EASY! Stable Diffusion 1.5 Photo bashing with img2img 1.5 Inpainting Model - FULL GUIDE
Vložit
- čas přidán 21. 10. 2022
- How to Photobash using Stable Diffusion 1.5 Inpainting Model
Negative Prompt Download: drive.google.com/file/d/1CcMW...
Install Stable Diffusion 1.5:
• Stable Diffusion 1.5: ...
Support my Channel:
/ @oliviosarikas
Subscribe to my Newsletter for FREE: My Newsletter: oliviotutorials.podia.com/new...
How to get started with Midjourney: • Midjourney AI - FIRST ...
Midjourney Settings explained: • Midjourney Settings Ex...
Best Midjourney Resources: • 😍 Midjourney BEST Reso...
Make better Midjourney Prompts: • Make BETTER Prompts - ...
My Facebook PHOTOGRAPHY group: / oliviotutorials.superfan
My Affinity Photo Creative Packs: gumroad.com/sarikasat
My Patreon Page: / sarikas
All my Social Media Accounts: linktr.ee/oliviotutorials - Jak na to + styl
As a traditional artist and a digital illustrator of 20 years. This is such a powerful tool. Ive noticed a trend of those concept artists that are of a high caliber are actually embracing this as a tool to help us create and a base to then take further the ideation of our designs and i must say this is an invaluable tool!
yes, absolutely. especially with all the automation and the ability to train your own styles
It's the modern equivalent of looking at the brushstrokes of old masters
That negative prompt you gave is EXCELLENT. The quality of my outputs has skyrocketed. Thanks!
awesome!
This method actually shows and proves that visual editing methods are working much better then text in terms of art. You gain so much control, where you can start to fix particular elements and mistakes, made by automation. I hope in near feature it gonna evolve into some sort of mix between 2d and 3d, where you literally "painting forms" and then rotating them on the canvas. Next you can paint it with some "materials" where materials are not just bland colors, but are also can contain volume, facture and even other 3d objects. It will be more realistic then a "traditional" 3d methods and a lot faster then 2d. Like "painting a photo", but without a need in photobashing and keeping gigabytes of libraries.
Visiting your old videos after regaining my interest in SD/A1111... and wow, I didn't know the technique of using composite photos actually has a name ("photo bashing")!
Thank you for taking the time to teach us about this powerful tool!
I thank you for the VERY easy to follow tutorial.
Very helpful and high quality production and description.
Thank you so much ;)
Thank you ❤
I was looking for a video on this topic 👍😊
Thank you so much for this tutorial.
Good stuff Olivio. I need to get into photo bashing a lot more 😃
thank you. it's a lot of fun :)
Thanks so much for your videos ♥
Thank you, great ideas and suggestions. 👍👍
you are welcome :)
good stuff.. great idea for the negative prompt file.. cheers :)
Thank you :)
i cant wait to put your negative prompt in the positive one
Not bad idea , i'll try it too :D
thanks for the negative prompt!
you are welcome :)
AMAZING stuff!
If it's actually "Upsilon" for Y, then I feel like you have to say "Chi" for X. I grew up saying "ex-why plot" but maybe my American schooling failed me there 😂
Tau Upsilon for your great work!
He is also saying "ex-why plot", just in German (the letter Y is called "Ypsilon" in German)
The inpainting model seems very cool 😎
yes, much better than the 1.4 model
@@OlivioSarikas no doubt this would make a fascinating video!? Same concept as this, but completely keep the original composition like with prompt2prompt, you can generate many variations of a single element with the inpainting algorithm, but in the same place in the image!..
❝Below❞ ❤ on one of first screen texts 🤣, only ❜cause u work *SO VERY FAST, OLIVIO* ‼️
lol, thank you. I did not see that
Main question, how to train the model with the own dataset?
Keeping all other parameters the same, did you try using just a prompt without the image reference to determine how much the image affects the final output?
You can use "Interrogate CLIP" to get a prompt of your image
I know, but there has to be something left we do ;) Also when you write your own prompt, you can write what ever you want
Can you add midjourney's parameters to it, like --ar --test/p --creative --aspect etc..??
Great video thank you, subbed... but plz eq your voice brih
do you have a video on how to get midjourney-style results but using stable diffusion ?
Yes, it's possible just by using proper .ckpt models.
I'd suggest Protogen, OpenJourney if you want these realistic looking results
Can’t wait to dive in, I’m on the shitter so it’ll have to wait
lol, to much information ;)
Great video. I wish so much that Serif would commit to AI incorporation directly into their Affinity products. If they did so, they’d finally have a capability that would place them apart and ahead of Adobe products … vs settling to be a poor man’s Adobe clone
i feel like "poor man's adobe close" is selling affinity a bit short. there are many things it does better than photoshop. as for AI generation, adobe already has AI tools in PS and has been researching it for years. they're a bit ahead of the curve here.
can it do img2img without text prompt? I just wanna see variations
Interesting, even on automatic1111 guide mentions that NOT TO USE so much negative prompts, it affects the outcome on a negative way, and you have MORE negative prompts than positives, LMFAO! lol
I'd love to have a "Prompt FAQ" video for Midjourney where people can write to you what they try to achieve but fail. I've tried a few times to "/imagine a bucket splashing colors in form of a butterfly onto a wall", but never got the desired outcome with lots of different prompts.
That sort of thing is too complex for the CLIP model that these ai’s are based on. A new version of CLIP that can do writing - and hopefully composition - is in the pipeline.
@@flyondonnie9578 Wow, thank you for the quick answer :D
So I can stop trying for now :)
I have been using this method for a long time. invented it myself.)the only downside is that the color scheme will be the same as the original one.
for example, if the image is black and white, it will generate black and white images
I had a bug “type error nonetype object is not subscriptable” while using the img2img + the script “img2img alternative text” does anyone had the same?
you probably wrote the wrong model names or selected the work input type for what your wrote in the script fields
Thank you for your answer! You were right, I fixed the bug by renaming the 1.5 file by “model” ~
Can you post your negative prompt please?
Sorry I see it
Why should the seed be -1?
-1 means it generates a random number every time
The whole part about having to input prompts is annoying. I wish it just worked with the image itself. I might as well just text2img. No criticism for the guide, it's good just can't wait for the software to be even better.
Why wouldn't you generate starting images yourself? 🥰
because photos have more details and are real. Like i said, you can stay as close to the original as your want. So if you want a very detailed astronaut with a original uniform, you can do that too
@@OlivioSarikas you're right, I've posted this comment before realizing that you need just parts for those images to create a mockup. Img2img is amazing.
Im trying to make a suggestion but yuotube shadow bans my comments every time very frustrtating :)
strange, they are not under "held for review" either. did you include links? because youtube filters hose comments out
If you use a Negative Prompt in which the word fingers and hands will be repeated leads to the fact that the system stops generating hands, feet at all. Creepy.
You can tweak it. The point is it does work rellatively well as ironing out a lot of issues. Think of it as a guide versus a rule.
@@KadayiPolokov Well, it makes a lot of good pictures if you need a large number of pictures where you definitely won't have grotesque limbs. And if you need one or two really good ones with full-fledged anatomy, so far I haven't found a better way than going through hundreds.