Loading video...

Video Failed to Load

There was a problem loading this video. This could be due to a temporary network issue or the video might be unavailable.

another example: a developer used the exact same $8 ESP32-S3 chip and local AI setup to build a real-time object recognition camera the fact that this is running entirely offline, on an eight-dollar microcontroller, without any cloud API calls, is wild... it captures the video feed, processes it locally... show more

ard

5,061 subscribers

230,333 views • 3 months ago •via X (Twitter)

Science & Technology

Anya Rossi• Live Now

Private livecam show

46 Comments

Marc The Shark3 months ago

I’m doing something similar but just running computer vision at the edge and then gets laid back to my agents as needed. I like what you were doing.

ard3 months ago

your implementation sounds cool!

ɢᴀᴅɢᴇᴛɢᴀᴢ ⌁⌁3 months ago

I used one to build this.

ard3 months ago

this is really cool, have you thought about working on this as a commercial product?

ɢᴀᴅɢᴇᴛɢᴀᴢ ⌁⌁3 months ago

Thanks, i have actually there is definitely a gap in the market for smart cat toys. The ones on the market all do the same thing just a laser or a bit that spins, cats are incredibly intelligent and soon predict the patterns and get bored. Thats what made me build it, this is just version 1 next I’m gonna add a camera so it knows when the cat is present and its location with the computer vision ai model. Then use that information to move the feather accordingly and automatically evading the cat like a real prey animal would.

Jurly3 months ago

that's a powerful demo of edge AI — the real test is how well it handles occlusion and lighting changes at that price po

ard3 months ago

i think at this price it doesn't matter at all

MAX3 months ago

Cheap hardware plus local AI is opening up a lot of possibilities.

ard3 months ago

fact

Misato3 months ago

the $8 chip doing this is still insane🔥

ard3 months ago

hell yeah

Tony Scott 🧄(🦆🐓🐵🧪🧬🪪)❌=↑🧄🧄🧄🥩🥚🧀↓👽👾🤖3 months ago

Object detection means autonomous targeting capability.

Kaworu3 months ago

we are giving dumb objects a brain now🙃

ard3 months ago

i'm waiting for this device to run Doom using a local AI

david3 months ago

Excellent. I have the same ESP32, and let me replicate.

ard3 months ago

you can try this

M1Lo3 months ago

How mucho local memory?

ard3 months ago

512 KB of fast SRAM and 16 MB of flash

Richard Rice3 months ago

Woot woot. Pieces 😉

christian coler3 months ago

Idiots who have a child's understanding of LLMs and zero understanding of microcontrollers talking about a "hardware barrier" being shattered by this or a tiny LLM running on an ESP32, as if the former is revolutionary or later is beneficial to an average person's usecase

Ahnaf Shahriar3 months ago

Yolo models are goated in local image recognitions 🔥

ɢᴀᴅɢᴇᴛɢᴀᴢ ⌁⌁3 months ago

This chip will get you running but the camera will have its limitations. Such as resolution and sensors this will give you at best VGA quality.

Aenish Shrestha3 months ago

Is this an open source project ?

🎶3 months ago

Important step towards singularity immersion

Nicolas S. 🛰️3 months ago

Not sure but you said “$8 chip *and local AI setup*”. So my understanding is that the ESP32-S2 only captures, streams to video feed to a nearby, potentially beefy, laptop and *that’s* what running the AI and doing the recognition, correct? Or are you saying the S2 is autonomous?

Osaker3 months ago

老实说，对比一个中国量产带PTZ的 CCTV，这个价格对这程度的功能，CP值并不算高…

CB3 months ago

Bit of an overstatement

EVIL:/GriN3 months ago

Ah the last time I tried esp32 cam it got turned to crisp 😭

Nico3 months ago

Well done to the dev! Inference rate is too slow for anything safety related but slap that onto your home security and for $10 bucks you’ve got a killer setup.

AI Mastery Guide3 months ago

Real time object recognition for under $10, that barrier is really gone now.

SushiDude3 months ago

my 10yo son made this a year ago for birds outside with Claude Cli

ard3 months ago

this is so cute...

Sebastian Buzdugan3 months ago

the hard part is not esp32-s3 inference, it is stable camera input in bad light

🏴‍☠️BOSSDOG🏴‍☠️3 months ago

lol Blue Iris has been doing this for 10 or more years with yolo.

