Good luck training FLUX without a 48GB graphics card. I could train with an A6000 but not a 4090. Instead of paying $4,000 for one though, I use Kohya SS on cloud like Run Pod and Massed Compute. Cost like a couple dollars to train a FLUX LORA and you get all the advanced options from local.
I appreciate you highlighting this service. I still think I prefer the one on CIVITAI, because of the choice of progressive EPOCHs. I'd ultimately like you to post something on improving quality of trained LORAs under KOHYA-based trainers, like FluxGym and the like, especially how to better caption.
Thank you. The OpenArt training is mainly for users who want good results fast and with no learning curve. I will look into the KOHYA training. But i suppose there are already a lot of videos about that on youtube
Thank you for your videos! A few months ago, I downloaded your ComfyUI Super Flux template and turned it into a Killer Flux. I use the FLUX 8-steps model in your template, generating just 8 steps to create the initial image, which is then directly processed through the upscaler set to 2 steps. The results are incredible, producing stunningly high-quality images. Thank you so much for sharing all this amazing work-truly, thank you!
I was expecting it to be a website, the things that turned me away is that the prices aren't lower, in fact they're a lot more expensive than others. It's also censored for training and the only option is monthly, you can't pay for credits as you need it, like with Replicate and CivitAI's training. Not sure what you're asking with the difference between loras, the others are usable on other models but yes, Flux is the best and easiest to train.
The openart lora training is optimized to be easy and give fast results. Also they offer a full envoriment of tools to edit, inpaint, adjust and upscale your images on their page, while keeping a archive of all your creations and settings. that also adds value. While Replicate is nice and cheap, you are pretty much left alone after training the lora. So you need to know your way around the ai tools, while on openart all the tools are already setup for you
@@OlivioSarikas That may be true and would be an advantage if you have no other way, I'm not knocking the site for that. However, if all you want is lora training (which is the main topic of the video0, it isn't good value, and many have all the other OpenArt features locally. It's training that is the big time sink that many can't do locally. Even so, 4 models for $26 is terrible. PixelDojo offers 10 lora models for the month for $25, is basically one-click. 🙂
yes, sure. no argument there. I talked with them and they lowered the prices. But they also said they don't want to be the cheapest on the market. Which is totally ok. Not every brand compets with prices. Think for example about Starbucks. They are not cheap, but I have also never seen a empty Starbucks. There is always people lining up in their Stores
@@OlivioSarikas Definitely and being the cheapest doesn't always mean being the best. With sites like OpenArt and PixelDojo you are paying for convenience as much as anything else. As you say, it's an all-in-one solution. Have you ever covered PixelDojo? If you haven't, would that interest you. I almost joined earlier this year.
thanks for the tutorial , is it possible to create a photo that contains two lora models? for example i created model 1 and model 2 , can i generate an image that contains model 1 and model 2 in the same image?
Yes, you can, but you have to experiment with workflows, no one on civitai or similar sites does it. I made a workflow for that and I'm still improving it
did you try to create a image with one person, expand the image and inpaint a second person with a different lora? not sure if that works directly on openart at the moment, but i will suggest it to them
this video shows, that no side pic of the face is used, because it has a bad impact, it also shows, no pic of emotion, so the kissing mouth of a woman, and you will get a lora, where the mouth is not wide enough. All the other lora training videos are wrong, to make you think, that your information is stored the right way. But there is another point; as long as the recognition is not detecting, that the lora could be a well known adult star or celebrity, it will not implement "deformed layers". When I tested a lora with a very low vram card, a 6GB GTX 1060, the output was better, than with a 12GB RTX 3060, where much more NSFW-layers where loaded from the same lora.
not sure what you mean by hand drawn style lora. if you mean the sketch style image of myself, that is not a lora. i trained my face on photos of myself and some styled images of myself to upen up the lora to have artisitc styles
Abit OT but Am I the only one wondering whats taking so long with Controlnets for flux? especially for Forge UI which still has none... :( last i heard any talk of it was october, complete silence since then.
please reach out to them via their discord. They are a/b testing a new pricing right now and you might have ended up with a different sub than you clicked on
"Train your model" button is disabled but under it says "Your first model training is FREE", How can I train first model if button is disabled for non upgrade users? THIS IS A SCAM!!!!
