Another good video Bob. I just want to share what I've been doing lately to clone myself. Since I have FaceFusion 3.0 and Dzine also has face swapping, I make a photo of a person who's built like me doing stuff and then I simply put my face on it. I found that I have more control over my images that way. I love this stuff !!!
I love the creative ways you put things together to create what your vision is!! This OpenArt model training with Flux is really something. I would say it blew my mind, but I'm sick of that hype about everything! But seriously, when I ran the model with my ugly face for the first time and the images popped up... I sort of didn't know what to do. I just sat there looking at it because (after having used so many platforms and models) I've never had one that just worked the first time (really well) and I didn't have to make all kinds of tweaks to get it to work. I share all that to say.... yeah, I love this stuff too!!
Thanks Bob, I was wondering how to train an AI for a whole person such as an influencer (not only the face). I have thousands of real pictures and want the influencer to appear in new scenarios, new dresses, new locations etc.
If you have real pictures, I think you'll find this to be a breeze! It's a little tougher when you're trying to 'invent' a person that doesn't exist and you have to create the training dataset 😳 When I fed this thing some images of me, I was shocked at how well it worked right away without much fiddling. And I'm not just saying that because I work with OpenArt... If it was "meh", I wouldn't sing it's praises. 😊
You're right, you can only use one model for an image generation. You can use image guidance with a model, either image to image, composition reference, or sketch reference. But that's not the same as what you're asking for.
@@grease112 At present, you wouldn't be able to generate with 2 consistent characters in the same image. The closest I can think of is 2 separate images and then try to blend them in the editor with the blend board.
I don't think so. Workflows, over in the ComfyUI workflow section of OpenArt can be downloaded (and some run on the cloud through OpenArt) but it does not appear that you can download the models created in the main section of OpenArt.
@excelerator good to know. Not for me too many online platforms in closed systems so you can keep using their platform. I run local as well so I want the ability to have my model or Lora online for easy generation a d also on my personal computer. At any point these systems could fail and go away with all your years of work. No Thanks! Just like whats been going on over at Playground AI.
with so many competitors & everybody screaming I'm The BeST, trust me bro when I say I've researched them ALLLLLLLLL. unless you really need smut& wanna upload your own images, firefly is the #1 best image generator, seems many don't want to admit that. on so many levels THIS is like the firefly version I've been hoping for. I hope NO OTHER CONTENT CREATORS REALIZE IT's potential
That's an interesting perspective - I don't think I've heard anyone else say firefly is the best image generator. I'm not disagreeing with you at all (I think the *best* depends on what you want to create and which platform you're most comfortable with) but it's great that you're speaking highly of Firefly when it doesn't get a lot of attention, especially in the noise of all the platforms that get the most attention. I appreciate you sharing this.
If I'm understanding your question correctly, no, not in one model/workflow. The fine-tuning uses one model and there isn't a way to configure it to add a LoRa for the realism style in the mix. Thinking out loud (and maybe I'm a dummy)... I'm thinking the way to achieve this look might be by styling the training images before training on them - but then again, I doubt the style would come through well if you're doing a face or character fine-tuning. So, maybe doing an image-to-image with your generated image to impart the style. If you're comfortable with a little bit more complication in the setup, you can create your own ComfyUI workflow - even in OpenArt.
Hi, what would you recommend for training model with my own pictures? Open art? Maybe Replicate? I need good quality, easy using with reasonable price :) and thanks for the tutorial btw!
OpenArt, without hesitation. Not just because they sponsor some videos of mine either. Their new training process is the simplest and the best results of any platform I've ever tried.
@AICodziennieI'm sorry you're having problems with it. When you say "creating other people" do you mean that you're generating an image with your trained models of your face and it's giving you an image that doesn't look like you? Or, that it's giving you an image that has you in it, but it also puts other people in the image with you, even when you didn't prompt for it to have anyone except you? If generated images don't look like you, take a look at the images you used to train the model an make sure your face is clear and not heavily shadowed (shadows are fine, as long as it's not completely covering half your face and obscuring all detail.) OpenArt's model training is the most flexible I've used, so the images don't have to be perfect - they just need to have you clearly visible and obviously the subject. Also, try moving the influence slider up to like 1.25. adjust as necessary. If you're getting extra people in your generated images that you didn't prompt for, that sounds like you have other people in your training images. When training a face or character model, use training images that have one person or character. If you want to use images that have other people, erase them out or use the "Remove" or in paint in the OpenArt editor to remove other people from the image. Cropping in on the subject (you) helps too. Also, I'm sure you did this right and checked it, but make sure you're selecting "face" for the model training type. I hope this helps. Without seeing what you're working with or getting, I'm doing a lot of guessing here. So I'm sorry if these tips don't apply.
@@excelerator I mean each time openart makes ANY other person as...myself. Also women, what is kind of funny. Have a look: aicodziennie.pl/wp-content/uploads/2025/01/openart-image_iIFCti7L_1735648357604_raw.jpg
OpenArt just revamped their upscaling options and honestly, it's a bit confusing right now. The ultimate upscale now has 'precise', 'refined', or 'creative' modes (how much you want the image to change) but the level of upscale isn't adjustable. But, from a generated image in your creation history or albums, on the right near the bottom, you should have "Choose Upscale Mode" right above Ultimate Upscale. Click the dropdown and it has options for 2X, 2X + Face, 4X, 4X + Face, and 4X + Anime.