𝛼 ✨3 months ago

how the hell is this possible

Proyecto5T ➡️ 2024-20303 months ago

@BrianRoemmele

Zak Allal3 months ago

fact-check this tweet by @ardchain. The $8 ESP32 isn’t running YOLO11, it’s a webcam feeding the PC that is running YOLO11. The LLM was trained only on TinyStories &its author says it “will not answer questions, nor follow instructions, nor write code, nor know facts.” The 512KB claim omits the 8MB PSRAM. No barrier was shattered.

Anthum AI3 months ago

$8 for real-time local object recognition is genuinely nuts. No cloud, no API cost, just glue it onto whatever you want.

安叫兽|Bird🕊️ 🔶 BNB3 months ago

八美元小板子跑成这样，离线这点太狠了

Modernus Cicero3 months ago

The S3 is a nice little package.

Vova3 months ago

I’ve built that using ESP32

ard3 months ago

looks stylish

Terri St. Quill3 months ago

I have GOT to know more about this!!

unchosen.eth3 months ago

edge ai keeps becoming more practical

MAGIC KOM chanell2 months ago

Cool stuf

Chuck Petras3 months ago

@BrianRoemmele

Related Videos

btw this robot runs on a fully open source AI framework called Xiaozhi it turns a $8 ESP32 chip into a fully autonomous voice assistant NO expensive cloud servers or high latency wake-word detection and audio streaming happen locally on the chip: > requires only an ESP32-S3, a microphone, and a basic speaker > supports real-time conversational voice interaction > completely open-source and customizable for any DIY robotics project anyone can now build a personal AI companion for the price of a cup of coffee. (github link in the replies)

btw this robot runs on a fully open source AI framework called Xiaozhi it turns a $8 ESP32 chip into a fully autonomous voice assistant NO expensive cloud servers or high latency wake-word detection and audio streaming happen locally on the chip: > requires only an ESP32-S3, a microphone, and a basic speaker > supports real-time conversational voice interaction > completely open-source and customizable for any DIY robotics project anyone can now build a personal AI companion for the price of a cup of coffee. (github link in the replies)

ard

42,210 views • 2 months ago

Every time I think the ESP32-S3 has hit its limit, someone finds another gear. This open-source Jet renderer is now pushing the S3 even harder. Real-time screen-space reflections, water ripple distortion, lens flare and 8 ships on screen, all running on a tiny MCU. Performance dropped from around 60 FPS to 45–50 FPS, but that's still wild for an ESP32-S3. And the developer thinks there is more performance left to squeeze out. No GPU. No desktop CPU. Just an ESP32-S3 doing real-time 3D graphics.

Every time I think the ESP32-S3 has hit its limit, someone finds another gear. This open-source Jet renderer is now pushing the S3 even harder. Real-time screen-space reflections, water ripple distortion, lens flare and 8 ships on screen, all running on a tiny MCU. Performance dropped from around 60 FPS to 45–50 FPS, but that's still wild for an ESP32-S3. And the developer thinks there is more performance left to squeeze out. No GPU. No desktop CPU. Just an ESP32-S3 doing real-time 3D graphics.

Praveen Kumar Verma

15,924 views • 2 months ago

ZERO-LATENCY AIMBOT? WORSE. A ZERO-LATENCY PHYSICAL TURRET someone built a rig in their garage that shoots exactly where you click on the live video feed no cloud APIs, no server lag. just raw local execution. a simple python tkinter GUI captures mouse coordinates and streams them over websockets. - the local $8 ESP32 chip processes the spatial data - the robotic arm pivots instantly to match the digital vector - the green laser locks onto the exact physical coordinate with zero delay you know where this is going? if the guy who built the head-tracking software ever meets the guy who built this physical turret... WE ARE SO COOKED.

