How to train your own AI image models without breaking the bank
Vložit
- čas přidán 25. 07. 2024
- Learn from Principal Systems Designer at GitHub, Maxime De Greve, about the techniques and easy-to-use tools that I've used to train my own avatar AI model without requiring expensive hardware. Through an easy-to-follow walkthrough, the talk aims to empower attendees to leverage AI and train models with their own objects, faces, or styles.
We are never going to replace the work of our illustrators or augment it away with AI. There really is no substitute for the creative problem solving and human touch that goes into a bespoke illustration. However, we are interested in how AI might quickly test and validate ideas as a brainstorming partner and how it could act as another tool to help expand the talents of our human artists.
0:00 - Intro
0:15 - About me
1:03 - Introduction into AI
2:53 - Examples of generative images
3:55 - Generative models
5:21 - Cost of training generative models from scratch
6:46 - Fine-tuning models
8:05 - Should I fine-tune a model?
11:25 - Fine-tune your own model (Tutorial)
20:16 - Touching up your images using AI
27:10 - Outro
#GitHub #AI #Innovation
Stay up-to-date on all things GitHub by subscribing and following us at:
CZcams: bit.ly/subgithub
Blog: github.blog
Twitter: / github
LinkedIn: / github
Instagram: / github
TikTok: / github
Facebook:
About GitHub
It’s where over 100 million developers create, share, and ship the best code possible. It’s a place for anyone, from anywhere, to build anything-it’s where the world builds software. github.com - Věda a technologie
You didn't mention how you uploaded the images for training the model, lets say I want to train the model on my pet, I thought that was the whole point of this video.
right? wtf
i can’t believe this clickbait useless content is on github official channel
I actually thought that I had missed this part so I watched the video a second time. I didn't see/hear it mentioned.
Idk how this video is recommended at all, it should have been 20 seconds long
Brillaint video... really clear and clarifies a lot of information
Thanks. Can’t wait to try some of these techniques.
definitely coming back to this
Bro, this helps me sooo much!!❤ Thank you so much ❗
this was an awesome demo! i am definitely doing this, thank you!
Thnx i was looking for a way to train ai with my own cartoon characters and now i can work so much faster❤
Thanks a lot, works really well but google collab keeps disconnecting after 1 or 2 generation of image telling that you can't use it for this. Can you run the code on your local computer ? If yes, How ? Thanks
I have this error when trying to generate images in Stable Diffusion locally:
Error: Could not load the stable-diffusion model! Reason: Unknown model! No config file path specified in context.model_configs for the "stable-diffusion" model!
I read somewhere that yaml file is missing. Where can I find it?
thank you and helped alot ! RESPECT
I have a question for those who where able to do it, is there a way to have the same gradio interface in hugging face with the same amount of options and sliders etc., when I uploaded my model it is just a simple text input with no sliders or any way to change the batch size, width,etc.
Great talk, Maxime!
1:34 1:36
this is awesome
Anyone have an idea why mine is running very slow or stops at a certain point while running the cell after cell?
I think the last part "testing the model" dont work anymore...
Legend!
Awesome video! one thing I haven't figured out: at what point and how do you feed you prepared images to the training / fine-tuning session?
Any videos on how to train text to video model ? Any help is appreciated. Thank you !
Thanks!
how can I use this code locally? To connect to my local paths with files, not google drive? Experiencing disconnections that don't let me finish execution of code blocks
Where do you put your images ? I didn’t see in the Colab notebook.
you run the cell "Instance Images" by clicking on the play button then there is a button to select your images on your computer
@@TechNbags i get an error about wget:
ModuleNotFoundError Traceback (most recent call last)
in ()
6 import ipywidgets as widgets
7 from io import BytesIO
----> 8 import wget
9
10 with capture.capture_output() as cap:
ModuleNotFoundError: No module named 'wget'
I want to create a model for text to image for fighting poses.
Is there a way to rent space in the cloud to train a model?
hi I am facing this issue: from pyngrok import ngrok, conf, kindly let me know how to solve it
Thanks
i didnt got it how to use this google colab thing, in end saud "something went wrong" and i have no idea why...i tried all you showed and said in the video, but for someone who has no idea bout all this, its not detailed enought :(
After a few minutes of playing with my models I always get a "Connection errored out." message in the test web page. Is this because I am not paying for google colab or some other thing I am missing ?
I have the same and I'm wondering if it's the same, did you find a solution ?
00:04 Learn to train AI image models on a budget
02:19 Stable diffusion is an open source model for generating diverse and high quality samples.
04:56 AI image models are expensive to train due to infrastructure costs.
07:23 Training AI models for specific styles or subjects
10:12 Training your own AI image models is necessary for specific objects or mascots.
13:14 Using Google collab for AI training with strong GPU access.
15:53 Training AI image models with a small dataset
18:05 Training your own AI image models improves results
20:22 Using in painting to improve generated images
22:30 AI image models need manual touch-ups for best results
24:43 Training AI image models for diverse collections of people and avatars
links?
Missing a lot of steppes, especially how to do basic operations with images. I've encountered at least 3 issues on that, which seems not letting me to go further & seems b'coz of that I don't have things like shown in step 10 & step 11. Also every time I'm clicking "play" on any cell it prompting google login window again & again even if I logged in & allowed/permitted everything through it. I would rly like to have detailed video about step by step setup with deep explanation on every step instead of introduction of what popular AI platforms are. I would like to get video about how to set all this stuff on own PC & use own hardware/GPU(S).
Wanted to see you make ground truths for your images.
cannot run "TEST THE MODEL"
Model Download fails with a ModuleNotFoundError
😎💙🙏🏽
Thank you so much! this is what i pay my internet bills for
Its ChatGPT mate. Not GTP.
i also notice that
I had one job: reading it off the slide.
many humans do this ..mate.. it's called dyslexia, many suffer mild forms. even in hearing.. imagine if you hear rhythm reversed sporadically.. id' show up as spastic dancing.. thus the term "white ppl can't dance".. it afflicts that demographic, just a tiny example.
Did you manage to do it ?
I give you 10 seconds to answer.
Thank you so much