Wow :D I'm really glad to see that they listened to the feedback and criticisms. It looks like they improved on the system for model training quite a lot from where it was on initial release and made themselves a much better value proposition against competitors in the space. Great work! And great video too :D
You can actually see in the video that it costs 2000 credits. What that means in dollar depends on the price you pay for the credits, which depends on the tier, discount or additional credits bought - so it's variable. Also price is relative to what you get. A cup of coffee can cost $1 or $8 - but without knowing what else is included in the full package, that price won't tell you if it is cheap or expensive. So I left out the pricing of this video, because it's misleading to the actual value of a page
Finally someone that teach this amazing topic! In your opinion can we caombine 2 different model? For example, a man, and a bag, to make photo shooting with the same model and the same product. Thanks for you amazing tutorial
on openart right now you can only use 1 lora at a time, but in comfyui you can use multiple flux loras together. also look for Kolors Virtual Try On google. i already covered that in one of my news videos
Well, in the sense that Heineke is just a different Budweiser - they are both beers, but they are not the same thing. Plus you need to take into consideration the enviroment this is in with the editing, inpainting, upscaling, canvas and a lot more. Plus this is made to be easy and fast
They come off really iffy for me. I think using a look alike chevron come off scammy. I like Civitai for making loras. And I think you have to pay to even try, and I don’t like giving out my CC number to try
@@KimiMorgam older models have a lot of plus points because they have been around for longer, but Flux can create very detailed images and text like no older model can. That said, I also still love the older models
![I.N "HALLUCINATION" | [Stray Kids : SKZ-PLAYER]](http://i.ytimg.com/vi/n5B5q1Hwt_U/mqdefault.jpg)
💾 GET A 50% Discount right now on all Plans:
openart.ai/
link gives "page not found"
chatGPT tells me it's because you copy pasted a zero-width space along with it at the end.
is this better then pulid ?
Hi
Hey would you know of a "one click lora trainer" that would work with the pony checkpoint?
local or nothing :D
This!
word up!
TRUE!!
Good luck training FLUX without a 48GB graphics card. I could train with an A6000 but not a 4090. Instead of paying $4,000 for one though, I use Kohya SS on cloud like Run Pod and Massed Compute. Cost like a couple dollars to train a FLUX LORA and you get all the advanced options from local.
Local or nothing!
What happens to the images you upload from yourself privacy wise?
can we Download trained lora/model?
I appreciate you highlighting this service. I still think I prefer the one on CIVITAI, because of the choice of progressive EPOCHs.
I'd ultimately like you to post something on improving quality of trained LORAs under KOHYA-based trainers, like FluxGym and the like, especially how to better caption.
Thank you. The OpenArt training is mainly for users who want good results fast and with no learning curve. I will look into the KOHYA training. But i suppose there are already a lot of videos about that on youtube
Thank you for your videos! A few months ago, I downloaded your ComfyUI Super Flux template and turned it into a Killer Flux.
I use the FLUX 8-steps model in your template, generating just 8 steps to create the initial image,
which is then directly processed through the upscaler set to 2 steps.
The results are incredible, producing stunningly high-quality images. Thank you so much for sharing all this amazing work-truly, thank you!
awesome! can you contact me through my email on the youtube about of my channel. please write "killer flux" into the email topic :)
@@OlivioSarikas I just sent you an email with the workflow. Have a great day! 🙂
@@ZanetwiceTV awesome, thank you
What about FluxGym?
Can you train products that have large and small writing on them?
that's just an ad for a website???? what's the difference with other lora, with flux the lora is one of the best.
