Hi, there is a great help forum here github.com/yoyo-nb/Thin-Plate-Spline-Motion-Model/issues.....but if you are not having any luck please try this alternative method: huggingface.co/spaces/CVPR/Image-Animation-using-Thin-Plate-Spline-Motion-Model
I accidentally had this on loud as I was out in public, some old guys head exploded as this was too much for his mind to comprehend. Data overload. Thanks for the great content
Upvoted and subscribed. For the content of the video, for the lack of clickbait, for the written and easy-to-follow steps, for giving credits and visibility to people who helped you... This is quality content. Thanks a lot.
This does NOT help my obsession with wanting to create more "cool stuff" :). You've put together the most complete workflow I've seen. It's a credit to your insights and passion and I'm really happy for you. It's becoming a bit of a crowded space of UA-camrs already, but you're bringing something different and in my opinion, unique and better.
Love how you've broken down in easy-to-understand terms so many different tools here. This gives me real hope for the future of AI, where communities work together to build tools that work not just for techheads, but for anyone. You're fulfilling an important role in that chain, which is as a very experienced non-technical user explaining things to other non-technical folks who want to get started. That's a link often missing, so thanks!
Brilliant stuff! I'm a former art/media student who has had a bad spell of many years, resulting in a lack of creativity and slight depression. AI and videos like this have really kicked my creative energy back into 1st gear! I'm now making scripts for all kinds of projects, images and design, and now, thanks to you, even animations! Thank you so much!
So happy to see this comment, i've been having an ongoing discussion with a friend about AI and it's effect on creativity... My friend thinks i'll be the end of creativity, and i argue the opposite.
@@BlackMoridin Definitely the opposite. I'm still using it and creating stuff. To be honest, it is mostly for myself to create stories and characters. However, until now, I have never had an instagram page in my life. Now I'm running two channels and putting stuff in there. Also, have written an article for Medium, which I hope to publish soon. So, all in all, am doing much better than before regarding my creativity.
In all of this UA-cam chaos, there are lights in the tunnel, your channel being one of them. Well articulated, no hokey lazy humor, deep demos and excellent instruction. If I had tons of money I would pour that into this resource, but alas, I'm here for the free teaching, so ahem, y'know. Keep rockin' AI artist sista!
Haha What a kind comment! I do it because I’m passionate about it. Also I love sharing the info and being involved in the community :) Thank you for your comment, it made my day !
This is the bright side of AI. Beautiful tools being explained by a competent and engaging person. Thank you for your hard work. It's channels like this that will teach the next generation to do amazing things.
What an ingenious way of creating a consistent character. It's certainly a lot quicker and easier than the way I've been doing it. Great work and subscribed! Thank you!
I have followed the video with attention, and I can say you are a great teacher. Unfortunately today I did not learn how to create a character in ANY pose, but I have learned how to create a consistent character. I hope in the future your videos will reflect the title's promise. Thank you, PromptMuse!
Glad to be able to say I was here at 4000-something subscribers! You are probably the best I’ve seen on YT teaching this stuff! And also someone buy this girl a 3090!
I've been so addicted to learning and running my own stable diffusion I forgot there were other AIs out there. Going to have to try midjourney after I finish training my model. Nice to see alot of people getting interested in it.
I have been working on a personal story that went from book, to unity, to UE and I've always struggle with being one person doing everything. I guess that after 15 years of starting my journey, I've finally found my perfect ally in AI. I can clearly see the path ahead and how I can finally make my vision real... of course there is always a caveat... we just got a baby and I can't put the free time in the computer hahaha Thanks for the tutorial, well explained and straight to the point! Subscribed!
