Stable Diffusion - Experimenting with Blending and Multi Line Prompts in Fooocus

Подписаться 12 тыс.

Просмотров 10 тыс.

50% 1

Master Fooocus and Stable Diffusion for Creative Image Generation! 🎨 In this Stable Diffusion video I will show you how to blend people or things using the multi line prompts in Fooocus. I have no fully figured out tthis feature so if you have more info feel free to leave a comment. #Fooocus #StableDiffusion #DigitalArt #ImageManipulation #CreativeTools
This is part of my ongoing series of videos on installing and using Fooocus for Stable Diffusion. Fooocus is a great interface for using Stable Diffusion and is very easy to get started. It also will run on systems with only 4GB VRAM.
I am running on an i5 32GB RAM Nvidia 3070 8GB VRAM
If you find the video useful and would like to tip, RU-vid does have the Super Thanks option. Those are greatly appreciated so I can have my coffee!!
☕ BUY ME A COFFEE: ko-fi.com/kleebztech
Download Fooocus: github.com/lllyasviel/Fooocus
How to install Fooocus video: • Stable Diffusion - Dow...
Fooocus Beginners Guide: • Super Easy Stable Diff...
Fooocus Upscales and Variations: • Stable Diffusion - Foo...
Checkpoints and LoRAs the Basics: • Stable Diffusion - Che...
Fooocus Poses and FaceSwap - Image Prompts: • Stable Diffusion - Pos...
Zoom with Outpainting: • Stable Diffusion - Zoo...
Inpainting with Fooocus - Don't Regenerate, Fix!: • Stable Diffusion - Inp...
Background Replacement with Fooocus: • Stable Diffusion - Bac...
Fooocus Tips and Tricks (and AI Hands) • Overlooked Fooocus Tip...
Stable Diffusion - Styles in Fooocus - How do they work?: • Stable Diffusion - Sty...
FaceSwap and Consistent Character TipsPart 1: • Stable Diffusion - Fac...
FaceSwap and Consistent Character Tips - Part 2: • Stable Diffusion - Fac...
00:00 Intro
00:00 Multi Line Prompts
01:45 Blending People
05:20 Blending People and Animals
08:20 Blending Animals with Animals

Наука

Опубликовано:

12 июл 2024

Ссылка:

Скачать:

Готовим ссылку...

Добавить в:

Мой плейлист

Посмотреть позже

Комментарии : 77

@KLEEBZTECH 3 месяца назад

If you find the video useful and would like to tip you can buy me some electricity for all these image generations. It is greatly appreciated! ko-fi.com/kleebztech

@AaliDGr8 Месяц назад

What if i want my own pic with lion face what can do through inpaint same technique u did not tell very stpd incomplete vid

@KLEEBZTECH Месяц назад

@@AaliDGr8 Because that will not work. So not a stpd incomplete vid.

@AaliDGr8 Месяц назад

@@KLEEBZTECH it is if you stop being nuisance stubborn , then I believe you can do gradually and share....!

@Arrrr-gj6kk Месяц назад

Man, that's so incredible! Thanks for sharing

@edward_fox 3 месяца назад

I had no idea about this function. Thanks a lot for sharing. I will experiment with this.

@KLEEBZTECH 3 месяца назад

You are welcome! I didn't either and really don't fully understand it yet. But I wanted to share so others could experiment.

@morsadis 3 месяца назад

this is just awesome! This is just a new breath of creativity! Thank you🙂

@KLEEBZTECH 3 месяца назад

You are so welcome! I do love experimenting with new things. Just created this video using that method: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-sHpckcyLWps.html And trying some other ideas.

