I'm glad because there isn't much information about Diffusion Bee. I'm Japanese and I'm glad that I found your video. I would like to see detailed training and videos showing how to use LoRA Tools. thank you.
Hello I'm glad you found my channel. I am waiting on the newest release so I can do a longer format video covering the app from beginning to end vs breaking it up in multiple parts as there's a lot of information that can be missed. The LoRa tool is definitely a powerful asset and I will try to cover it as closely as I can.
I just downloaded this today after finding out the new term ‘local llm’ … what a world. It’s difficult to understand all of it, like lora’s…. I will try to study a bit more. What a time be alive to be able to do all this.
Have you been trying to get the new Flux model to work on your Mac? If so, have you had any success? I get all the way up to the sampler box in ComfyUI and then get the old "BFloat16 is not supported on MPS" error.
I am pleasantly surprised. After it seemed for a long time that nothing really useful was going to happen, things are now happening in leaps and bounds. I had almost given up on DB, but the update looks like fun.
Thanks for posting these videos. When training a person have you found the optimal input? You had mentioned 30 images. What percentage should be closeup vs. medium vs wider shot? Still 100 iterations per image? What should go into the txt file for a person? Is it higher the model size the higher the quality/resemblance? Sorry for all of the questions but you seem to be only person online who talks about Diffusion Bee.
I wish there was one setting that worked every time however, this is not the case I've gotten varying results with different settings each time with the same data set. i covered some of this on in the preview previous video on captioning as captioning will play a pretty heavy role in how stable diffusion understands your data set th-cam.com/video/EoXyhgf6pCg/w-d-xo.html When training another person, the app mentions to have at least 12 images I found anywhere from 30 to 40 images seem to be pretty good keep in mind have images that are close up, far away, etc. with varying colors and backgrounds for a flexible model. Yes I would still probably use the hundred iterations per image so if you have 40 images it would be 40×100. No, the higher model size does not necessarily mean quality and resemblance ...again a lot of things play in and factor in how the overall model will look choosing the right base model will also affect how the model will look in the end as well if you're going for something realistic than you want to aim for a realistic base model
@@obscuremusictabs5927 no problem. There is a newer pre release version available however, I didn't notice too many new features other than being able to import from hugging face
@@theprotoart Is it possible to put two separate people into a model? 10 photos of one person and 10 of another as long as they are distinctly labelled in the text files? I ask instead of trying just because it takes so long to train.
@@obscuremusictabs5927 are you trying to be able to blend between the two different people? If so, I would probably create a LoRa versus a full dream booth model.
I hope Diffusionbee adds the new higher-quality motion video being demonstrated now. I don't expect Sora level yet, but the open source versions are getting very good.
App Store? The link is in the description it’s on github.com/divamgupta/diffusionbee-stable-diffusion-ui/releases/tag/2.5.0 what version of Mac are you running? might be a compatibility issue if you have an older OS make sure you’re running at least Ventura 13.2.1 or later
I have DiffusionBee version 1.7.4, how to I update it to 2.5? Do I have to download it again and delete my current version? Or can I update my already installed version? Please help.
Hello, you have to download it through the GitHub. I provided the link for this already, The website itself unfortunately isn’t up-to-date so that version is still the older version.(1.7.4) you don’t necessarily need to delete the old one if you do not want to just change the name in the apps folder itself. It will require you within the app to download some new models and whatever dependencies it requires so be sure that you don’t have a VPN blocking or any malware or antivirus things that may potentially stop it from downloading the new models this is a known thing if you were to go to the Discord, this is explained thoroughly.
So to answer your first question, get it from the GitHub. for your second question you don’t need to delete the version you have. You can keep that as well, although it’s not needed.
True there are a lot of features I’m looking forward to and some basic ones. I’m still hoping for like the option to train just a lora file or extract it from previous dream booth models
@@theprotoart Are you talking about Diffusionbee or Stable Diffusion? Either way neither of them are as solo projects. Divam even credits the people who helped him with db on his github.
hello, I’m not sure if you got to watch the video in it’s entirety but the same way you were able to import models in the previous version now you can import LoRa in the same tab.
@@theprotoart that was indeed the problem, for some strange reason the default was an old version... must be all this artificial intelligence displacing organic intelligence lol
I'm glad because there isn't much information about Diffusion Bee. I'm Japanese and I'm glad that I found your video. I would like to see detailed training and videos showing how to use LoRA Tools. thank you.
Hello I'm glad you found my channel. I am waiting on the newest release so I can do a longer format video covering the app from beginning to end vs breaking it up in multiple parts as there's a lot of information that can be missed. The LoRa tool is definitely a powerful asset and I will try to cover it as closely as I can.
I just downloaded this today after finding out the new term ‘local llm’ … what a world. It’s difficult to understand all of it, like lora’s…. I will try to study a bit more. What a time be alive to be able to do all this.
