Midjourney & Stable Diffusion - how to fix hands and eyes!
ฝัง
- เผยแพร่เมื่อ 8 ธ.ค. 2022
- The secret quick fix for messy hands and eyes in Midjourney and Stable Diffusion artwork. And a magic quick fix in Photoshop! This is a Midjourney, Stable Diffusion Tutorial about in-painting and out-painting with DALL-E, Stable Diffusion and Photoshop.
SUBSCRIBE
▶ handle: / @levendestreg
▶ You can subscribe to our channel here: th-cam.com/users/levendes...
▶ Read more: levendestreg.dk/en
▶ Download cheat sheet: levendestreg.dk/en/tutorials/
-------------------------------
Links
▶ Documentation: midjourney.gitbook.io/docs/us...
▶ Midourney's website: midjourney.com/home
▶ DALL-E: labs.openai.com/editor
▶ Invoke AI: invoke-ai.github.io/InvokeAI/
▶ monstermash.zone/
▶ huggingface.co/spaces/pharma/...
-------------------------------
This video is created on the Wacom Cintiq Pro 32 with Macbook Pro, Atem Mini Pro, Midjourney, Stable Diffusion, Invoke AI, DALL-E, Premiere Pro, After Effects and Photoshop.
00:00:00
How to fix messy hands and eyes on the artwork you’ve created with Midjourney and Stable Diffusion.
00:00:19
I’ve been so excited to do this video. Because so many of you have asked, what the best way to fix hands and eyes is.
00:00:49
So let’s start by doing a prompt in Midjourney.
00:01:17
So if you’re wondering what Invoke AI is, you should check out the video I did on the install of that.
00:01:39
The "out painting" process refers to generating a new image based on a given description or concept, while the "in painting" process refers to filling in missing parts of an existing image.
00:01:51
So first you load your picture and accept the view here in DALL-E to start out-painting and in-painting.
00:03:08
Remember to only do small areas at a time. And use the eraser to erase what you want the engine to fix.
00:03:16
So let’s just try the same process in Invoke and Stable Diffusion and see if that works better.
00:04:26
But what about her eyes. Because I promised you there was a magic fix in Photoshop.
00:04:52
By the way - if you want to see more about how I created that with Stable Diffusion - and see how Midjourney fares against Stable Diffusion well hit the subscribe and bell button. Because I’ve got an upcoming episode on that.
00:05:11
And in Photoshop I’ll open up neural filters. - ภาพยนตร์และแอนิเมชัน
![Fix Midjourney imperfections using DALL-E for FREE [Outpainting Tool Tutorial] - Part 1](http://i.ytimg.com/vi/q77RAUJOsJI/mqdefault.jpg)
Thanks for the info! Great point about using Photoshop's neural filters. May I ask how to pronounce your channels name? I'd like to give you credit if I ever mention that tip and I don't want to mispronounce your name.
Thank you kindly @FutureTechPilot. In this video: th-cam.com/video/peWFUxiMUNM/w-d-xo.html at 08:31 I pronounce the name Levende Streg. It’s Danish and means lively outlines or outlines that are brought to life.
@@LevendeStreg Awesome, thanks!
@@FutureTechPilot I'm the one thanking you🙌Have a great day!
Wow, can't believe there's actually a tool that fixes what you create on Midjourney. This is great! Thank you!
Yeah it’s pretty great, I agree! And you’re welcome🙌
I love your positive energy and enthusiasm! It's contagious.
Thank you so much Chris! I’m glad you think so. It really means a lot to me!🙌
Great video. As an artist, MJ has been a gamechanger, but the hands issue tends to be a pain. This really helps!
Thank you kindly @JayParker - I’m glad you like the video! 🙌
@@LevendeStreg Yes it is as you said we have to do more outpainting work not to mention inpainting hands and eyes, But if we, from the get-go already have that headroom clearance that would be a great help and would speed up our work. Thank you for your reply.
@@axelgreen7255 hahahaa. Yes you’re right. I think in a couple of month we won’t need to use outpainting so much anymore👍
Only using Stable Diffusion on my 4090 and it makes the card even hotter than gaming does haha. Nice tuturial getting those fingers and hads right is gonna take some time for an AI to learn I guess.
Just found out yesterday about this kinda software.
Guess it will be photoshop haha to correct eyes and fingers.
Welcome to a brand new world of AI then. And thank you for watching and commenting. Yes, it takes a lot of GPU power to run Stable Diffusion. I've given up on running it locally - I can only use cloud service for it.
@@LevendeStreg yeah indeed but they are all paid services and I can run it locally just fine with the gpu.
Im still in early testing stages myself but your vids help a lot cheers.
@@dannyvanlaarhoven Great to hear that. And I would love to hear about your progress. 🙌
Very helpful. Lots to learn, but this was a big help.
I’m glad to hear it Tiger 🙌 thank you!
Nicely put together.
Thank you kindly🙌
Awesome tips!
Thank you kindly @TheWebStylist - glad you found it useful! 🙌
Great instructional video, may I ask how we can type in a "properly controlled prompt" in stable diffusion to give us some "headroom" in our comic cover so that we can put titles or text above our subject? So far I haven't seen anyone teaching this. Maybe you can make this your next video. Thanks.
