This prompt manipulation looks way better than the previous img2img that exists. But i know that if i spend time installing manually, then it will be added to 1111 within a few days.
Hey! Thanks for the great breakdown :) I have a quick question: Is it possible to run prompt-to-prompt on images that weren't originally SD generated via prompt? (i.e instead of generating an original image of potates, just start from a real picture of potatoes and add the prompt to it).
Based on my prediction, stabilization of the MetaVerse will definitely depend on our prompt to prompt interactions. It’s becoming clear that each computer connection will have its own visual interface, but the rules must remain constant. Nice video
Thanks a great deal for your videos! I saw the other one about bringing the AI generated characters to video! When you bring your different avatar images to life on every video, do you use the sane technique or able to use another solution to record your videos using different avatar images in real time ?
Your best bet is running cmdr2's Stable Diffusion UI. I run it on GTX 1050 with 3 GB VRam without a hitch. In fact, I am watching this video while Stable Diffusion is rendering a batch of images in the background.
I was just trying to install it on a cloud gpu provider using a jupyter. When I reached step "4.) conda activate diffusers2" I got the following error: CommandNotFoundError: Your shell has not been properly configured to use 'conda activate'. To initialize your shell, run $ conda init Currently supported shells are: - bash - fish - tcsh - xonsh - zsh - powershell See 'conda init --help' for more information and options. IMPORTANT: You may need to close and restart your shell after running 'conda init'. Any help would be much appreciated!
Question for more advanced users. What if I want to train SD with multiple different images of my own. For example: suppose I want to train SD on blue eyes and on green eyes. Currently it's easy enough to train on either and get a .ckpt, but if I want to include both trainings in my model, what then? 1. Using checkpoint merger - gives bad results, destroys oldest training the most. 2. Train on blue eyes, use output .ckpt as input .ckpt for training on green eyes - seems to work but older trainings do get lost over time. 2b. Same method as above but use different class images for each training. Even though both trainings are eyes (and therefore all the class images are images of eyes). The idea being that loss is reduced as new trainings are added, I am just not sure what the best way is to expand my model with multiple new trainings. Hope someone here has any pointers.
I'm new here, but let me ask this question but how to make you face like this wend you toking I know the is face to face but I need to know who you did which a webcam or phone
I haven't bothered with Automatic's web ui yet. But i'll get into it when this is added. I don't have the time to bother with troubleshooting installation with this method cause I always run into an issue.
@@NerdyRodent Been trying all morning. Managed to convert the model to diffusors, but now getting all types of errors trying to invoke it using. device = torch.device("cuda" if torch.cuda.is_available() else "mps" if torch.backends.mps.is_available() else "cpu") pipe = StableDiffusionPipeline.from_pretrained("C:/Users/chlyw/Desktop/Automatic1111/prompt-to-prompt-main/diffusors/model_index.json") pipe = pipe.to(device) ** Repo id must be in the form 'repo_name' or 'namespace/repo_name':
He's using the Thin-Plate Spline Motion Model for Image Animation. I sent a link in the first comment, but I think TH-cam got rid of it in the spam filter. You can type in the name of the model with "hugging face" after it to see a hosted version.
Second time your channel has taught me something amazing.
Awesome, and many thanks! 😀
Im on my 6th time.
This prompt manipulation looks way better than the previous img2img that exists. But i know that if i spend time installing manually, then it will be added to 1111 within a few days.
1111 is wild
1111 is giving us Christmas twice a week
You sir are a national knowledge sharing treasure.
Hey! Thanks for the great breakdown :) I have a quick question: Is it possible to run prompt-to-prompt on images that weren't originally SD generated via prompt? (i.e instead of generating an original image of potates, just start from a real picture of potatoes and add the prompt to it).
nice to finally see your actual face. thanks for the tutorial!
Making it all look just too easy, as usual. Thanks.
Great video, thank you :) Is it possible to run this on Collab, please?
Thank you, thanks Google as well!
At this rate, we're gonna have holodeck-style natural language image editing before the end of the year...
Just downloaded it on my Android phone and it's amazing!!!🖖😁
@Mr Mavvy malware
Maybe a reference to the dude that created an app to run Stable Diffusion on mobile. Called "Maple Diffusion" or something like that.
@Mr Mavvy a demo was shared on Twitter, though I wouldn't recommend downloading it unless you're okay with waiting 10 minutes per a single image 😂
Based on my prediction, stabilization of the MetaVerse will definitely depend on our prompt to prompt interactions. It’s becoming clear that each computer connection will have its own visual interface, but the rules must remain constant. Nice video
Thank you very much for this knowledge, I will wait for the Colab link please.
+1
Thanks a great deal for your videos! I saw the other one about bringing the AI generated characters to video! When you bring your different avatar images to life on every video, do you use the sane technique or able to use another solution to record your videos using different avatar images in real time ?
Yup, same ;)
Fantastic, thank you so much
Thank you so much for this one!! 🤩
Outstanding channel!
Can you suggest me a best method to run stable diffusion local in 4GB NVIDIA card
That's a good question! Whenever I have free time, I'll have to figure it out too!
Automatic1111 runs it reasonably well on a 2070 with 4GB.
Your best bet is running cmdr2's Stable Diffusion UI. I run it on GTX 1050 with 3 GB VRam without a hitch. In fact, I am watching this video while Stable Diffusion is rendering a batch of images in the background.
@@mrdragonrider thanks man
@@mrdragonrider what resolution can you go upto?
I was just trying to install it on a cloud gpu provider using a jupyter. When I reached step "4.) conda activate diffusers2" I got the following error:
CommandNotFoundError: Your shell has not been properly configured to use 'conda activate'.
