Видео

No problem, thank you for watching

what versions are you trying to install? If they are XL versions not available from the in app download they are not usable just yet the dev already mentioned this long ago and the update should be coming. I don’t know when I’m not involved in that. I would join the discord to ask any other questions you may have for troubleshooting.

Much appreciated ✊🏾

That’s awesome!! I’m glad it helped

Thank you for watching ✊🏾

Settings for? There is no one set of settings for a perfect model for LoRa or for a safesensor, etc.. It’s all trial and error.

@@theprotoart Sorry, I meant I was trying to find out what each setting is. Like, what's a bucket. What's a checkpoint? What's an epoch? What's an alpha? Etc.

@@WifeWantsAWizard are you using it on a desktop computer because if you hover over it it a tells you this info already

@@theprotoart Well. This has been average. Thanks for your time.

around 2:34 I answered this already 😉

No problem I appreciate you taking the time to check out the video ✊🏾

Are you unable to access them or … you using the mobile version?.. might be a slight difference in how to get there..

Sorted @@theprotoart cheers

Same !!

Yeah, it could be a little frustrating having to deal with that aspect. I think you’re allowed 100 credit by liking photos each day, an additional 50 in the generation tab and then you can also get 10 more credit by adding people so total you can acquire around 160 each day if you choose the free route. (you could however make many accounts cough cough)

Make two account and do the daily rewards you would have 500 in on day if you donate to your main account

For sure if you have any tips or tricks, let me know. I’m planning to make an updated video whenever the Dev drops a new update( will probably be more of a long format video covering the entire app from front to back as I’ve broken it up into a lot of pieces before)

You can’t m. DB is unique and uses its own unique format. This is all answered on the Discord.

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

Much appreciated. yeah, there’s a lot of troubleshooting to get this thing to work when I was trying to figure it out myself 😅

not only have I already provided a link to an updated version with an entirely new video I replied to the same kind of comment a few different times already. please look at the description ✊🏾

No problem

What version are you installing? The latest is 2.5.1

Double check with the github. doesn’t really keep up with updating the website consistently

@@theprotoart 1.7.4.... that explains it, i just clicked the first download button. Will have another look... thanks 👍

@@m12652 no problem. You’re not the first one who’s encounter this unfortunately as they don’t seem to update the website that often even when putting on the new releases

@@m12652 there’s been a lot of changes since then, so you may need to watch a few of the tutorials 😅

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

Hey, just a few comments ago I’ve kind of addressed this already with an alternative option ruclips.net/video/JpdNv5SHNWc/видео.htmlsi=HveoOm74mvYsRYiN

@@theprotoart This link Autostrip its not free but work fine . Have only 10 minute on moth if i not wrong.

@@IndigaVP not sure if there are many free ones left that let you process long audio files unfortunately 😅 might need to break it up into small sections and piece it back together in post

Do what?

thanks

Idk 🤷🏾‍♂️

I don’t believe it would affect if you had it the short version or the longer format as long as the images correspond with the text ie ORIK1.jpg then the text would be ORIK1.txt with prompting with a person as I mention if not in this video, I have an others whatever element of the image you would want to change I would include that in the descriptor. (hair color if you plan on changing the hair color later on, hairstyle etc ) but if it’s unique to that person, i.e. the wrinkles on Gordon Ramsay’s head, or some thing, then I would not have it in the description of the image

It also depends on what you’re trying to achieve…you could just have it as “ORIK PERSON” as you want to have the flexibility to make it into a painting a photo or sketch later on

@@theprotoart Thanks for your reply. Yes I think the "ORIK PERSON" option is closer to what I want to do (I basically just want as much flexibility as possible but with an accurate reproduction of the person, but mostly going for realism/photography... but nothing specifically I want to change like hairstyle). Quick followup, with that in mind, should the txt files have just "ORIK PERSON" or something like "PHOTO OF ORIK PERSON" in them do you think? (In my mind "Photo of" seems sensible since those are what my training images actually will be-just photographs, not paintings or drawings etc-but I don't know enough about how it all works to know if that's necessary or if you would even need to or want to "tell" the model that these are photos of a person vs just "a person").

@@danvorosmarty9854 you should be fine with photo of ORIK person next would be what you’re choosing for the base model. I’ve had some success using epic realism. I usually have a data set of around 40 photos of a person, different clothing, backgrounds, etc..

