Can you actually run a generative AI model on a $5 microcontroller?
In this video, I build, train, optimize, and run a tiny AI model directly on an ESP32-S3 camera device. No GPU. No powerful computer doing the inference in the background. The model runs on the microcontroller itself.
We look at how the model is reduced using downsampling and INT8 quantization, how TinyML works within extremely limited memory, and what happens when we compare 32x32, 64x64, and 128x128 models.
I’ll also show you the browser-based interface I built for training models, testing them on your computer, creating firmware, selecting different visual styles, and uploading everything directly to the ESP32-S3 over USB-C.
This is not Midjourney running on a microcontroller. It is an experiment in how far we can shrink an AI model and still get useful visual processing from extremely inexpensive hardware.
Project files and resources:
https://github.com/GeekatplayStudio/Nano-Ai
Board:
https://amzn.to/4wdYi00
If you build something with this project, modify the code, or create your own tiny AI model, let me know in the comments. And if the project is useful, consider giving the GitHub repository a star.
More tutorials and projects:
https://www.geekatplay.com
00:00 Generative AI on a $5 ESP32-S3
00:40 Downloading and Running the Project
01:37 How This Tiny AI Model Works
02:32 Downsampling and Quantization Explained
04:16 Training Data and INT8 Quantization
05:18 AI Model Size, Speed and Quality
06:23 ESP32-S3 Hardware and Memory Limits
08:26 TinyML, Camera and Firmware
11:06 Building Your Own AI Model
13:37 Training the Model
14:13 Uploading AI to the ESP32-S3
15:02 Testing 32x32, 64x64 and 128x128 Models
16:14 Connecting and Controlling the Device
16:56 Project Files and Source Code
⚠️ AFFILIATE DISCLOSURE: We may earn a commission from purchases made through the links below at no extra cost to you.
My recommendations:
Hitem3d: https://www.hitem3d.ai/?utm_source=aff&utm_medium=cps&utm_campaign=geekatplay
My Patreon webpage - https://www.patreon.com/geekatplay
Tutorials and packs - https://gumroad.com/geekatplay
Tutorials website - https://www.geekatplay.com
Photography - https://www.chopinephotography.com
#GenerativeAI #ESP32 #ESP32S3 #TinyML #EdgeAI #EmbeddedAI #MachineLearning #ArtificialIntelligence #AIProjects #Microcontroller #GeekatPlay
In this video, I build, train, optimize, and run a tiny AI model directly on an ESP32-S3 camera device. No GPU. No powerful computer doing the inference in the background. The model runs on the microcontroller itself.
We look at how the model is reduced using downsampling and INT8 quantization, how TinyML works within extremely limited memory, and what happens when we compare 32x32, 64x64, and 128x128 models.
I’ll also show you the browser-based interface I built for training models, testing them on your computer, creating firmware, selecting different visual styles, and uploading everything directly to the ESP32-S3 over USB-C.
This is not Midjourney running on a microcontroller. It is an experiment in how far we can shrink an AI model and still get useful visual processing from extremely inexpensive hardware.
Project files and resources:
https://github.com/GeekatplayStudio/Nano-Ai
Board:
https://amzn.to/4wdYi00
If you build something with this project, modify the code, or create your own tiny AI model, let me know in the comments. And if the project is useful, consider giving the GitHub repository a star.
More tutorials and projects:
https://www.geekatplay.com
00:00 Generative AI on a $5 ESP32-S3
00:40 Downloading and Running the Project
01:37 How This Tiny AI Model Works
02:32 Downsampling and Quantization Explained
04:16 Training Data and INT8 Quantization
05:18 AI Model Size, Speed and Quality
06:23 ESP32-S3 Hardware and Memory Limits
08:26 TinyML, Camera and Firmware
11:06 Building Your Own AI Model
13:37 Training the Model
14:13 Uploading AI to the ESP32-S3
15:02 Testing 32x32, 64x64 and 128x128 Models
16:14 Connecting and Controlling the Device
16:56 Project Files and Source Code
⚠️ AFFILIATE DISCLOSURE: We may earn a commission from purchases made through the links below at no extra cost to you.
My recommendations:
Hitem3d: https://www.hitem3d.ai/?utm_source=aff&utm_medium=cps&utm_campaign=geekatplay
My Patreon webpage - https://www.patreon.com/geekatplay
Tutorials and packs - https://gumroad.com/geekatplay
Tutorials website - https://www.geekatplay.com
Photography - https://www.chopinephotography.com
#GenerativeAI #ESP32 #ESP32S3 #TinyML #EdgeAI #EmbeddedAI #MachineLearning #ArtificialIntelligence #AIProjects #Microcontroller #GeekatPlay
- Категория
- Партнёрские программы
Комментариев нет.