@morsadis 3 месяца назад

@@KLEEBZTECHLately I've been experimenting with adding anime elements to real photos through the inpaint function. For example, I found out that focus recognizes surfaces, geometry and physics, and also recognizes the location of the light source in the photo. Thus it correctly and organically places drawn soft, loose and liquid objects on real surfaces, as well as light and shadows. For example focus drew snow on the car from the photo, with less snow on the windshield and more on the hood. Also focus recognizes the essence of the object, for example on the electric stove in the photo, focus draws a pot or pan with food cooking. So far, I continue experimenting with recognizing the essence of the object🙂

@KLEEBZTECH 3 месяца назад

Yes it does do very well with that. I am always being amazed with what it can do. Of course then it goes and just does the opposite sometimes. lol.

@michaelanthony4750 День назад

This is great!

@meikmeiker2615 3 месяца назад

Great as always

@KLEEBZTECH 3 месяца назад

Thanks again!

@didiereloy9843 3 месяца назад

Thanks for sharing, I'm going to try it, it's a new approach that's very interesting. Fooocus is a very good tool.

@armondtanz 3 месяца назад

yes, i was banging my head with 1111 and sd... getting 2 year old looking AI results... Ha, loaded up foooocus and BANG, some great images right from the start

@a_vin7644 3 месяца назад

Many tnx. I will wait for your video about new features on v2.3 of Fooocus.

@KLEEBZTECH 3 месяца назад

Oh the pressure. lol. I may have a short video in the next couple days. Not a huge number of changes like the last one.

@Doddibot 3 месяца назад

Oh I had been splitting my prompt across multiple lines just for easy of readability, but looks like I'll have to be careful as this is changing the way the prompt is understood.

@abzaloff_ru 3 месяца назад

Great 👌

@steffen_r 3 месяца назад

Thx for sharing. I should give fooocus another try. I do most of my generations in comfyUI for now..

@KLEEBZTECH 3 месяца назад

I use it for most. I still use others but not as often.

@timbersavage90 3 месяца назад

thanks this helps a lot also you can weight wildcards

@seavyi22 Месяц назад

U are the best

@robertcarlsson1781 3 месяца назад

I found that you can get nice results if you create similar prompts and put them into parenthesis and then applying the weight like this: ( (a hyper realistic, 8k, extreme detail photo of a dinosaur sitting on a rock):0.57 ) ((a hyper realistic, 8k, extreme detail photo of a crab sitting on a rock):0.5)

@KLEEBZTECH 3 месяца назад

Interesting.

@robertcarlsson1781 3 месяца назад

@@KLEEBZTECH I also discovered that the weight depends alot on what data the model been trained on. You might need to empathize some more for specialized models

@KLEEBZTECH 3 месяца назад

Yes that is what I have discovered as well.

@robertcarlsson1781 3 месяца назад

@@KLEEBZTECH Great video BTW, forgot to say that!

@KLEEBZTECH 3 месяца назад

Thanks! Of course now you have me diving into this a little more. 😀

@Morpheus981 3 месяца назад

It will be interesting using it with image prompt, if it is possible

@KLEEBZTECH 3 месяца назад

hmmmmm.... 🤔

@KLEEBZTECH 3 месяца назад

It does.... Sort of...

@fishpickles1377 3 месяца назад

Oh wow, i will have to try this out, i wonder if this can be used with an existing photo of a real person. Say i wanted to merge a picture of myself, toss it in image prompt? Hmm... will have to experiment.

@KLEEBZTECH 3 месяца назад

I have not tried much else yet with it.

@ximorro5247 13 дней назад

But you cannot swap between a text prompt and a photo prompt (or two photos) like you do with two text prompts, can you?

@marcelobaus 3 месяца назад

It wasn't working for me but when I chose the lightning performance (new version) it worked pretty well.

@KLEEBZTECH 3 месяца назад

Interesting. I have not tested that much but I did try originally on speed setting and with less steps seemed to not work as well for me. I will have to check it out since I am working on making a video and need to generate it frame by frame and doing that on quality setting takes forever to do a thousand images. I wonder if it can get the slow gradual adjustments with less steps. 👍

@kzmtsk 3 месяца назад

Try this: portrait, a cat that looks like (Marilyn Monroe: 1.05), close-up. Sometimes it's cat-heavy, sometimes Marilyn-heavy. Sometimes, it's just adds ears to Marilyn. Don't use any style.