@@theprotoart Awesome thanks, and good to know about epic realism. When I used Dreambooth previously I used a model called "dreamlike-photoreal" I had found on huggingface which worked ridiculously well (I trained a model for a friend and some of the results would have fooled me had I not made them myself, really just super accurate and realistic). And weirdly, I got the best results using only 5-10 images, and without even very much variation in background, facial expression, clothes etc. When I tried to use more training photos it didn't come out nearly as well. Not sure why exactly that is (maybe had to do with my number of training steps I was using? Or a quirk of that particular base model perhaps?). In any case I'll try your suggestion of epic but also, do you know if it's possible to use an XL base model? (I have a massive queue of like 500 images lined up at the moment that I want to wait for it to finish before I try the training lol).

No not at the moment, hopefully the dev includes that option

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.

No, everything is local. 👍🏾

@@theprotoart Right on. Thanks. I'll give it a try.

Could always put in a request on the discord

No problem

Oh no problem

I appreciate it

Awesome!! I’m hoping there’s a easier way to extract LoRa from trained models created in Bee because I would love to upload those to civitai at some point

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

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

😂

Hey there not sure if you saw that there’s an update 2.5.0 I have it covered on my channel. the dev has been informed about this already, hopefully they update all the links

ruclips.net/video/YeD9fSEKkRs/видео.html

No problem 😉

No problem I’m glad you found it helpful, and greetings ✊🏾 🇲🇽

It really is a powerful little app with a lot of useful features. There is still plenty of room for improvement ✊🏾

You’re welcome. new video coming soon

No idea unfortunately

​@@theprotoart Apparently it has to do with how the individual applications access the audio driver. Oddly, the only VoIP software that works is Zoom. The unfortunate part is Zoom is not one of the applications that is integrated with any routing software, so the mic can ONLY be used by Zoom. Long story short: iPad is great for recording into, but not recommended for Podcasts unless you pay for an application that both records and streams. For iPad, Zoom Pro is one of those but I find the quality to not be ideal. I will be using a laptop in the future to use zoom and record at the same time and will deal with slight fan noise. I will be using the iPad for local recording / auditions. Thank you for the video!

@@KevinSkocypec I can see how that could be very frustrating and a little disheartening because you can get some pretty good quality audio using the Focusrite into this thing

As far as I am aware of You don’t … you can do that using other software like RipX

So to your first question, yes, there is a Hidden folder that is created and there’s instructions on how to access it on the Discord. To your second question I’m not quite sure what you mean by deleting one by one…you can’t delete individual images…you can delete the batch set that is generated. This, however, does not remove the images from your computer just from the app itself the image is generated are also in the hidden folder that is created as mentioned in the previous statement. You can also see the list of images using other apps like “breadboard”, which I’ve covered on here as well.

From the discord “Diffusion Bee creates a hidden directory called .diffusionbee in your home directory. To open this folder, open a Terminal window and type `open ~/.diffusionbee` and press Return. Inside this folder you will find folders for custom models, downloads, and images, along with a .json file with application settings; as of version 2.2.1 there are now multiple JSONs, each of which tracks a different kind of data. For the most part you should leave this folder alone, but if you need to find any of the raw files, you can find them here. New versions of Diffusion Bee should automatically inherit information from this folder.”

@@theprotoart Thanks for the quick response. I've noticed Bee reacts differently (in a good way) to the same models I've always used in ComfyUI and Auto1111. Even when I have style set to none. Just further adds to the mystery of how all of this works.

@@obscuremusictabs5927 yeah I’ve heard that across the board for a a few apps like a1111 on how doesn’t react the same as comfy ui etc. very interesting. All very fun learning it although.

There is already in implementation for drawing tools like AI canvas and in painting which do most of what you’re talking about. Photoshop is an entirely different kind of tool ✊🏾

​@@theprotoart Hmm, they def dont though. Inpainting only has one brush with adjustable size, but cant edit prompt or mask without starting over, or any of the other things I mentioned. And just tried the AI canvas, and is same, but unfortunately even worse, with even less control. Inpaint has the single brush, but here cant adjust size even, and the eraser and undos dont correlate at all to the inpainting brush

@@theprotoart Obviously not complaint to you though! love the app, Its just if these things were possible to add, it would absolutely elevate it dramatically. The rest of the updates havent seemed as much of game changers to me, since like version 1.4, but just inclusion of more models. Photoshop was just an example of an app that includes eraser, undo, and masking tools, wasent saying it should, or could be photoshop though

Yo

How@@theprotoart

@@timelapsebydk how what ?

@@theprotoart how now

@@Lekt0r I don’t quite understand what your question is. Sorry

I appreciate that. I just uploaded it a few minutes ago. ruclips.net/video/ilmsu22745M/видео.html as far as LoRa's in DiffusionBee... you can't use them individually you have to merge them with a base model. at least for now..

It’s crazy how fast the tech keeps evolving 😂😋