Who else is amazed by AI yet at the same time feels like we've unlocked something forbidden and it is already way too late to prevent chaotic events from happening... our future seems crystal clear now
I'm amazed childhood ideas I feel like I'm slipping behind it's like Star Trek holodeck you're only limit as your imagination and it's probably a good idea if you have a degree art history I feel like being 37 years old I'm too old. Do you have to have a desktop or can I do this on my iPad
@@promptmuse Indeed, it will be okay in the long run. I'm still an AI enthusiast although I stopped posting my ai-generated visuals online ever since I realised how unethical Stablediffision deep learning models actually are... I feel guilty for contributing to the ai community lately, it's a conflicting situation since I am also an artist
@@promptmuse That's a dismissive and ingenuine response. Sure, blackholes are literal tears in space. catastrophic failures of reality to maintain time and reality in a way that's measurable or observable. But that has NOTHING to do with what is happening in our lives right here and now on Earth. Sure if the entire history of humankind was erased, it would be a little blip and all would be fine, but that's in the eyes of God and infinity, not the eyes of us right here. It would be catastrophic for us for our world to collapse. And saying that that doesn't matter because...oh universe be big....nihilism...yadda its fine... isn't just ignorant its dangerous. We can't ignore one dying or hurt person, because oh yeah, there are 8 billion more. We can't ignore our own planet because, the big old universe will get on fine without us. And we shouldnt ignore the potential horrors AI could cause....Just because it's fun to use...and yadda let's just say it will be fine when really it clearly won't be. Ignorance of the inevitable is what makes it inevitable. Knowledge and resistance is the key to mitigating the issues we see coming.
My inner battle was won when I realized I'm like a songwriter.. I can't sing, I can't draw or paint, but my imagination is 1000mph and I deserve to have it imagined into reality:) if not just for my own sanity lol.
I think what a lot of us are interested in is how to upload a posed image and have all those poses have the character rendered consistently using the reference images. Sort of akin to tracking data for animation.
I would imagine that this could work for that, just with a lot more steps? Like if you can fine-tune it to have an “idea” of your character, then you could use controlNet to make it make images of your character in whatever pose? Though I would imagine that it would be possible-but-difficult for someone to make a model that would like, copy a character from one image and make it match another pose, instead of needing to fine-tune separately for each character?
Holy shit, now this is how you do a tutorial. I was intimidated af at first seeing all of this, but you made it so easy to follow along with and I had a successful first go. Can't wait to see where I can take this! You earned a new sub. Can't wait for the next vids!
Great video and thanks for making it easier for more people to get access to these tools. Your manner helps make a daunting "techy" activity actually doable. Well done!
Thank you for this incredibly detailed step-by-step guide. Exactly what I've been looking for. Thanks for distilling this convoluted process down and taking the time to put this tutorial together!
I’ve watched several other videos trying to explain dream booth and just didn’t click for me. Thank you for this fantastic, concise and understandable tutorial!
OMG thank yeeeeeeeeeee :O I saw CS bloom while I was in my senior year of phil and wanted to jump on the train and I just started my first semester of CS but it's already gotten so far ahead of what interested me back then. O-O so cool
Yeah, we will be making ai driven short films here soon and I’m all for it. I tried learning 3D modeling but had a terrible time trying to model a realistic 3D character.
This is Brilliant! Thank you for taking the time to explain the workflow clearly and precisely. Will Definitely give this a shot. I am thinking if I train a model using my own Pictures, then I can become anyone using the power of AI. I can create talking head videos of myself as an astronaut, a professional fighter, a Rockstar, a pirate etc. I think we are not too far from being able to create our own homemade AI-enabled animation short films. Soon we'll have motion tracking and such. What a time to be alive!
A.I. is already a game changer and the revolution is still silent. I was born to early in history. I went from abacus to DALL-E over 60 years, this is the brave new world I was hoping for especially in creativity, but hopefully also in rational public policy as an advisory tool.
BRAVO! you are an amazing teacher! thank you very much for sharing those information in a very efficient way. I will try with a drawing, or a painting soon!
Just a suggestion for anyone who isn't get very good results, 1. Use images which have mostly the face, as the model can handle the body using positive and negative prompts. I do recommend you scale the images and have them High Quality 500x500 with a high DPI should work for you perfectly. 2. Make a simple facial recognition script, and what to do is put all the images you fed into the AI in 'known_person' folder, and then from here generate your images after each result before it adds it into the list, run facial scan against it, if the result says it isn't the person, or you're using a percentage, have it regenerate an image until you have the amount you require. You could train the model more but this kind of cheats that area where you're able to just have an automation to pick the best ones that match the persons face. 3. Always use negative prompts and remember () around the keyword adds to the weight so it leans on this more than ones outside of this for example you could put ((poorly drawn face)), ((poorly drawn eyes)) and you will get better results, and ((non symmetrical eyes)) works better for the eyes as sometimes it looks like it over laps eye lashes. Peace out.
