di...
My workflow to Perfect Images
• Revealing my Workflow ...
ControlNet tutorial and install guide
• NEW ControlNet for Sta...
Ultimate Stable diffusion guide
• Stable diffusion tutor...
The Rise of AI Art: A Creative Revolution
• The Rise of AI Art - A...
7 Secrets to writing with ChatGPT (Don't tell your boss!)
• 7 Secrets in ChatGPT (...
Ultimate Animation guide in Stable diffusion
• Stable diffusion anima...
Dreambooth tutorial for Stable diffusion
• Dreambooth tutorial fo...
5 tricks you're not using
• Top 5 Stable diffusion...
Avoid these 7 mistakes
• Don't make these 7 mis...
How to ChatGPT. ChatGPT explained:
• How to ChatGPT? Chat G...
How to fix live render preview:
• Stable diffusion gui m...
I think your channel will have 100K+ subscribers within a year if you keep creating superb content like this. Thank you!
Thank you, that's very kind of you to say, you absolute rainbow factory you 😊🌟
@@sebastiankamph 1million in 2 years
@@HowToRealise Remind me in 2 years. If I have 1 mil, I'll send you a gift. And Morgan too.
@@sebastiankamphI absolutely want in on this too! 😂
@@McMartinLC Gifts for everyone! But then you guys gotta help out spreading the word 😂
Future videos should be sponsored by "Yeirkok Bath"
😂
Very useful as usual. With this you have solved the text to logo issue found in Midjourney. Automatic 11111 is getting better and better... Looking forward to your next video Seb :)
Waiting for that Sebastian Kamph /Olivio Sarikas/Nerdy Rodent crossover any day now …
🤫😏
Thanks Sebastian! This is exactly what I've been struggling with. I was ready to accept that AI just isn't there yet, to generate text in images. But this solution is probably even better than a random text shape.
Love your videos, keep it up!
I'm glad you found it useful and I really appreciate your support, you absolute hidden gem! Thank you! 💎
Thank you Sebastian!
You the Controlnet Master 🙌
You know it! Thank you my friend 🦾🤖
i was thinking about it thanks man
i would also love to see a desk setup tour
Very cool technique - text has been such a pain in the butt previously!
Very interesting and useful tip for working on and improving image details. 👍
Excellently explained. I also really appreciate the calm-toned explanation. It seems a minor detail, but I think it is important: it gives time to understand and assimilate the words. 👍
TX Seb 😺
Glad it was helpful! And thank you for the positive feedback, I appreciate it 😊
Your videos are always informative and I love the lil jokes you throw at the beginning of each!!
Thank you! :)
Excellent idea! I feel like the first example with Masked content as original instead of fill mode gives better results, as the sign and its light are better matches that way.
Just about to take a break from binging SD videos. And this video came out.
Tell the developers to stop working and we'll both take a break, deal?
Thank's works well. I found that the text in the image file you make needs to take into account where in the final image you are masking for inpainting. I could not get the inpaint to accept the text from my image correctly until I put the text in the same physical position, that I was masking for inpaint, in my image.
You read my mind. I was just thinking controlnet would work well with text
Good idea but I would say the end result is lacking, there is no such thing as a black neon light. Maybe blurring the text would help
Certainly neon black appears an oddity, but from technological development I also expect this possibility. People have been looking for the perfect black in LED screens for years.
In any case, anything is possible in a Sci-Fi image. 😸 Also, the Cyberpunk genre tends to be a bit surreal, almost a kind of post-impressionism where colours are more an idea than a description of reality. 😸
Can you do another lesson on controlling text in an image? I wanted to put the brand name on the image, I did exactly as you did, but Stable Defusion, regardless of any settings, either misspelled the word or didn't display it at all. Used the exact mask on the image in ControlNet, marked the place of replacement in Inpainting, weights (weight, guidance start, guidance end) to the maximum and less, but still ended up having to replace everything in Photoshop. You're just the only one who touched on the subject of accurate text in the image among all the other bloggers who talk about Stable Defusion in one way or another.
"Yeirkok" sounds like somethin an old mad english man would say xD
Hah, now that you mention it 😄
Buddy you have the highest deree of Black belt in AI kungFu from the Shaolin temple of Stability AI. thanks a lot , Big fan.
Warm words of encouragement, thank you my friend! 🌟
Omg i love you :)
😉
That's a good idea for an extension. Hard part I can think of would be how to access the fonts. Since they need to be installed on a user's system to be seen on their system, unless inside a container format that can embed the font with the image/document such as pdf. But a tab for making text depth maps with a simple web interface similar to the sketch but for text can't be that hard.
That sounds like an extension I would use!
@@sebastiankamph looking into it with GPTchan this morning and it appears the gradio library has the function necessary built into it already. The sketchbox should be a viable container for this function. I'll give it a try.