I was expecting it to be a website, the things that turned me away is that the prices aren't lower, in fact they're a lot more expensive than others. It's also censored for training and the only option is monthly, you can't pay for credits as you need it, like with Replicate and CivitAI's training.
Not sure what you're asking with the difference between loras, the others are usable on other models but yes, Flux is the best and easiest to train.
The openart lora training is optimized to be easy and give fast results. Also they offer a full envoriment of tools to edit, inpaint, adjust and upscale your images on their page, while keeping a archive of all your creations and settings. that also adds value. While Replicate is nice and cheap, you are pretty much left alone after training the lora. So you need to know your way around the ai tools, while on openart all the tools are already setup for you
@@OlivioSarikas That may be true and would be an advantage if you have no other way, I'm not knocking the site for that. However, if all you want is lora training (which is the main topic of the video0, it isn't good value, and many have all the other OpenArt features locally. It's training that is the big time sink that many can't do locally.
Even so, 4 models for $26 is terrible. PixelDojo offers 10 lora models for the month for $25, is basically one-click. 🙂
yes, sure. no argument there. I talked with them and they lowered the prices. But they also said they don't want to be the cheapest on the market. Which is totally ok. Not every brand compets with prices. Think for example about Starbucks. They are not cheap, but I have also never seen a empty Starbucks. There is always people lining up in their Stores
@@OlivioSarikas Definitely and being the cheapest doesn't always mean being the best. With sites like OpenArt and PixelDojo you are paying for convenience as much as anything else. As you say, it's an all-in-one solution.
Have you ever covered PixelDojo? If you haven't, would that interest you. I almost joined earlier this year.
thanks for the tutorial , is it possible to create a photo that contains two lora models? for example i created model 1 and model 2 , can i generate an image that contains model 1 and model 2 in the same image?
Yes, you can, but you have to experiment with workflows, no one on civitai or similar sites does it. I made a workflow for that and I'm still improving it
@adrianmunevar654 is there a tutorial for this bro?
@@pantoomeme I haven't seen one yet :/ They just recommend using Photoshop and stuff, but that's too complicated
did you try to create a image with one person, expand the image and inpaint a second person with a different lora? not sure if that works directly on openart at the moment, but i will suggest it to them
@@OlivioSarikas thanks man 😍
whats the least number of images I'd need for a decent lora
I would say 4-5 images
locally trained in Flux Gym? Around 10.
@@AIKnowledge2Go is more better? like 15 maybe
this video shows, that no side pic of the face is used, because it has a bad impact, it also shows, no pic of emotion, so the kissing mouth of a woman, and you will get a lora, where the mouth is not wide enough. All the other lora training videos are wrong, to make you think, that your information is stored the right way. But there is another point; as long as the recognition is not detecting, that the lora could be a well known adult star or celebrity, it will not implement "deformed layers". When I tested a lora with a very low vram card, a 6GB GTX 1060, the output was better, than with a 12GB RTX 3060, where much more NSFW-layers where loaded from the same lora.
thank you.
could you please share that "hand drawn style lora"?
not sure what you mean by hand drawn style lora. if you mean the sketch style image of myself, that is not a lora. i trained my face on photos of myself and some styled images of myself to upen up the lora to have artisitc styles
can I sale the LORA?
do you have any tips on training a style of clothing
OpenArt has a Style Traning Mode - But personally I have not trained any clothing items
I tried training clothing. It won't be as good as face. Use captions. The model won't train well without captions.
Abit OT but Am I the only one wondering whats taking so long with Controlnets for flux? especially for Forge UI which still has none... :( last i heard any talk of it was october, complete silence since then.
Right? taking forever. We really need them controlnets
how to download the model?
It's in the video. Once you've trained it, click the three dots in the top right of the preview image.
