Create Consistent Character Face/Body/Clothes From Multiple Angles
Вставка
- Опубліковано 18 чер 2024
- Tips For Creating Consistent Characters in Stable Diffusion With Fooocus. If you are looking for a way to create multiple angles of a character body and or face this grid method might be for you.
Fooocus is a very simple beginner friendly stable diffusion image generator with endless possibilities. Most AI Image Generators are hard to tackle and quite overwhelming but the simplified UI of Fooocus, along with it’s versatility, is what sets it apart.
Anatomy360: anatomy360.info/anatomy-scan-...
Face Swap in Image Prompt: • Fooocus Tutorial Part ...
Face Swap in Inpaint: • Fooocus Tutorial Part ...
If interested in any of the 4 Grid images I used download them in a zip file here: www.mediafire.com/file/totpz3...
Help support the channel!
www.buymeacoffee.com/jumpintoai
ko-fi.com/jumpintoai
-----
00:00 Intro
01:03 Explanation
02:30 Add Resolution
04:06 Face Grid
07:40 Full Body Grid
08:37 Face Fix
11:03 Wildcards
----- - Наука та технологія
If you don't already have Fooocus, download from here: github.com/lllyasviel/Fooocus
If you want any of the grid images I used you can download them here: www.mediafire.com/file/totpz3ok9hdw0yk/Character-Sheet.zip/file
Hey, you videos on Fooocus are most detailed and very insightful. Thank you. I am trying to swap the object in the image with my own object what do you recommend I do?
Do you run it from pc? Vscode?
I ran it from collab and link expires in like 15 mins
And you seem getting fast results is it gpu or cpu or ram needed on pc to do it fast ? I have 16gb ram with 3070 ti gpu
@@Poppinthepagne I run it on my Windows 11 machine with a 4070 Ti. You should be able to run this fine on your computer with a 3070 Ti. FYI (I speed up the video when its generating to save time) but my generations usually take around 10-14 seconds per image on speed setting.
@@JumpIntoAI thanks for reply, does the link expires like in collab or it’s completely different
@@Poppinthepagne No, the colab is entirely online, If you download and install the local version onto your computer it will produce as many images you like.
Have been doing the same for months but you are way more clear in explaining and actually uploading the video for it. Kudos to your upload 🙌🏻🙌🏻
This video alone I learned so much more about Fooocus. Thank you!
This is what I've been looking for, I was tired of getting the same face looking at the camera, thanks for the video!
Hey! Love the content and thorough explanations of different tools and use cases. Keep up the great work!
Excellent job! finally a real good in dept explanation of fooocus, by far the best for now, THANKS a lot!
Great videos, I love how you are straight and to the point and I don't have to listen to you ramble for 4 minutes before getting to content, great job! Subscribed and hope to see some new tips.
Your videos are very insightful and full of new tips. Thank you!
gawd damm bro, finally someone with a scientific method of making videos and doing work.
This video was great in terms of looking at ideas for character generation. Thanks.
Brillliant video! Extremely new insights and information! Thank you
The wildcard tip is... Wild! Thank you for that!
No problem!
Wow, the mixing image prompt + inpaint trick is epic.
Great tips with grids! Thanks!
Great video!! Lots of great tips... very grateful
wow, i learned many new things today, time to start generating. thanks for the tips!
Thanks for sharing awesome hints
Learned more here than in two weeks by myself
Such quality. Much appreciated
Thanx a lot your videos are really inspiring and clear !!
Thnaks for your great insights and for your generosity in sharing your knowledge and tools, such as images for the grids! 🌟
No problem! I'm glad they were helpful.
Thank you! Very useful video!
you really tried to help, i can say that. thanks mate!
Brilliant! I learnt so much from this, thankyou :D
Incredible video !!!
Fantastic tutorials!!! Straight to the point. Concise. Video and audio professionally well done. So much useful tips and tricks never heard of before. I like it a lot. Keep it up please. More..more..more.!! A hundred thanks.