Hello! I just started learning about AI generated images, and I have to say that your UA-cam channel has been incredibly helpful in my learning journey. Thank you for all of the valuable information and insights you provide! In addition to the great content on your UA-cam channel, I really appreciate that you are willing to allow a FAQ in the comments section. Not many UA-camrs are open to this, and it has been a huge help for me as I am just starting to learn about the subject. I just wanted to let you know that this comment was actually generated by an AI, but it does accurately reflect my thoughts and appreciation for your UA-cam channel. Thank you! :)
0:00 Intro ==== 1:06 Step 1 - Midjourney: Create the character face. 2:40 Setp 2 - Stable Difusion: Create a video from character. 10:20 Setp 3 - Adding body to character. 15:23 Setp 4 - Training AI on generated images. ====
This is very cool, the only weakness is that the model will always generate face on images since thats what it was trained on. Need to see if there's a way to generate profile shots of a face based on a face on generation then you would be able to make a character that could be in any pose looking in any direction :)
@Mark H that would work for a model based on something but this one is complete ai. Perhaps rng brute force is they way. Generate a few hundred images and hope one is a good match to be the profile shot
@@youngwolf0 Yup - can't think of another way at the moment. Brute force + decent enough pshop skills to eke out the missing profile shot and add it to the final batch.
I've thought about using all these different tools together to do this, but I've never actually done it. It's so cool that you spent the time putting this together. It's also going to be interesting what they do with SD V2.0 when it's fully released, because they have said there is a much easier and faster way to train models now. I think they might actually be talking about only needing one image, which has been talked about before.
That would be a very neat development ! I’ve not installed SD 2 yet. Looking forward to trying it myself, as the feedback from the community so far is that is… meh 😂 So looking forward to forming my own opinion. Hoping it’s good 🤞🤞
@@promptmuse they decided to remove all artists & known characters + famous people from the default SD 2 model(s). So people are getting bad results because of that. The good news though is they've done that so they have a clean model to train from and they say more detailed models are on the way + an easier method to train your own models. So althought SD 2.0 seems bad to people at the moment, I personally think people are going to be shocked how good it suddenly becomes in the next few weeks. I have alot of hope for it 🙂 I'm also really interested in its Depth2Img feature, which will hopefully give us way more control on where things appear in the rendered image. I can't wait to see what happens 🙂
This is amazing! Perfectly condensed useful information and really well explained! This is going to be one of the top channels on this subject! Glad to be here from (almost) the start of the ride, when we were not millions of subscribers yet! XD
Here are some tips (it has helped me for complex poses and it may not be the most practical but it helped me) with the clipstudio program I made the exact pose with the camera angle that I wanted, then I took it to photoshop and painted a little, I basic so that SD understands, then in SD with the cara model that I had trained, use img2img some mask and that's it, it might seem like a lot of work but it's actually quite fast.
You should have more subscribers, really great videos! I do think someone will make it drop dead simple to 3D convert a photo/ai photo making it simple to drop into animation software(lots of steps here). I've seen some newer platforms making an attempt at this but wondering what you and others here have seen...
I think I saw a tutorial on building a space shuttle that took less steps... but it was an amazing video and you did a great job presenting all that madness. Wow... I hope it's useful in a lot of ways after all that.
These tutorials are gold. I'd be really keen to see what happens when you use 2 to 4 models in a prompt. What does it look like when you ask stable diffusion to create a scene with multiple models interacting with each other?
Great explanation! There should be a service to just address this task in one shot. Upload 10 perspectives and emotions of a character, hit submit and done. 🙂
Hello. That wonderful tutorial was working till the 9th of February. But since the 10th of February, I am getting an error on step 4 (upsize and export video): NotImplementedError: A UTF-8 locale is required. Got ANSI_X3.4-1968 Any ideas? Thanks.
Hi thank you for the amazing video. I just had a question because the program successfully ran all the way through step 3 but then stopped working at step 4 saying that "A UTF-8 locale is required. Got ANSI_X3.4-1968". I never learned any coding so I have no idea what this means. If you have any idea on how to fix this I would greatly appreciate it!
Thanks for great tutorial. It's only frustrating that I can't complete it because of annoying error. As you kindly mentioned in the end of the video, that we can post them here, I'm gonna use this opportunity. I run into this error: "EOFError: Ran out of input", on the step 3. Perhaps you can help me to fix it
It's because the first step didn't run correctly. The checkpoint file is no longer being hosted. If you Google, you can find the original Thin-Plate-Plate-Spline-Motion-Model and find the original Colab and replace the code in that cell with the ones to download from Yandex disk.