@KLEEBZTECH 3 месяца назад

Yup that will work for creating hybrids as well. I started messing with the multi line prompts when trying to blend people. There are multiple ways of doing it that all lead to slightly different results. I see one as asking for what you want and the other is more like forcing the results you want.

@KLEEBZTECH 3 месяца назад

Plus with the multiline prompt I find once you find that spot you can really tweak it to get slightly different results. In my testing I actually did one that I created about 50 images that showed it slowly morph from a frog to a spider with all the minor changes along the way. If I can find a way to automate the changing of the weights I would like to do a couple hundred images and make into a video.

@kzmtsk 3 месяца назад

fooocus updated (2.3.0). We are also expecting a video about it :)@@KLEEBZTECH

@KLEEBZTECH 3 месяца назад

And I just figured out how to automate what I said I could not in my last comment... lol. I am thinking about making a video on it but some of the changes I don't know enough about yet. Dealing with some major plumbing issues in my house so not sure if I will be able to get to that for a couple days. Shall see.

@ravenbandit 3 месяца назад

That's pretty cool.. must give it a go... have you looked into sdxl lightning checkpoints for fooocus?

@KLEEBZTECH 3 месяца назад

I just started messing with SDXL Lightning more the last few days. So far it seems decent but not sure if it will replace regular SDXL for me.

@2010Edgars 3 месяца назад

Now we can made own catwomen. 😁

@KLEEBZTECH 3 месяца назад

Yeah but they look freaky as hell. lol.

@musicacuantica 3 месяца назад

Is it function new on fooocus? I never realized it was there

@KLEEBZTECH 3 месяца назад

Not new.

@DinkoDarlinn 3 месяца назад

can you do that between two image prompts? one image prompt and one prompt? can you apply style to finished "morph"?

@KLEEBZTECH 3 месяца назад

You can use image prompts but it won't work the same way. As for Styles you can use them when generating but it's a little more difficult to get it to always do what you want. But you could add style information manually to each prompt yourself. Then it would alternate between them.

@DinkoDarlinn 3 месяца назад

@@KLEEBZTECH tried it yesterday with two images as character styles + prompt in MJ...working pretty good right from the start

@TerronHD 20 дней назад

Is it possible to delay the prompt? in A1111 you can use [prompt:steps] to delay the prompt for x amount of steps but it doesnt seem to work in Fooocus

@KLEEBZTECH 20 дней назад

It is not.

@a_vin7644 3 месяца назад

Additional question. What is difference between "::" and ",". Generally I use comma, but today I tested "::" and I get same results. Where I'm wrong?

@KLEEBZTECH 3 месяца назад

:: itself doesn't do anything special with Fooocus. If you are referring to the screenshot I posted at the start that was from the Midjourney docs for their multi prompts which is what the multiline prompt in Fooocus is supposed to be similar to. When using weights you just use one : like (keyword:1.1) for example.

@darekspy 3 месяца назад

what model you use ?

@KLEEBZTECH 3 месяца назад

JuggernautXLV8 for these but others should work as well. Although I imagine some might work better than others.

@dpgedward5947 3 месяца назад

I notice that you are not using the current version, which has extra setting, like speed - Lighting etc

@KLEEBZTECH 3 месяца назад

Recorded before the latest update.

@armondtanz 3 месяца назад

ive only just started on fooocus, deleted 1111 and SD, boy that was a minefield, uggghh.... Any vids for LORA training, i dont mind paying premium site if its not too expensive ... Cheers

@KLEEBZTECH 3 месяца назад

Not yet. I am looking for a site that does it well myself but just have not really had the time to dig into that yet.

@armondtanz 3 месяца назад

@@KLEEBZTECH the only one I've found is tensorart. But I'm pretty new to this and the tensorart youtube channel gives a 3 min talk over so it's very very vague. unless u know what hes talking about it's a pretty useless so called tutorial.