Thanks! I don't deserve such praise but I do appreciate it. I'm glad so many are finding it useful. 😊
Amazing tutorial! 👏👏👏 thank you very much! 🙂
Super! Thank You Very Much!
This really very very useful! Thank you a lot! 💕
What a nice video!
Good tips, thanks
Thank you 🙏 very useful
Great work man
Thank you very much!
excellent ! thanks a lot
Excelent Tutorial THX a lot
Excellent.
Thank you. I will never do any of this but it was an interesting and informative video.
At least you are honest. 😁
Thank you !
Phenomenal
perfect tutorial.
★★★★★
Perfect video
great video
Great Tips, Tks
Welcome! Glad it was helpful.
Awesome
Video idea: " Dall e How to upload image-to-image consistent characters." Oh love your video.
So many great things in this video, I've been using an older copy of Fooocus MRE. Thank you showing all these cool udpated to v2. I like your editing, I make tutorials as well, which program do you use? I never bother with the zooms or arrows, maybe I should start.
Thanks! I am using the free version of Davinci Resolve. Very new to editing still and have been trying to improve little by little.
Amazing video, thanks so much, i followed step by step and ended with the character sheet created, but I'm not sure what to do with it now
Great video man, can you make a video or send a link to one that gives a guide on fixing hands properly in Fooocus? I could really use a hand with this
Hey mate, would you be able to do a video on LORA training and effecitveness. You're videos are great and easy to understand.
7:12 in you can change Text prompt weights can be changed to add weights to emotions:)
Increíble trabajo!!! Aunque lamentablemente tengo que usar la versión de colab y no puedo editar los archivos.. en fin, saludos desde Argentina
Great, informative video. Thank you. Like. Subscription.
best
Hi! Thanks!
What can we do with this grid pictures after?
thank you always for your advanced tutorials on Foccus, is that right there is a new updated Fooocus with controller pose, and depth ?
Nope, same controlnets as before. Depending on what version/fork you are using. I use the Original.
Ahah, wildcards is a trick I knew already. And you can combine nest them, which is very powerful to generate several random images.
I love wildcards. Perfect for inspiration. They are so simple yet easily missed.
Sorry I did see it before you responded but forgot to delete. Thanks!
Why dosent youtube have a Super Like button💯 :)
😁 Thanks!
This is absolutely brilliant!
This approach may take more time, however the results are far more consistent than what other youtubers are sharing.
Thank you!
Thanks. A bit of a different question, is there a way to naturally synthesize the character files I want in an existing image background file rather than a text prompt?
super interesting and very well designed video. Just one question. How does one use the already downloaded models from Auto1111 in Fooocus in order not to take more disk space?
Yep! edit the config.txt in the fooocus folder, set the paths to your Automatic 1111 folder locations.
can see it in this vid: ua-cam.com/video/828wYIp2HAM/v-deo.html
HI. So to have full body consistency do you use this full body grill? We need to build a storyboard with three characters who have the same body in different scenes "artflow" esemple, and then create a video with runaway or Pica Labs, thank
Hello, excellent video.
But once you have created a character using this procedure, how can we reuse it please ?
How to replace it in other situations ?
I'd like to build the best possible consistent character. Here's the flow I have in my mind at the moment.
1.) Generate a ton of these grids, with different expressions/outfits etc.
2.) Generate a huge variety of final images that feel as consistent as possible
3.) Use these images to then train a Lora
4.) Use the trained Lora to generate all future images. Additionally using faceswaps/inpaint with the best possible match according to the angle/expression if needed.
Does this make sense?
Yes
Yes it does, especially creating a Lora.
Thanks, I understand you explain how to create a grid, but what do you use the grid for? Do you have any other video on how to leverage that multi angle grid, and even multi expression grids? Thanks in advance. Very cool video, learned many things in this one.