I can bet that soon, in just Midjourney, you will be able to create a whole character. You will be able to generate and pick how character's face looks on both sides, as well as it's body. Then, you will be able to save that character with unique I'd to it and use it to generate scenes with it. Next big step it will be generating animations. God...I'm so excited!!! I always wanted to be a movie director! And it seems... that I will be...SOON! xD
Dont depend on ai for that, just use 3d softwares to all that stuff. Dont be limited by ai when you are free to do everything you want with 3d softwares, you wont like your characters face glitching in the most emotional scene in your mivie right?
Hi, there is a great help forum here github.com/yoyo-nb/Thin-Plate-Spline-Motion-Model/issues.....but if you are not having any luck please try this alternative method: huggingface.co/spaces/CVPR/Image-Animation-using-Thin-Plate-Spline-Motion-Model
the form no longer works :'(
Can you redo this where its completely on a local machine?
I accidentally had this on loud as I was out in public, some old guys head exploded as this was too much for his mind to comprehend. Data overload. Thanks for the great content
You are a gem!
@@Arkhkalos she accidentally added the "....but" into the link. just copy the important part and paste it into the address bar for it work.
Finally! A video that doesn't assume I have any coding experience and goes slow enough for a numpty like me to follow step by step.
Upvoted and subscribed. For the content of the video, for the lack of clickbait, for the written and easy-to-follow steps, for giving credits and visibility to people who helped you... This is quality content. Thanks a lot.
This does NOT help my obsession with wanting to create more "cool stuff" :). You've put together the most complete workflow I've seen. It's a credit to your insights and passion and I'm really happy for you. It's becoming a bit of a crowded space of UA-camrs already, but you're bringing something different and in my opinion, unique and better.
Thank you very kind. I haven’t been outside due to these obsessions, is the sky still blues ??? Thank you for the kind comment, really nice. Cheers!
She showed us, but lets give some love to Sixhaunt for putting it together. Can't have one without the other. Both are helping the community immensely
@@rne1223 Totally !! I wouldn’t of known this workflow without him!! Also he shared everything with me. A true asset to the AI community 🙏
She could have saved herself a lot of time and trouble by just filming herself, as she has the look of an AI character.
That is how it works by design tho
Love how you've broken down in easy-to-understand terms so many different tools here. This gives me real hope for the future of AI, where communities work together to build tools that work not just for techheads, but for anyone. You're fulfilling an important role in that chain, which is as a very experienced non-technical user explaining things to other non-technical folks who want to get started. That's a link often missing, so thanks!
This is another level, didn't expect to see something like this on UA-cam at all. I am blown away! Please keep it up!
Holy moly, this is so good! Thank you for this genius research and development. So many people looking for this.
I was blown away just by the animation part alone. So many possibilities.
Brilliant stuff! I'm a former art/media student who has had a bad spell of many years, resulting in a lack of creativity and slight depression. AI and videos like this have really kicked my creative energy back into 1st gear! I'm now making scripts for all kinds of projects, images and design, and now, thanks to you, even animations! Thank you so much!
Yes, love that for you. So much to look forward to in this space, this is just the start :)
So happy to see this comment, i've been having an ongoing discussion with a friend about AI and it's effect on creativity... My friend thinks i'll be the end of creativity, and i argue the opposite.
@@BlackMoridin Definitely the opposite. I'm still using it and creating stuff. To be honest, it is mostly for myself to create stories and characters. However, until now, I have never had an instagram page in my life. Now I'm running two channels and putting stuff in there. Also, have written an article for Medium, which I hope to publish soon. So, all in all, am doing much better than before regarding my creativity.
It was nice to see such a complete, full workflow for using these new tools and putting them towards practical uses. Great video!
In all of this UA-cam chaos, there are lights in the tunnel, your channel being one of them. Well articulated, no hokey lazy humor, deep demos and excellent instruction. If I had tons of money I would pour that into this resource, but alas, I'm here for the free teaching, so ahem, y'know. Keep rockin' AI artist sista!
Haha What a kind comment! I do it because I’m passionate about it. Also I love sharing the info and being involved in the community :) Thank you for your comment, it made my day !
one the best, most thorough tutorial vids i've ever seen. . . what a game changer!
This is the bright side of AI. Beautiful tools being explained by a competent and engaging person. Thank you for your hard work. It's channels like this that will teach the next generation to do amazing things.
What an ingenious way of creating a consistent character. It's certainly a lot quicker and easier than the way I've been doing it. Great work and subscribed! Thank you!