Training is only available on paid plans. OpenArt is changing their subscription plans. Previously, you were allowed a fixed number of trained models based on your plan. If you're an existing subscriber, you may still have that plan. The new plans don't have a fixed number but use a number of credits.
Hi bob. I am new user for openartai. I am struggling to create my own face character. And the ai always make my picture become slim, i don’t like it.. can you make video for how to train for a face and body character, please..
That's odd. I can't say I've had the issue of it making me skinny (I wish!) But, I'm glad to do a video on training a face model. I'll get it on the list. In the meantime, I find that when I'm using a face model in OpenArt, it usually does very well but sometimes I can get even better results by describing the subject with a little detail in my prompt. For instance, I can use my face model and prompt "a man sitting on a park bench" or I can say "a man with short blonde hair with dark roots wearing clear-framed glasses sitting on a park bench" and that can really dial in the output results. Also, experiment with the weight slider - that makes a huge difference in how much the generated image matches the trained subject. Too high, it will want to use a strong image from the training dataset, too low and it starts to diverge notably from the original character. Somewhere in the 1.0 to 1.4 gets me to the result I like best.
Those are rare, but do exist. Like ArtSpace (I have a video on it) that has a lifetime deal. It's OK, especially for beginners, but you won't get all the features and models and things like training custom models that OpenArt has. Another option is something like Replicate where you deposit like $10 and then cost is deducted as you use it. Tons of models and generations can cost anywhere from less than a penny to close to 10 cents each. But, the interface isn't as pretty and it's better suited for techies.
Excellent video, very nicely explained. I've been pretty impressed with the training so far, it's definitely levelling up!
Thanks for checking it out!!! I love having a celebrity visit my channel 🤩
@@excelerator haha, I don't know about that but I very much appreciate the sentiment :) Doing awesome man!
Thanks Bob for the detailed explanations! The monster model is soooo cute! The model of yourself is impressive.
What, my model isn't cute? 🤣🤣🤣. Seriously, I was stunned at how well it performed with no hassle.
loving this 🔥🔥and the candy model will be my go to for the coming holidays 😁
Nice work enjoyed it
Awesome! I'm happy to hear that!! 😁
Thanks for the great video!
I'm glad you liked it.
Another good video Bob. I just want to share what I've been doing lately to clone myself. Since I have FaceFusion 3.0 and Dzine also has face swapping, I make a photo of a person who's built like me doing stuff and then I simply put my face on it. I found that I have more control over my images that way. I love this stuff !!!
I love the creative ways you put things together to create what your vision is!! This OpenArt model training with Flux is really something. I would say it blew my mind, but I'm sick of that hype about everything! But seriously, when I ran the model with my ugly face for the first time and the images popped up... I sort of didn't know what to do. I just sat there looking at it because (after having used so many platforms and models) I've never had one that just worked the first time (really well) and I didn't have to make all kinds of tweaks to get it to work.
I share all that to say.... yeah, I love this stuff too!!
Thanks Bob, I was wondering how to train an AI for a whole person such as an influencer (not only the face). I have thousands of real pictures and want the influencer to appear in new scenarios, new dresses, new locations etc.
If you have real pictures, I think you'll find this to be a breeze! It's a little tougher when you're trying to 'invent' a person that doesn't exist and you have to create the training dataset 😳 When I fed this thing some images of me, I was shocked at how well it worked right away without much fiddling. And I'm not just saying that because I work with OpenArt... If it was "meh", I wouldn't sing it's praises. 😊
great video! How do you use a style, a character, and an object model in one generated image? It looks like it only gives you the option for 1
You're right, you can only use one model for an image generation. You can use image guidance with a model, either image to image, composition reference, or sketch reference. But that's not the same as what you're asking for.
@excelerator how would you be able to out 2 consistent characters into a scene?
@@grease112 At present, you wouldn't be able to generate with 2 consistent characters in the same image. The closest I can think of is 2 separate images and then try to blend them in the editor with the blend board.
Can you download the model safetensor so that you can use it on other platforms?
I don't think so. Workflows, over in the ComfyUI workflow section of OpenArt can be downloaded (and some run on the cloud through OpenArt) but it does not appear that you can download the models created in the main section of OpenArt.
@excelerator good to know. Not for me too many online platforms in closed systems so you can keep using their platform. I run local as well so I want the ability to have my model or Lora online for easy generation a d also on my personal computer. At any point these systems could fail and go away with all your years of work. No Thanks!