@@monkeyfish227 if you haven’t already check out the latest video as I cover everything from beginning to end
Have you been trying to get the new Flux model to work on your Mac? If so, have you had any success? I get all the way up to the sampler box in ComfyUI and then get the old "BFloat16 is not supported on MPS" error.
@@obscuremusictabs5927 I haven’t but there’s a good amount of people detailing it on Reddit
I am pleasantly surprised. After it seemed for a long time that nothing really useful was going to happen, things are now happening in leaps and bounds. I had almost given up on DB, but the update looks like fun.
It really is a powerful little app with a lot of useful features. There is still plenty of room for improvement ✊🏾
Thank you very much for sharing your knowledge, greetings from Mexico!
No problem I’m glad you found it helpful, and greetings ✊🏾 🇲🇽
Thanks for posting these videos. When training a person have you found the optimal input? You had mentioned 30 images. What percentage should be closeup vs. medium vs wider shot? Still 100 iterations per image? What should go into the txt file for a person? Is it higher the model size the higher the quality/resemblance? Sorry for all of the questions but you seem to be only person online who talks about Diffusion Bee.
I wish there was one setting that worked every time however, this is not the case I've gotten varying results with different settings each time with the same data set. i covered some of this on in the preview previous video on captioning as captioning will play a pretty heavy role in how stable diffusion understands your data set th-cam.com/video/EoXyhgf6pCg/w-d-xo.html When training another person, the app mentions to have at least 12 images I found anywhere from 30 to 40 images seem to be pretty good keep in mind have images that are close up, far away, etc. with varying colors and backgrounds for a flexible model. Yes I would still probably use the hundred iterations per image so if you have 40 images it would be 40×100. No, the higher model size does not necessarily mean quality and resemblance ...again a lot of things play in and factor in how the overall model will look choosing the right base model will also affect how the model will look in the end as well if you're going for something realistic than you want to aim for a realistic base model
@@theprotoart Thank you for such a quick response.
@@obscuremusictabs5927 no problem. There is a newer pre release version available however, I didn't notice too many new features other than being able to import from hugging face
@@theprotoart Is it possible to put two separate people into a model? 10 photos of one person and 10 of another as long as they are distinctly labelled in the text files? I ask instead of trying just because it takes so long to train.
@@obscuremusictabs5927 are you trying to be able to blend between the two different people? If so, I would probably create a LoRa versus a full dream booth model.
I hope Diffusionbee adds the new higher-quality motion video being demonstrated now. I don't expect Sora level yet, but the open source versions are getting very good.
Could always put in a request on the discord
If you train DiffusionBee on your own images, are those images available to the public or are they kept locally?
No, everything is local. 👍🏾
@@theprotoart Right on. Thanks. I'll give it a try.
I use drawthings too but increasingly like diffuisonBee more simply for the simplicity of the UI
Same !!
The App Store doesn't offer me this update so this version must not run on my model of Mac. 😢
App Store? The link is in the description it’s on github.com/divamgupta/diffusionbee-stable-diffusion-ui/releases/tag/2.5.0 what version of Mac are you running? might be a compatibility issue if you have an older OS make sure you’re running at least Ventura 13.2.1 or later
I have DiffusionBee version 1.7.4, how to I update it to 2.5? Do I have to download it again and delete my current version? Or can I update my already installed version? Please help.
Hello, you have to download it through the GitHub. I provided the link for this already, The website itself unfortunately isn’t up-to-date so that version is still the older version.(1.7.4) you don’t necessarily need to delete the old one if you do not want to just change the name in the apps folder itself. It will require you within the app to download some new models and whatever dependencies it requires so be sure that you don’t have a VPN blocking or any malware or antivirus things that may potentially stop it from downloading the new models this is a known thing if you were to go to the Discord, this is explained thoroughly.
So to answer your first question, get it from the GitHub. for your second question you don’t need to delete the version you have. You can keep that as well, although it’s not needed.
@@theprotoart Ok. Thanks..
No problem just look at the description it’s all there for you
Now SD 3 is out; fingers crossed for Diffusionbee to use it.
True there are a lot of features I’m looking forward to and some basic ones. I’m still hoping for like the option to train just a lora file or extract it from previous dream booth models
SD3 wasn't released when you wrote this and it still isn't available to the public...
@@697_ that’s correct. I believe it’s like one developer you’re more than welcome to talk to him on discord
@@theprotoart Are you talking about Diffusionbee or Stable Diffusion? Either way neither of them are as solo projects. Divam even credits the people who helped him with db on his github.
Can we run stability cascade on Diffusion Bee?
No not at the moment, hopefully the dev includes that option
How you have imported lora?
hello, I’m not sure if you got to watch the video in it’s entirety but the same way you were able to import models in the previous version now you can import LoRa in the same tab.
At about 1 minute in I show you
This interface doesn't look anything like what i just installed...
Be sure to download the latest from GitHub
@@theprotoart that was indeed the problem, for some strange reason the default was an old version... must be all this artificial intelligence displacing organic intelligence lol
@@m12652 😅 yeah they haven’t updated the website for a while so there has been a lot of confusion from others too