I have followed the video with attention, and I can say you are a great teacher.
Unfortunately today I did not learn how to create a character in ANY pose, but I have learned how to create a consistent character.
I hope in the future your videos will reflect the title's promise.
Thank you, PromptMuse!
Glad to be able to say I was here at 4000-something subscribers! You are probably the best I’ve seen on YT teaching this stuff!
And also someone buy this girl a 3090!
Thank you for being here :) I hope you stay with me !
I've been so addicted to learning and running my own stable diffusion I forgot there were other AIs out there. Going to have to try midjourney after I finish training my model. Nice to see alot of people getting interested in it.
Thanks!
You are incredible. I, too, am obsessed with workflows but you're actually making videos about it and you're on another level.
love it very resourceful. NGL these kind of videos should go as viral imo
Thanks
Thank you so much 💕 That is really appreciated 🙏💕🙏🙏
This is just the beginning. Insane
I have been working on a personal story that went from book, to unity, to UE and I've always struggle with being one person doing everything. I guess that after 15 years of starting my journey, I've finally found my perfect ally in AI. I can clearly see the path ahead and how I can finally make my vision real... of course there is always a caveat... we just got a baby and I can't put the free time in the computer hahaha
Thanks for the tutorial, well explained and straight to the point! Subscribed!
This woman is simply my dream woman, what a personality and professional! Cheers!
We’re in the early days of AI, this channel will likely have millions of subscribers with in the next 5 years.
haha thanks, I just do it as a hobby as it’s really interesting and the best way I learn is by teaching 😂 so it’s a win win! But, thank you !
In Planet earth they are not 5 millions of people smart enough to understand what this channel speaking about
@@blender_wiki lol
Totaly agreed! Simply because she is a great tutor! I will defenitly follow and recomend!
Millions of subscribers where AI now wants to learn how to render itself?
Who else is amazed by AI yet at the same time feels like we've unlocked something forbidden and it is already way too late to prevent chaotic events from happening... our future seems crystal clear now
Out universe is full of unpredictable chaotic events. This is just a small blip. It will be okay 🙏
I'm amazed childhood ideas I feel like I'm slipping behind it's like Star Trek holodeck you're only limit as your imagination and it's probably a good idea if you have a degree art history I feel like being 37 years old I'm too old. Do you have to have a desktop or can I do this on my iPad
@@promptmuse Indeed, it will be okay in the long run. I'm still an AI enthusiast although I stopped posting my ai-generated visuals online ever since I realised how unethical Stablediffision deep learning models actually are... I feel guilty for contributing to the ai community lately, it's a conflicting situation since I am also an artist
@@promptmuse That's a dismissive and ingenuine response. Sure, blackholes are literal tears in space. catastrophic failures of reality to maintain time and reality in a way that's measurable or observable.
But that has NOTHING to do with what is happening in our lives right here and now on Earth. Sure if the entire history of humankind was erased, it would be a little blip and all would be fine, but that's in the eyes of God and infinity, not the eyes of us right here. It would be catastrophic for us for our world to collapse. And saying that that doesn't matter because...oh universe be big....nihilism...yadda its fine... isn't just ignorant its dangerous.
We can't ignore one dying or hurt person, because oh yeah, there are 8 billion more. We can't ignore our own planet because, the big old universe will get on fine without us. And we shouldnt ignore the potential horrors AI could cause....Just because it's fun to use...and yadda let's just say it will be fine when really it clearly won't be.
Ignorance of the inevitable is what makes it inevitable. Knowledge and resistance is the key to mitigating the issues we see coming.
My inner battle was won when I realized I'm like a songwriter.. I can't sing, I can't draw or paint, but my imagination is 1000mph and I deserve to have it imagined into reality:) if not just for my own sanity lol.
I think what a lot of us are interested in is how to upload a posed image and have all those poses have the character rendered consistently using the reference images.
Sort of akin to tracking data for animation.
Did you have any luck figuring it out? I'm still having trouble!
@@Queenwraith nope.
I would imagine that this could work for that, just with a lot more steps? Like if you can fine-tune it to have an “idea” of your character, then you could use controlNet to make it make images of your character in whatever pose?
Though I would imagine that it would be possible-but-difficult for someone to make a model that would like, copy a character from one image and make it match another pose, instead of needing to fine-tune separately for each character?