I think i prefer the "YEIRKOK" sign. xD
Kind of looks like you just slapped the text in from a image editor and it really doesn't fit the scene. But I'll give it a go and see if I get better results myself.
And unless I'm mistaken , you don't need to match the image resolution for inpainting only the masked area. So you should probably stick to a 512x512/768x768/1024x1024 resolution when using that option.
Hi! Have you tried using controlnet for creating 2 characters in specific pose when one of characters is covering other one a bit? And without openpose?
I did a scene with about 15 characters. It's a lot of inpainting steps.
@@sebastiankamph thanks! I guess i just still need to learn that tool more.
I would have liked to see the text just become a nicer version of the "YEIRKOK BATH" sign. I think if maybe feathering the mask might be beneficial for neon signs like this. Also, it might be possible to run it through a slight change where the controlnet is disable and you keep the inpainted area just so that it blends it a bit. Low enough denoising and it won't change the text.
It is possible to get text to work "as is" inside stable diffusion, it's just VERY hard. You can try prompts like "text that says XXXXX", where X is whatever word you want, maybe also adding "logo".. but again, it might just work 1 out of 100 generations, so, hopefully the Stable Diffusion people will get this figured out one day.
@@MaxKrovenOfficial Oh yeah it is definitely possible. Especially if you use img2img and do a bit of just typing right over the area you want text, but if there is already text that requires a bit more work. I think for the speed that controlnet confers is really where it is at. Things were possible before, but needed intricate and time consuming methods.
We need to wait Deep Floyd to be released someday. In this example "YEIRKOK BATH" was good looking so we would a feature to change THAT text and not generate new that is very different from the original.
did you consider using sketch mode instead of depth, that way you dont have to convert foreground to black.
hey i want to run stable diffusion but i am new to the pc world my macbook isnt capable enough to run stable diffusion any suggestions for a pc within the price range of $2500? please help me out
Seems like an 'easy' addition to ControlNet to be able to add text to an image instead of the pose model? Would certainly help with workflow bloat.
Im spoiled. If its possible do it like this now, next week there will be a quicker way 😛 I dont know W 😂😂
I wonder if you could show a tutorial on how to take an image and change its style to where it’s something new, but you can still tell what it originally was. Like for instance a power ranger. Changing the armor but keeping its originality
just do img2img with low denoising strength (~0.5)
@@dailyfunnytv358 i actually can’t run stable. Don’t have a fast computer
@@buzbeediddle7846 you can use a google collab online
I think the inpaint model would give better results.
Is it possible to do graffiti font, like if you train specific graffiti style
Yep!
Dude, what AI-Wizarding school did you go to? I need my letter stat!
I think you need too add some variation in your generation, the text looks bad in the example, like someone new to photoshop made it. Amazing idea though, might have to mess with this. I tried generating my own font by jsut creating a template for it to follow, but it was messy, but possibly this might be a cool solution.
Yooo that means we can now play with existing logos!
Anything goes! Real freedom™
giving me ideas for making logos
Thank you for this, and all other, videos.
1 thing here though: A neon sign is fluorescent gas that lights up. Here you have an outline that has a colour but the sign itself is black. Thats not a neon sign
Also: why did the scarecrow get a medal?
You're welcome! And that's good information! Tell me, why did it get a medal?
@@sebastiankamph it was outstanding in its field
halo sir, how about text in shirt? thats my problem rightnow
Came to this channel only for the pun 😂
😂
I liked it before i even started the video
That's some real superstar moves right there
At this point i only come to your videos for the jokes... yeah cool tutorial, whatever ;)
Maybe I need to do a full dadjoke episode.
Not sure who these other commenters are, redditors probably, they obviously have some form of brain damage
This is amazing Seb, keep up the good work.
Why thank you very much 😊
That’s a lot of work to have little control of final result
Try painting or photoshopping.
Love the jokes! 😂
To be honest, I expected better results. Even just with Photoshop we can do better. But I'm sure AI will get better at this in time.
You can prompt anything. If you feel the end result is lacking, try another model and a more advanced prompt than what I did in this example. I wanted to show that in its absolutely simplest form, base model and basically no prompt, it still delivers. It's basically me driving the car at 30km/h and saying "I expected it to go 120km/h" 😊
0:15 i know 29 letters🧐
I like the approach. The result is kinda meh.
You have an Instagram?
No, RUclips only for now.
@@sebastiankamph We wanted to tag you in our artwork but I couldn't find you. Just wanted to shout you out. Build the community on social media.
This would probably work better if your mask fully covered the text.
Y
😂
No offense but this just looks like a paint pasted text on the image. it would be better if it looked anything like your original sign on the first picture and if it had stylized text