Training A Diffusion Model - Stable Diffusion Masterclass

This is a preview lesson from the deeplizard Stable Diffusion Masterclass!

Welcome to this deeplizard course, Stable Diffusion Masterclass - Thoery, Code, & Application! In this course, we learn the theory behind Stable Diffusion and get hands on experience with code and applications.

🔥 EARLY ACCESS SALE 🔥
Be the first to enroll in the course, and get the limited-time early access sale price!

✨ See the course page to get an overview of the course and enroll:

✨ Check out more course details and other resources in the course intro:

💥🦎 DEEPLIZARD COMMUNITY RESOURCES 🦎💥

👋 Hey, we're Chris and Mandy, the creators of deeplizard!
👀 CHECK OUT OUR VLOG:

💪 CHECK OUT OUR FITNESS CHANNEL:

🧠 Use code DEEPLIZARD at checkout to receive 15% off your first Neurohacker order:

❤️🦎 Special thanks to the following polymaths of the deeplizard hivemind:
Mano Prime

👀 Follow deeplizard:

🎓 Deep Learning with deeplizard:

🎓 Other Courses:

🛒 Check out products deeplizard recommends on Amazon:

📕 Get a FREE 30-day Audible trial and 2 FREE audio books using deeplizard's link:

🎵 deeplizard uses music by Kevin MacLeod

❤️ Please use the knowledge gained from deeplizard content for good, not evil.

[marufahmed]

Really good explanation, please continue to produce these videos.

[deeplizard]

✨ See the course page to get an overview of the course and enroll:

✨ Check out more course details and other resources in the course intro:

[stefanobraghettocatoni]

I have a question. Is fair to use the same noise scheduler for inference than for training? I am asking because when the model is predicting noise during training it has just the correspond noise step acording to that timestep. But when inference, the model has 2 noises: The timestep noise (same as training) + noise that wasn't able to delete since the model isn't perfect and always will make tiny mistakes. For diffusion process with a lot of steps I fell is unfair to treat both as the same.

Thank you for your clarification!

[alikadhm]

we need more discount for this course, it’s expensive compared with udemy

[shanhuang]

This is really explained poorly in that

1. It assigns unnecessary complexity: rather than say the model subtracts the predicted noise and adds it back, you could have just said it subtracts 10% of the predicted noise.

2. The explanation goes on and on with the computation steps without explaining the intuition.

3. It is simply incomplete. You did not explain what are we doing with noise scheduler with multiple “t”s, if during training, the model is already learning denotes in incremental steps