Holy shit, now this is how you do a tutorial. I was intimidated af at first seeing all of this, but you made it so easy to follow along with and I had a successful first go. Can't wait to see where I can take this! You earned a new sub. Can't wait for the next vids!
This tutorial is awesome. It covers so many things that I didn't know.
Great video and thanks for making it easier for more people to get access to these tools. Your manner helps make a daunting "techy" activity actually doable. Well done!
Thank you ! I cannot even turn on a tv tbh haha
Thank you for this incredibly detailed step-by-step guide. Exactly what I've been looking for. Thanks for distilling this convoluted process down and taking the time to put this tutorial together!
We are also working on Al Art recently, Al Art is the future! Thanks to the blogger for sharing.
I’ve watched several other videos trying to explain dream booth and just didn’t click for me. Thank you for this fantastic, concise and understandable tutorial!
OMG thank yeeeeeeeeeee :O I saw CS bloom while I was in my senior year of phil and wanted to jump on the train and I just started my first semester of CS but it's already gotten so far ahead of what interested me back then. O-O so cool
Yeah, we will be making ai driven short films here soon and I’m all for it. I tried learning 3D modeling but had a terrible time trying to model a realistic 3D character.
that is the vid that will let me do AI, thanks!! workflow is gold
This is Brilliant! Thank you for taking the time to explain the workflow clearly and precisely. Will Definitely give this a shot. I am thinking if I train a model using my own Pictures, then I can become anyone using the power of AI. I can create talking head videos of myself as an astronaut, a professional fighter, a Rockstar, a pirate etc. I think we are not too far from being able to create our own homemade AI-enabled animation short films. Soon we'll have motion tracking and such. What a time to be alive!
That is very cool. I’d love to see the finished results ! For sure, exciting times are here 👍
This is insane!!! In a good way of course, I have a Mac though….😅
A.I. is already a game changer and the revolution is still silent.
I was born to early in history. I went from abacus to DALL-E over 60 years, this is the brave new world I was hoping for especially in creativity, but hopefully also in rational public policy as an advisory tool.
Thanks for sharing! It's so well explained that I have no prompts for describing it😁
BRAVO! you are an amazing teacher! thank you very much for sharing those information in a very efficient way. I will try with a drawing, or a painting soon!
Just a suggestion for anyone who isn't get very good results,
1. Use images which have mostly the face, as the model can handle the body using positive and negative prompts. I do recommend you scale the images and have them High Quality 500x500 with a high DPI should work for you perfectly.
2. Make a simple facial recognition script, and what to do is put all the images you fed into the AI in 'known_person' folder, and then from here generate your images after each result before it adds it into the list, run facial scan against it, if the result says it isn't the person, or you're using a percentage, have it regenerate an image until you have the amount you require. You could train the model more but this kind of cheats that area where you're able to just have an automation to pick the best ones that match the persons face.
3. Always use negative prompts and remember () around the keyword adds to the weight so it leans on this more than ones outside of this for example you could put ((poorly drawn face)), ((poorly drawn eyes)) and you will get better results, and ((non symmetrical eyes)) works better for the eyes as sometimes it looks like it over laps eye lashes.
Peace out.
thanks for sharing these techniques. it is a very clear tutorial
I work in IT handling files with PHI. This stuff is so much more exciting. I am so amazed by your videos. Thank you so much!
Thanks for the high density informative content!
Hello! I just started learning about AI generated images, and I have to say that your UA-cam channel has been incredibly helpful in my learning journey. Thank you for all of the valuable information and insights you provide! In addition to the great content on your UA-cam channel, I really appreciate that you are willing to allow a FAQ in the comments section. Not many UA-camrs are open to this, and it has been a huge help for me as I am just starting to learn about the subject. I just wanted to let you know that this comment was actually generated by an AI, but it does accurately reflect my thoughts and appreciation for your UA-cam channel. Thank you! :)
0:00 Intro
====
1:06 Step 1 - Midjourney: Create the character face.
2:40 Setp 2 - Stable Difusion: Create a video from character.
10:20 Setp 3 - Adding body to character.
15:23 Setp 4 - Training AI on generated images.
====
And for this one! Thank you for taking the time to sort that :)
@@promptmuse You're most welcome.
This is very cool, the only weakness is that the model will always generate face on images since thats what it was trained on. Need to see if there's a way to generate profile shots of a face based on a face on generation then you would be able to make a character that could be in any pose looking in any direction :)
I wonder if you could instead generate whatever and the inpaint the face in.