Just like whats been going on over at Playground AI.
with so many competitors & everybody screaming I'm The BeST, trust me bro when I say I've researched them ALLLLLLLLL. unless you really need smut& wanna upload your own images, firefly is the #1 best image generator, seems many don't want to admit that. on so many levels THIS is like the firefly version I've been hoping for. I hope NO OTHER CONTENT CREATORS REALIZE IT's potential
That's an interesting perspective - I don't think I've heard anyone else say firefly is the best image generator. I'm not disagreeing with you at all (I think the *best* depends on what you want to create and which platform you're most comfortable with) but it's great that you're speaking highly of Firefly when it doesn't get a lot of attention, especially in the noise of all the platforms that get the most attention. I appreciate you sharing this.
What's the website for firefly??
can you use openart but also apply the realism lora on top? so the pictures look amateur?
If I'm understanding your question correctly, no, not in one model/workflow. The fine-tuning uses one model and there isn't a way to configure it to add a LoRa for the realism style in the mix. Thinking out loud (and maybe I'm a dummy)... I'm thinking the way to achieve this look might be by styling the training images before training on them - but then again, I doubt the style would come through well if you're doing a face or character fine-tuning. So, maybe doing an image-to-image with your generated image to impart the style. If you're comfortable with a little bit more complication in the setup, you can create your own ComfyUI workflow - even in OpenArt.
Hi, what would you recommend for training model with my own pictures? Open art? Maybe Replicate? I need good quality, easy using with reasonable price :) and thanks for the tutorial btw!
OpenArt, without hesitation. Not just because they sponsor some videos of mine either. Their new training process is the simplest and the best results of any platform I've ever tried.
@@excelerator Thanks a lot for your advise!
Hi again, do you have any hint to avoid creating other people on the pic that look exactly like myself? I try and try and can not deal with it :(
@AICodziennieI'm sorry you're having problems with it. When you say "creating other people" do you mean that you're generating an image with your trained models of your face and it's giving you an image that doesn't look like you? Or, that it's giving you an image that has you in it, but it also puts other people in the image with you, even when you didn't prompt for it to have anyone except you?
If generated images don't look like you, take a look at the images you used to train the model an make sure your face is clear and not heavily shadowed (shadows are fine, as long as it's not completely covering half your face and obscuring all detail.) OpenArt's model training is the most flexible I've used, so the images don't have to be perfect - they just need to have you clearly visible and obviously the subject.
Also, try moving the influence slider up to like 1.25. adjust as necessary.
If you're getting extra people in your generated images that you didn't prompt for, that sounds like you have other people in your training images. When training a face or character model, use training images that have one person or character. If you want to use images that have other people, erase them out or use the "Remove" or in paint in the OpenArt editor to remove other people from the image. Cropping in on the subject (you) helps too.
Also, I'm sure you did this right and checked it, but make sure you're selecting "face" for the model training type.
I hope this helps. Without seeing what you're working with or getting, I'm doing a lot of guessing here. So I'm sorry if these tips don't apply.
@@excelerator I mean each time openart makes ANY other person as...myself. Also women, what is kind of funny.
Have a look: aicodziennie.pl/wp-content/uploads/2025/01/openart-image_iIFCti7L_1735648357604_raw.jpg
Upscale level option (2k or 4k) is not showing while upscaling images, could you please assist
OpenArt just revamped their upscaling options and honestly, it's a bit confusing right now. The ultimate upscale now has 'precise', 'refined', or 'creative' modes (how much you want the image to change) but the level of upscale isn't adjustable. But, from a generated image in your creation history or albums, on the right near the bottom, you should have "Choose Upscale Mode" right above Ultimate Upscale. Click the dropdown and it has options for 2X, 2X + Face, 4X, 4X + Face, and 4X + Anime.
@excelerator thanks for the reply
lit stuff
🔥
you have to pay to train the model right ?
Training is only available on paid plans. OpenArt is changing their subscription plans. Previously, you were allowed a fixed number of trained models based on your plan. If you're an existing subscriber, you may still have that plan. The new plans don't have a fixed number but use a number of credits.
Hi bob. I am new user for openartai. I am struggling to create my own face character. And the ai always make my picture become slim, i don’t like it.. can you make video for how to train for a face and body character, please..
That's odd. I can't say I've had the issue of it making me skinny (I wish!) But, I'm glad to do a video on training a face model. I'll get it on the list. In the meantime, I find that when I'm using a face model in OpenArt, it usually does very well but sometimes I can get even better results by describing the subject with a little detail in my prompt. For instance, I can use my face model and prompt "a man sitting on a park bench" or I can say "a man with short blonde hair with dark roots wearing clear-framed glasses sitting on a park bench" and that can really dial in the output results. Also, experiment with the weight slider - that makes a huge difference in how much the generated image matches the trained subject. Too high, it will want to use a strong image from the training dataset, too low and it starts to diverge notably from the original character. Somewhere in the 1.0 to 1.4 gets me to the result I like best.
Man I wish there was an AI that wasn't subscription based and you could just buy it with a $100> one time purchase.
Those are rare, but do exist. Like ArtSpace (I have a video on it) that has a lifetime deal. It's OK, especially for beginners, but you won't get all the features and models and things like training custom models that OpenArt has.
Another option is something like Replicate where you deposit like $10 and then cost is deducted as you use it. Tons of models and generations can cost anywhere from less than a penny to close to 10 cents each. But, the interface isn't as pretty and it's better suited for techies.