To initialize your shell, run
$ conda init
Currently supported shells are:
- bash
- fish
- tcsh
- xonsh
- zsh
- powershell
See 'conda init --help' for more information and options.
IMPORTANT: You may need to close and restart your shell after running 'conda init'.
Any help would be much appreciated!
I wouldn’t use conda in colab personally as it’s already a separate environment
I want this!
Is fun 😀
Please give us the jupyter notebook to work with
Link is in the description ;)
How are you doing the talking head? Is it DFL? Please, I need to know.
I’m using the thin plate spline motion model for image animation. Link is in the video description 😉
@@NerdyRodent Thank you!
Great video, do u know if this is already implemented in automatic1111 webui
It’s been 1 day so I’d be amazed if it hasn’t 😉
how do you make your face cam... what program do you use to remap your video onto that anime looking character... stable diffusion? what else?
Link in the description 😉
@@NerdyRodent that’s very awesome… thank you for all you do 😎💪
Question for more advanced users. What if I want to train SD with multiple different images of my own. For example: suppose I want to train SD on blue eyes and on green eyes. Currently it's easy enough to train on either and get a .ckpt, but if I want to include both trainings in my model, what then?
1. Using checkpoint merger - gives bad results, destroys oldest training the most.
2. Train on blue eyes, use output .ckpt as input .ckpt for training on green eyes - seems to work but older trainings do get lost over time.
2b. Same method as above but use different class images for each training. Even though both trainings are eyes (and therefore all the class images are images of eyes). The idea being that loss is reduced as new trainings are added,
I am just not sure what the best way is to expand my model with multiple new trainings. Hope someone here has any pointers.
I just used different tokens and daisy chained, but there is a repo that uses image filenames to allow multiple things at once :)
I hate to ask a question off topic but what program are you using for your avatar? And not obs.. I mean just the mocap Avatar
Link in the description 😉
Is this already implemented in AUTOMATIC1111 web-ui? or weighted prompts are different than this?
The code has already been out for a day so it should be implemented already!
@@NerdyRodent hmmm so how do i install itt for local windows?
@@fabgeb667 Run the 7 commands in the description 😉
When can we expect Stable Diffusion for Radeon?
Already works on AMD cards 😄 Easiest way is to use the rocM pytorch.
@@NerdyRodent thanks
Is strange that I have a problem running this on colab?
Should run ok on a T4, but if you get OOM try the low vram setting
@@NerdyRodent no my problem is more about a ptp_utils module not found, even if I clone the repo
can it img2img?
now there only needs to be a gui for this lmao
Other than the jupyter web ui? 😆
@@NerdyRodent My smooth brain self is barely able to use automatic's web GUI so ... probably yeah
@@RHYTE Lols XD
How did you make a V-Avatar like that from the AI Image?
Link is in the description 😉
One of these days, ill fifure out how to do a face swap and voice change
I’d check out my videos for voice cloning and avatar animations 😉
GUI for this???
Probably Automatic's webui will have it.
rodent with 3 legs and lion with 12 fingers.
Best type!
Just waiting for it to be optimized to work with 8GB of VRAM and added to Automatic1111's repo... I give it a week 😁
I love that pytourch works on every platform and os. And u don't have to port the application.
Why does the visualization split "rodent" into "ro" and "dent"? These two syllables make no sense by themselves.
I'm new here, but let me ask this question but how to make you face like this wend you toking I know the is face to face but I need to know who you did which a webcam or phone
Link is in the description 😉
@@NerdyRodent thx i will look up later
this next gen coding when there's the parameter "croutons"
Thanks Nerdy ♥
I haven't bothered with Automatic's web ui yet. But i'll get into it when this is added. I don't have the time to bother with troubleshooting installation with this method cause I always run into an issue.
Yeah, I prefer just using the normal conda method for A1111 as well!
Any chance you could put this into a colab?? I'd be super thankful
Comes as a notebook :)
@@NerdyRodent seems to run into some problems, at least on colab
Ohh, can't wait till I press git pull on Automatic's webui and it magically just runs like a charm.
We'll probably have to wait at least 10 minutes for that though! ;)
is it possible to just use a local CKPT instead of diffusors token?
Yup, you can convert the ckpt to diffusers format then use that directory instead!
@@NerdyRodent Been trying all morning. Managed to convert the model to diffusors, but now getting all types of errors trying to invoke it using.
device = torch.device("cuda" if torch.cuda.is_available() else "mps" if torch.backends.mps.is_available() else "cpu")
pipe = StableDiffusionPipeline.from_pretrained("C:/Users/chlyw/Desktop/Automatic1111/prompt-to-prompt-main/diffusors/model_index.json")
pipe = pipe.to(device)
**
Repo id must be in the form 'repo_name' or 'namespace/repo_name':
"hold on to your papers" 🤣
I love Two Minute Papers as well - awesome channel!
how do you make your avatar?
Link is in the description 😉
Can't wait for this to be in auto's gui
Dude, your videos are the best, thanks for teaching us all this cool stuff.
Glad you’re having fun! 😀
Can you make a tutorial for streaming like you . What Programm, how to choose what you Look like etc.
He's using the Thin-Plate Spline Motion Model for Image Animation. I sent a link in the first comment, but I think TH-cam got rid of it in the spam filter. You can type in the name of the model with "hugging face" after it to see a hosted version.
He has done a tutorial already. No streaming though, it's done after recording; not fast enough for real-time just yet.
this is really amzing, may i ask how to make the talking avatar realtime at the right side corner? thanks a lot
Link is in the description 😉
@@NerdyRodent ok thanks a lot : )