I guess u would just take some extra side profiles and add them to the dataset?
@Mark H that would work for a model based on something but this one is complete ai. Perhaps rng brute force is they way. Generate a few hundred images and hope one is a good match to be the profile shot
@@youngwolf0 Yup - can't think of another way at the moment. Brute force + decent enough pshop skills to eke out the missing profile shot and add it to the final batch.
Phenomenal walkthrough and results!
This is a great tutorial! Can't wait to try it out myself. Thank you!
thank you so much my muse, amazing work
I've thought about using all these different tools together to do this, but I've never actually done it. It's so cool that you spent the time putting this together.
It's also going to be interesting what they do with SD V2.0 when it's fully released, because they have said there is a much easier and faster way to train models now. I think they might actually be talking about only needing one image, which has been talked about before.
That would be a very neat development !
I’ve not installed SD 2 yet. Looking forward to trying it myself, as the feedback from the community so far is that is… meh 😂 So looking forward to forming my own opinion. Hoping it’s good 🤞🤞
@@promptmuse they decided to remove all artists & known characters + famous people from the default SD 2 model(s). So people are getting bad results because of that.
The good news though is they've done that so they have a clean model to train from and they say more detailed models are on the way + an easier method to train your own models.
So althought SD 2.0 seems bad to people at the moment, I personally think people are going to be shocked how good it suddenly becomes in the next few weeks. I have alot of hope for it 🙂
I'm also really interested in its Depth2Img feature, which will hopefully give us way more control on where things appear in the rendered image.
I can't wait to see what happens 🙂
You're doing incredible work for the space and we appreciate you infinitely!
Using prompts from data models that we trained and really original art. Where artist didn’t get any consent ? Really?
thanks so much!
This is amazing! Perfectly condensed useful information and really well explained! This is going to be one of the top channels on this subject! Glad to be here from (almost) the start of the ride, when we were not millions of subscribers yet! XD
Thank you so much!
Very nice workflow!
This is ingenious. Good job and thank you. Subscribed.
Thank you. That is A LOT of info but fortunately I can come back
Wow... You are amazing, please keep it going
Wow, amazing video!!! Thank you for taking all the time to create a tutorial. You are doing an awesome job!
Wow. You are an awesome instructor. Thank you so much for putting this together.
Wow! Again, excitingly helpful and informative! thank you!
Here are some tips (it has helped me for complex poses and it may not be the most practical but it helped me) with the clipstudio program I made the exact pose with the camera angle that I wanted, then I took it to photoshop and painted a little, I basic so that SD understands, then in SD with the cara model that I had trained, use img2img some mask and that's it, it might seem like a lot of work but it's actually quite fast.
just found your channel and wow i am impressed and THANK YOU so much for sharing. I really like your straightforward vids. Love from California !!
Hey, thankyou 🙏
You should have more subscribers, really great videos! I do think someone will make it drop dead simple to 3D convert a photo/ai photo making it simple to drop into animation software(lots of steps here). I've seen some newer platforms making an attempt at this but wondering what you and others here have seen...
Awesome! Thanks a bunch for making the video, showing how to pull it all together :)
Excellent video. Very complete and well explained. You are the BOSS.
very complicated steps but they worth , thank you for that huge effort....and for this amazing tutorial
Thank you very much for this tutorial. It has given me a an idea for something that I hope works out.
I think I saw a tutorial on building a space shuttle that took less steps... but it was an amazing video and you did a great job presenting all that madness. Wow... I hope it's useful in a lot of ways after all that.
WOW! Amazing results - from a bleeding edge process map...
This was FANTASTIC! Thank you so much!
These tutorials are gold. I'd be really keen to see what happens when you use 2 to 4 models in a prompt. What does it look like when you ask stable diffusion to create a scene with multiple models interacting with each other?
Omg thank you so much! I run these locally never thought using these tools together like this to get ref video!
Great explanation! There should be a service to just address this task in one shot. Upload 10 perspectives and emotions of a character, hit submit and done. 🙂
Great tutorials, jam packed with clear info.
Very cool! Lots of great information simply and nicely explained. Thank you so much!
this is exactly the type of thing i've been looking for. i wanted to make some music videos to my songs with characters that arn't me.
Good job, Amelia, good luck and freshy inspiration to you
Your tutorials are inspiring. Keep the good work!
Thank you for a well explained and useful material.