@KLEEBZTECH 3 месяца назад

I have not tried this... civitai.com/models/train

@SpaceRonin242 3 месяца назад

@@KLEEBZTECHI just tried it, didnt have enough credits, but I did not find where to add your keyword anywhere.

@xellostube 3 месяца назад

I don't know if it's me that I'm missing something or just that the latest update made this more difficult. Because my results are not as good as yours

@KLEEBZTECH 3 месяца назад

I am using the new update today and doing more experimenting with it and I am seeing no difference. You really have to play with the weights and make sure to use the same seed while adjusting. It does also depend on what you are trying to merge. People is easiest I find and a good place to start.

@xellostube 2 месяца назад

@@KLEEBZTECH with latest latest update it's not working anymore for me btw

@joechip4822 3 месяца назад

I really like your videos about Fooocus and learned a lot from them. But the one thing I don't like is that most users of AI image generation are so focused on 'photorealism' and detail perfection. Why they even try to let totally fantastic objects look as if they were photos. When it comes to art, photorealism has always been a very narrow niche, while on the other hand famous styles like cubism or impressionism played on a totally differently level. Picasso or Van Gogh would never have become famous if they merely tried to replicate photos. They became famous because they explicitly NOT tried to show reality 'as it is', but tried to show angles that normally are hidden from the eye. Maybe the current fooocus (pun intended) on 'realism' very often comes from the fact that most early users of AI come from a more technical background, which traditionally has problems with everything too abstract and metaphorical. I hope this will change over time when more really inspired and creative people are beginning to use it. But for this the systems need to be much more intuitive and must get rid of all the numbers stuff in the GUIs which totally turns artists off. Real art is and to my opinion will always be something vague and spontaneous, even something random. Trying to force as many control on it as possible - especially mathematical control with numbers - totally kills it. We live in a world where great engineers very rarely make good artists, and this is for a reason.

@KLEEBZTECH 3 месяца назад

I think part of it is that many people are doing it for other reasons than just for art. I know often many are using these tools to assist with work that involves photographic images like advertising or planning things out. I myself actually rarely create photo type images for myself but when doing many of my videos I know many people do and also it can sometimes be easier to assess how well it works for some things. For example a painting of a person vs a photo type image it will be easier to notice when the AI messes up. Of course some of my favorite stuff is when it messes up. I myself come from a technical background so my interest in AI is for many different things and not just limited to image generation. But I do prefer generating non realistic stuff myself in general. But I love experimenting with it all and finding new ways of doing things.

@joechip4822 3 месяца назад

@@KLEEBZTECH Thank you, same here... Also coming from a technical, computer science background and having been a professional photographer for quite some time - but also being an avid art lover, I am looking forward to AI use shifting more to the 'art for art's sake' side, away from the mere commercial focus on 'drastically reducing the time to create image material for business'. But this will only happen if somebody creates a convincing GUI that can be used much more intuitively. The currently available GUIs are nothing my artist friends are willing to enjoy and use at the moment. A first step of course is the integration of an LLM which translates simple text commands into rich prompts for AI image generators, like ChatGPT 4 already does with Dall-E 3. If I tell ChatGPT: 'create a modern cover for a book with the title xyz' and I am not satisfied with the first results, I can simply say 'make it a bit more modern!', and most of the time it will work. Tweaking the image creation step by step by simply 'talking' to a LLM mediator and thus juggling with nuances is clearly the way to go. A shame that on the other hand OpenAI still and seemingly in principle does not allow to set even the seed for Dall-E. If they don't open up here Dall-E won't ever be more than a toy or a first starting point for creativity.

@KLEEBZTECH 3 месяца назад

@@joechip4822 Yeah I think there will be lots of changes and improvements in the near future. As for the UIs I think that will also come slowly. Fooocus for example is much better that the others when it comes to using it. But really excited to see some of the almost instant generation and drawing on screen to change the image.