Thank you so much. That’s a great idea for a video. I would just add space with outpainting in Dall-e or invoke. That would be the easy answer. Hope it works for you. And thanks for watching!
Great stuff 🔥
Thank you kindly. Glad you like it! 🙌
I don't understand ... I erase part of an image and DALL-E does ... nothing. We're supposed to delete part of the description to get it to generate corrections in the image?
Have you remembered to write a prompt up overhead, and write what you want it to correct? Hope that works! 🙌
I'd love to see a video on organising your AI art locally, I've got one folder that is growing with each AI session I go through. It would be nice to have system to organising images and by category/date so you have something to go back to and reference which would be way faster.
Hey @MaxWeir - that’s actually a good idea. I consider how it fits in with the other stuff I’ve planed. Thank you for taking the time to comment and suggest! 🙌
Tak! =)
Det var da så lidt🙌
Doesn't Dall-E each image generation require credits?
Yes every generation costs credits!🙌
Hi! Thank you for this manual and comparison between dall-e and invoke.
I want to add just one thing: while Invoke team is working on faster generation process, there is no solution for the «older Macs». Intel Macs are slow because of their video cards, unless user buy a good eGPU. So it’s like any other fork on any other OS, it depends on a hardware (and Invoke runs very good on Macs, to be honest).
But M1/2 and future Macs have beautiful ML chips inside that can run generations two times faster without RAM usage. And we’re working on inplementation of this feature too 🙂
Thank you for sharing your knowledge. I really appreciate it! 🙌 I am looking into buying a new MacBook or a Mac Studio. Any thoughts on which to choose for best performance with Invoke. As I am actually quite eager to use Invoke AI🤩
@@LevendeStreg I tried to write the answer 4 times and it’s just disappears after that 🙂 So, short answer: Mac studio is ok, M1 Max + 32gb RAM or more is ok.
@@artnetsvetaev great - thank you for taking the time to answer (4 times) 🫣 I really appreciate you reaching out and sharing your knowledge!🙌
@@artnetsvetaev by the way - do you know where to find info on how to update to SD2 in InvokeAI - and how to import models on a Mac? Can you point me in any direction to read about it or a tutorial for Mac? Because all documentations are for PC - and the folder structure and scripts are different on Mac… any hint is appreciated🔥
@@LevendeStreg good questions! First of all, we’re working on implementation of the 2.0 (currently 2.1). I recommend to just wait for the Invoke release. It also will be faster than 2.2.
You can import models by editing the models.yaml file in /config/ or with the !import_models command in Invoke console. But we’re making the new gui for import.
Feel free to ask questions in our discord. I can’t insert links here, so please google it ;-)
Great tutorial, thanks for adapting to AI that bring superhuman artistry out in us humans.
Glad you like it! 🙌 Thank you!
Great Video! I've been experimenting with SD to fix eyes but can't find the idea way. I usually generate 512x512 images, select the ones I like, the use inpainting with face fix to fix before retouching in photoshop. I just can't figure out if I should upscale before fixing and what settings to use. If you have any insight would appreciate that!
Thank you kindly. And you should always upscale before fixing. I have annul coming video where I show the process in SD. I’ll release it in a couple of days!
@@LevendeStreg Awesome look forward to it especially any tips on workflow especially to automate as much as possible! I've kind of got things figured out but it's too time intensive!
invoke needs a Nvidia to be fast. Tried it on my m1 mac and it was unusable. Everything takes forever.
I agree, but you can use Invoke on RunDiffusion. Works like a charm on their cloud GPU. Be sure to check out the promo code in some of my other videos to get 15% off on creator’s club, if you check it out!
Yeah, turned out RTX3090 is good not only at gaming, it takes mere seconds to fix stuff with in-painting in Stable Diffusion using it.
Really nice to know @wrmusic8736 - I'm saving to buy a PC for this instead of Mac. There are just so many issues with Mac. And now Dreambooth is broken. There's a fix - but only for PC. This is an uphill battle with Mac.
How to make dall e works in my country?
Thank you for asking🙌 Well basically you sign up to their service. You get 50 free tokens - and the you can buy more tokens, when you need it. Every token is one prompt or one correction with outpainting.
ya invoke ai is running slow anyways and i am on pc i5 but ya generating 4 images and altering takes like 30 minutes, so i guess i would recommend maybe borrow gpu power or dalli reset.
Oh wow - thank you for sharing Dustin. I wasn’t aware of that it was slow on pc too. And yeah, looking into getting a setup with virtual gpu🙌
Unfortunately, InvokeAI or any other AI generator was never designed to run on i5 processor. All AI tools are highly dependent on the GPU (graphics card). Buy at least an nvidia 3060 10 or 12gb and it will run 7-10 seconds for a single image.
@@artnetsvetaev thank you for your insights. And yes, most of us should probably look into getting virtual gpu! Or better graphics cards!
@@LevendeStreg theres a api shows how much the render cpu and gpu is costing processors wise, probably best solution is google colabs or googlecloud
@@dustinsc2023 agreed! Thanks for commenting. Once I figure out a solution I make a video about it… I’m actually looking into buying a new Mac anyway. So that’s probably what me solution will be. ☺️
👋
Hello Louis. Great to see you back.
and my secret with neural filters are public :)
Hahaha - well I guess so. Thanks for commenting🙏
@@LevendeStreg all for the community
Hahahaha. I'm so sorry