Your tutorials are perfect
Great effort. I can see it take a a lot of focus on the steps. Thank you. ❤
Very good demo and to the point. Subscribed.
Thank you so much!
This is awesome! Thank you for the tutorial.
You are welcome
I work in tech and i feel so behind with this. WOW. The animation blew me away and there was more!
It’s moving fast than I can handle 😂🥴 Got some great stuff in the next video !
Holy shit! You are beautiful. I'm not into ai generated art, but I'm subscribing anyway.
Hello. That wonderful tutorial was working till the 9th of February. But since the 10th of February, I am getting an error on step 4 (upsize and export video): NotImplementedError: A UTF-8 locale is required. Got ANSI_X3.4-1968 Any ideas? Thanks.
same
Amazing, glad I found your video from reddit.
Awesome, thank you 🙏
Hi, on the 4th step in the collab, I got the following issue:
NotImplementedError: A UTF-8 locale is required. Got ANSI_X3.4-1968 .How can I solve it?
Same
im in love with this woman
This was a really clever workflow. Great job.
Creative job!
Hi thank you for the amazing video. I just had a question because the program successfully ran all the way through step 3 but then stopped working at step 4 saying that "A UTF-8 locale is required. Got ANSI_X3.4-1968". I never learned any coding so I have no idea what this means. If you have any idea on how to fix this I would greatly appreciate it!
Same problem here
@@jeanettefromchina remains the same
-------------------------------------------------------------------------
NotImplementedError Traceback (most recent call last)
in ()
2 #@markdown ## Output file is upsized.mp4
3 get_ipython().run_line_magic('cd', '..')
----> 4 get_ipython().system('apt install ffmpeg')
5 get_ipython().system('apt install libmagic1 python3-yaml')
6 get_ipython().system('apt install libvulkan-dev')
2 frames
/usr/local/lib/python3.9/dist-packages/google/colab/_shell.py in system(self, *args, **kwargs)
96 kwargs.update({'also_return_output': True})
97
---> 98 output = _system_commands._system_compat(self, *args, **kwargs) # pylint:disable=protected-access
99
100 if pip_warn:
/usr/local/lib/python3.9/dist-packages/google/colab/_system_commands.py in _system_compat(shell, cmd, also_return_output)
451 # is expected to call this function, thus adding one level of nesting to the
452 # stack.
--> 453 result = _run_command(
454 shell.var_expand(cmd, depth=2), clear_streamed_output=False
455 )
/usr/local/lib/python3.9/dist-packages/google/colab/_system_commands.py in _run_command(cmd, clear_streamed_output)
165 locale_encoding = locale.getpreferredencoding()
166 if locale_encoding != _ENCODING:
--> 167 raise NotImplementedError(
168 'A UTF-8 locale is required. Got {}'.format(locale_encoding)
169 )
NotImplementedError: A UTF-8 locale is required. Got ANSI_X3.4-1968
That's actually pretty cool.
Great job, you won a new subscriber.
Incredible and so helpful 🙌💖💖💖 Thank you!
Really useful. I subbed and no idea how I didn't find you before.
Thank you so much 🙏🙏 appreciated
Thanks for great tutorial. It's only frustrating that I can't complete it because of annoying error. As you kindly mentioned in the end of the video, that we can post them here, I'm gonna use this opportunity. I run into this error: "EOFError: Ran out of input", on the step 3. Perhaps you can help me to fix it
It's because the first step didn't run correctly. The checkpoint file is no longer being hosted. If you Google, you can find the original Thin-Plate-Plate-Spline-Motion-Model and find the original Colab and replace the code in that cell with the ones to download from Yandex disk.
What a video!!! Thanks for sharing so much knowledge 🫶
I can bet that soon, in just Midjourney, you will be able to create a whole character. You will be able to generate and pick how character's face looks on both sides, as well as it's body. Then, you will be able to save that character with unique I'd to it and use it to generate scenes with it. Next big step it will be generating animations. God...I'm so excited!!! I always wanted to be a movie director! And it seems... that I will be...SOON! xD
I’m doing this pipeline atm for my next tutorial but …. sadly have to take it out of midjourney to do it. But you are so right, it will be very soon!
@@promptmuse Yes, ai is learning extremely fast.
Dont depend on ai for that, just use 3d softwares to all that stuff. Dont be limited by ai when you are free to do everything you want with 3d softwares, you wont like your characters face glitching in the most emotional scene in your mivie right?