Didn't have any specific purpose when first creating it, more of an idea. But I know some have used it to create 3d models and even a LoRA is possible with some effort.
HI. Can you make a video of how to create consistency between two character actors?
Hi , why does the generator always churn out different face of me . i can't get it to copy a face of me in different clothes
I notice they might work for western face. My asian face not doing well. Probably it is because the model the AI is trained with.
Hello, I am a big fan of your content, you have helped me a lot, but in this video it mentions using the "realistic vision" refiner, it is already downloaded, but I don't know how to install it or in which folder to put it? you would help me a lot! greetings from Mexico!
that goes into the same folder as your base model (juggernaut) so the fooocus\models\checkpoints folder
then it will show up in the refiner pull down.
@@JumpIntoAI Perfect! Tnks a lot! Blessings!
can you turn the images into a video where the model shows the clothes on a white back ground?
can the grid outputs be used to train specific lora for that charecter?
Great videos. How is the character sheet made in the first place?
Searched the internet for things like "face reference sheet" till i found images with the same face/body of multiple angles. and then cobbled together what i needed in the right resolution and formation with a paint program.
So what can you do with those multiple angles? Any resources as I’m interested in learning more and what is that called again? A character grid?
It can be an easy way to integrate into a 3d model, or even be the basis to create a Lora. It's not for anything specific, more of an example of what's possible. The main attraction is creating multiples of the same character in a single frame makes them all extremely similar VS every new image generation will never match the last that closely, especially a different angle.
Please create a video about how to well in image generation prompting, like putting the prompts,,,,, (Prompt Engineering:1.4)
Hello! great content, where i can download refiners? mine comes with default one
refiners are just checkpoints you can use together. Example is you might have juggernaut to start, and then Realistic Vision as a refiner to take over halfway. You can get interesting results.
Anything you put in the fooocus\models\checkpoints folder will show in the refiner slot
Major difference is you can use SDXL 1.0 and SD 1.5 models as refiners (only SDXL 1.0 models can be used as your base model)
All of these can be downloaded at CivitAI
Hey man, I rly love your vids.
Is it possible to create your own grid from an already created face in Midjourney? I would probably mix the FaceSwap method with PyroCanny, wouldn't I? I've tried this before but have failed so far and would appreciate any tips!
Trying to face swap onto the grid all at once will usually give poor results, you can try to inpaint a face one at a time and face swap from that (using the debug "mixing image prompt and inpaint" setting) make sure to use the "improve detail" preset and you might want to turn the "inpaint denoise strength" in the inpaint debug setting higher, like .7 or more. The higher it is the better the face swap but it might start losing the angle you want.
Thank you so much! I'll give it a try@@JumpIntoAI
can you explain wildcards more?
Could you make video generate the realstic animal like a cat or dog !
what makes fooocus better/different then 1111, what in your opinion does it do better? i have just discovered fooocus and havent looked into it too much. thanks
It's the ease of use for most. Simpler UI, but I'd say the Inpaint/outpaint is worth it alone.
@JumpIntoAI thanks , I think I might try and see if it works for my workflow.
broo i need all the characters like shinchan doraemon ben10 nad such cartoons which model or lora should i install
Thanks for the tutorial. One question, with people fine but what about characters like if i create a cute looking unicorn how do i create it in different angles, i mean same face but doing different actions/angles?
I'd have to experiment myself as I haven't tried to recreate a non humanoid character consistently.
face swap in fooocus isnt working anyway, the issue in the denoising streight, to make a similar face we need high value, but with the high value the inpaint will not be blended with base picture,
Thx for the nice videos of the amazing Fooocus!
A quick tip
Directly in the Fooocus folder, you have two files, config.txt and config_modification_tutorial.txt, for easy and safely customizing your Fooocus experience!
Yes! This is the user friendly route for sure. I stopped using it only because it kept resetting on me occasionally. But that was a while ago and it might even have been my error.