Its not true, training is not free! You have to pay even tho they lie 1 training is free on website.
please reach out to them via their discord. They are a/b testing a new pricing right now and you might have ended up with a different sub than you clicked on
GIGA CHAD OLIVIO
The big question is: did he skip leg day? XD
very good, another great video
thank you :)
i sign up free with free credit and it wont let me train my lora models says i have to pay
That's because a training needs 2000 credits. Runnung servers costs money. Just like you had to pay for your computer and your electricity at home
"Train your model" button is disabled but under it says "Your first model training is FREE", How can I train first model if button is disabled for non upgrade users? THIS IS A SCAM!!!!
@@DivinityIsPurity dude, relax. Ever heard of technical issues. Not everything is a evil plot against you
Wow :D I'm really glad to see that they listened to the feedback and criticisms. It looks like they improved on the system for model training quite a lot from where it was on initial release and made themselves a much better value proposition against competitors in the space.
Great work! And great video too :D
Was hard work to convince them. But now it's pretty cool 🥰
- How many Olivio do you want?
- Yes!
There can never be enough Olivio. Especially the Giga Chat Version XD
Have wonderful holidays Olivio! ^-^
I'm not on holiday - unless you mean happy christmas ;) I wish i had some free time ;) - but thank you :)
People will come to them with free limits?
Bearded, not beared.
But it shows that Diffusion models aren't too sticky about what you write in your prompt.
😆
you should have mentioned in the video how much it actually cost to train it.
You can actually see in the video that it costs 2000 credits. What that means in dollar depends on the price you pay for the credits, which depends on the tier, discount or additional credits bought - so it's variable. Also price is relative to what you get. A cup of coffee can cost $1 or $8 - but without knowing what else is included in the full package, that price won't tell you if it is cheap or expensive. So I left out the pricing of this video, because it's misleading to the actual value of a page
Good!
thank you
Finally someone that teach this amazing topic! In your opinion can we caombine 2 different model? For example, a man, and a bag, to make photo shooting with the same model and the same product. Thanks for you amazing tutorial
on openart right now you can only use 1 lora at a time, but in comfyui you can use multiple flux loras together. also look for Kolors Virtual Try On google. i already covered that in one of my news videos
Can you do text to videos
they have a area with workflows and some of them are web hosted, but I don't think video ai is their focus right now
No local, no fun, haha. Just kidding-it's a good tool for saving time.
true. i started to use more online tools too. especially with these giant models that don't even run local anymore. models blow up so fast these days
It's a pity that the skin texture is not that good though 😐
i think its a neat online solution... but to be honest, i would rather rent a online cloud pc desktop and do the heavy lifting there.
so this is just like the civitai trainer on a different site? I'm really not that impressed :/
Well, in the sense that Heineke is just a different Budweiser - they are both beers, but they are not the same thing. Plus you need to take into consideration the enviroment this is in with the editing, inpainting, upscaling, canvas and a lot more. Plus this is made to be easy and fast
No… it is actually worse and more expensive!
advertisement
They come off really iffy for me. I think using a look alike chevron come off scammy. I like Civitai for making loras. And I think you have to pay to even try, and I don’t like giving out my CC number to try
I stopped the video at 5:13, after one click.
local or nothing
Sd1.5 sadly still best base model.
I dont understand why people are loving flux... slower... much weight than others and dont understand prompt weights...?
what are the vantages?
i think pony realistic are very better than flux
@@KimiMorgam older models have a lot of plus points because they have been around for longer, but Flux can create very detailed images and text like no older model can. That said, I also still love the older models
@@OlivioSarikas thanks, my major problem with flux is the speeds... is not for who have 3060... =/
First
yay! :)
I came as fast as I could to inform you that your title has a typo! Its training not traning! :D
Thank you very much!!!
Honestly a wasteof time… both the platform and the video
talk about animatedifff
I will look into that. But i highly suggest you look up Purz on youtube
💾 GET A 50% Discount right now on all Plans:
openart.ai/