Do you think, that this is useful with original anime chars? thanks
If I'm doing it in webui, what kind of prompt shall I provide to do make a grid. ?
Hi is possible I have render 3D model and after I just only change body as real character not the cloth all cloth should be not changed just changed model please make one video
So, what happens when you put the same face but maybe in different positions in the image prompt in either all 4 spots with Faceswap selected or at least 3 of them and then a Pyracanny selected on one? Does that give any benefit of doing that or not?
I have tried that, putting 3-4 faces from different angles and setting them all to face swap. Can't say its better or worse without testing it more, but I got a better result for what I wanted at the time with just using a single face.
During inpaint process...why do you remove the refiner?
how to keep face the same for other image generate?
Would there be a way with Fooocus to enhance an existing face in a photo adding skin texture, and imperfections (possibly from another photo, but not swapping the face). Just taking some details of the other face or maybe skin photo. I don't know how to do that.
You could try loading a photo into Inpaint, mask the face, and put it on "improve detail" and then go into the debug setting> Inpaint and turn the "inpaint denoise strength" down to 0.2 or so then in text prompt maybe add whatever imperfections you want. if its too much or too little, increase or decrease the denoise strength.
This video came out just in time. Although, I did not understand why you used PyraCanny instead of the usual Face Swap. Can you please explain?
The face grid plus pyracanny was just to use the structure/angles of the faces in order to create a character with the faces in those positions. You could in combo with the grid image using pyracanny, add an image of a face and face swap onto the grid of faces. If thats what you mean.
@@JumpIntoAI Thank you. You're extremely helpful with these tutorials. I crave these in-depth videos, since I'm only learning AI generation and it feels like I'm wasting a lot of time and resources (including emotional) due to the lack of knowledge and understanding what these settings do.
How can I change the Theme to dark in Fooocus ?
It should be automatic as the dark theme was written into the program a few versions ago (unless you are using a different one then me) otherwise you can add /?__theme=dark in the address bar after 127.0.0.1:7860... ex. "127.0.0.1:7865/?__theme=dark"
how can i create consistent scenarios, e.g. from one background with different angles?
HI. I only have fooocus, how do you make a grill with fooocus?
Hello! In the section "Model Base (SDXL ONLY)" and "Refiner (SDXL OR SD 1.5)" I only have one option that is: Juggernautxl_V8rundiffusion.
What I can do? I can't get the same face.
Greetings. Thank you
with your level of instruction, i wish you would do comfyui.
lol I understand, Comfy still baffles me most times. Just when I think I have a good grasp , I watch something from Latent Vision and get humbled.
I will be branching out from Fooocus, but lack of free time is what makes it a slow process.
what kind of system are u running this on?
GPU is 4070 Ti 12gb
(lots of the video is sped up for convenience) but my image gens take about 10-15 seconds.
I use Fooocus via Kaggle to avoid the error because time runs off while generating process running. Is the any advice for adding 1536 resolution?
Since creating this it is easier to add extra resolution if necessary through a preset file or in the config.txt itself. Assuming you can edit the config.txt in kaggle (i have never used it)
Hi!
I have a question/challenge. I have been trying to recreate this ability, but purely for a face. So, I have used controlnet for generating the faces in the same pose (being a head and shoulders, facing forward pose), but then, I want the ability to add glasses, hats, earrings, etc., and have them be the SAME every time. I would then want to extend this to hair, face shapes etc., so that I could have 10 different faces, all wearing the same pair of glasses, or have the same face, showcasing 10 different types of glasses. Is this possible? Believe me, I have been trying...
Thanks!
Product showcasing is a high request, biggest issue is it's near impossible to get "the exact" product in an image, and even more difficult in multiple angles. There are ways to get closer but can be more trouble than its worth. such as training a lora for each product. Even that wont be 100% accurate though.
Once we have the faces and the body, how we can use them to create a picture of that character??