ZERO-LATENCY AIMBOT? WORSE. A ZERO-LATENCY PHYSICAL TURRET someone built a rig in their garage that shoots exactly where you click on the live video feed no cloud APIs, no server lag. just raw local execution. a simple python tkinter GUI captures mouse coordinates and streams them over websockets. - the local $8 ESP32 chip processes the spatial data - the robotic arm pivots instantly to match the digital vector - the green laser locks onto the exact physical coordinate with zero delay you know where this is going? if the guy who built the head-tracking software ever meets the guy who built this physical turret... WE ARE SO COOKED.

ard

49,616 views • 3 months ago

I made a Draconia Malfoid Visual Novel. I used Vivix's new W1 model, which is unique for it's real-time interaction. No more static cut-outs of characters and still backgrounds. Vivix's W1 model costs $0.003 per second and is fully interactive, shattering the cost barrier while being high quality and real-time. Vivix W1 is also different. It's not a simple clip model, its beta API offers something unique - it lets you modify the video on the fly, a real-time interactive video model. Most AI video is prompt in, clip out, then repeat. Every prompt a break or cut in the video. Vivix W1 is continuous, creating a seamless video experience. That means you can have an LLM handle the conversation, and stream the new chat and emotions and scenes as signals into the API to direct the character's video in real-time, without pauses. You can take characters anywhere in the world, any setting. That's kinda nuts. Vivix reached out to see if I was interested, so I put together a quick AI Visual Novel system with their API. The prototype video is impressive at this speed. The W1 model is directable in real-time. The responses, outfit changes and the dances are all live. The W1 seamless responsiveness is something no clip generator offers. At just about 10 bucks/hour of video, gaming applications are within reach. I do wish the audio was better, it tends to have drop-outs, and it currently doesn't do seamless scene transitions. But the model is in full scale training now, and I hope to have characters moving from location to location with the player. The site is clip-only for now, real-time is the beta API, browser interaction is later. This is the part I am most interested in. You can check it out at: *Sponsored by Vivix.AI

I made a Draconia Malfoid Visual Novel. I used Vivix's new W1 model, which is unique for it's real-time interaction. No more static cut-outs of characters and still backgrounds. Vivix's W1 model costs $0.003 per second and is fully interactive, shattering the cost barrier while being high quality and real-time. Vivix W1 is also different. It's not a simple clip model, its beta API offers something unique - it lets you modify the video on the fly, a real-time interactive video model. Most AI video is prompt in, clip out, then repeat. Every prompt a break or cut in the video. Vivix W1 is continuous, creating a seamless video experience. That means you can have an LLM handle the conversation, and stream the new chat and emotions and scenes as signals into the API to direct the character's video in real-time, without pauses. You can take characters anywhere in the world, any setting. That's kinda nuts. Vivix reached out to see if I was interested, so I put together a quick AI Visual Novel system with their API. The prototype video is impressive at this speed. The W1 model is directable in real-time. The responses, outfit changes and the dances are all live. The W1 seamless responsiveness is something no clip generator offers. At just about 10 bucks/hour of video, gaming applications are within reach. I do wish the audio was better, it tends to have drop-outs, and it currently doesn't do seamless scene transitions. But the model is in full scale training now, and I hope to have characters moving from location to location with the player. The site is clip-only for now, real-time is the beta API, browser interaction is later. This is the part I am most interested in. You can check it out at: *Sponsored by Vivix.AI

Grummz

76,723 views • 16 days ago

we are introducing a state-of-the-art real-time object detection model, RF-DETR RF-DETR outperforms all existing object detection models on real world datasets and is the first real-time model to achieve 60+ Average Precision on COCO talked w/ NVIDIA about it at GTC:

we are introducing a state-of-the-art real-time object detection model, RF-DETR RF-DETR outperforms all existing object detection models on real world datasets and is the first real-time model to achieve 60+ Average Precision on COCO talked w/ NVIDIA about it at GTC:

Joseph Nelson

18,231 views • 1 year ago

Introducing DeepSeek R1:1.5b Running Locally in Cursor! In less than 4 minutes, I set up the DeepSeek R1:1.5b model, download it, and run it locally to seamlessly work with code in Cursor The video is shown in real-time, with the model impressively running on an Intel Core i5 😱 using Ollama and the CodeGPT extension 👇

Introducing DeepSeek R1:1.5b Running Locally in Cursor! In less than 4 minutes, I set up the DeepSeek R1:1.5b model, download it, and run it locally to seamlessly work with code in Cursor The video is shown in real-time, with the model impressively running on an Intel Core i5 😱 using Ollama and the CodeGPT extension 👇

Daniel San

584,392 views • 1 year ago

The QVAC SDK puts the "brain" directly into your pocket. From real-time on-device translation to multimodal understanding, build apps that work everywhere, even 30,000 feet in the air. Local AI is here: 💡Offline-First: No cloud, no latency, no "Department of Truth". 💻 Universal API: One codebase for iOS, Android, macOS, and Linux. 🔍 Multimodal: Understanding text, audio, and images without a server. If you can dream it, you can build it. The era of Stable Intelligence has begun. Start building:

The QVAC SDK puts the "brain" directly into your pocket. From real-time on-device translation to multimodal understanding, build apps that work everywhere, even 30,000 feet in the air. Local AI is here: 💡Offline-First: No cloud, no latency, no "Department of Truth". 💻 Universal API: One codebase for iOS, Android, macOS, and Linux. 🔍 Multimodal: Understanding text, audio, and images without a server. If you can dream it, you can build it. The era of Stable Intelligence has begun. Start building:

QVAC

36,652 views • 6 months ago

is your player object the same as your character object? learn why you should be separating them, and how to implement it in your #gamemaker projects. in fact, this technique can be used for any #gamedev project

is your player object the same as your character object? learn why you should be separating them, and how to implement it in your #gamemaker projects. in fact, this technique can be used for any #gamedev project

~gentoo~

24,152 views • 2 years ago

$72/day for a voice that never sleeps or build a local AI device around an $8 class ESP32-S3 honestly, i prefer the second option you can teach the robot to fit your habits and needs, and that makes it feel more alive – even though building one is far from simple this little robot already shows the appeal: it catches a capacitor polarity mistake, warns about reverse polarity, and reacts with a face i have an article in my profile about running local AI on this kind of chip 😉

ard

25,476 views • 1 month ago

for everyone wondering how zero-delay tracking is even possible on an $8 chip: it is all thanks to the Kalman filter. this 1-minute clip from Steve Brunton perfectly explains why it works (full lecture in reply) instead of just observing a target, it mathematically predicts where it will be in the next millisecond. here is what happens under the hood: - balances physics of motion against noisy camera data - corrects the trajectory instantly - moves the robotic arm before the next frame processes this is the exact same math used in rockets and autonomous driving, stripped down for a cheap microcontroller.

for everyone wondering how zero-delay tracking is even possible on an $8 chip: it is all thanks to the Kalman filter. this 1-minute clip from Steve Brunton perfectly explains why it works (full lecture in reply) instead of just observing a target, it mathematically predicts where it will be in the next millisecond. here is what happens under the hood: - balances physics of motion against noisy camera data - corrects the trajectory instantly - moves the robotic arm before the next frame processes this is the exact same math used in rockets and autonomous driving, stripped down for a cheap microcontroller.

ard

53,268 views • 3 months ago

Qwen3.5-35B-A3B running locally on an M4 chip at 49.5 tokens per second. A 35B model. On a laptop. In real time. LOCAL AI IS GETTING SCARY FAST.

Qwen3.5-35B-A3B running locally on an M4 chip at 49.5 tokens per second. A 35B model. On a laptop. In real time. LOCAL AI IS GETTING SCARY FAST.

0xMarioNawfal

478,644 views • 8 months ago

What if you could pick up an ESP32-S3 and play with it without touching the hardware? I made an interactive 3D ESP32-S3 simulator that runs directly in the browser. You can click GPIO pins to toggle them, lift the shield can, watch the board boot, switch camera angles and even export screenshots. Try it.

What if you could pick up an ESP32-S3 and play with it without touching the hardware? I made an interactive 3D ESP32-S3 simulator that runs directly in the browser. You can click GPIO pins to toggle them, lift the shield can, watch the board boot, switch camera angles and even export screenshots. Try it.

Praveen Kumar Verma

19,213 views • 1 month ago

People Occlusion arrived as a beta feature in Varjo Base 4.16, using real-time AI object recognition to detect people in the passthrough feed and composite them into the simulation. For crew training, instructor-led sessions and any scenario where a trainee needs to see a real teammate, that's one less thing standing between the simulation and the task. Read more in our blog:

People Occlusion arrived as a beta feature in Varjo Base 4.16, using real-time AI object recognition to detect people in the passthrough feed and composite them into the simulation. For crew training, instructor-led sessions and any scenario where a trainee needs to see a real teammate, that's one less thing standing between the simulation and the task. Read more in our blog:

Varjo

20,390 views • 2 months ago

Want to add offline voice AI to your ESP32? Start with ESP-Skainet. It is an open-source speech AI framework for building always-on voice assistants that run entirely on ESP32 devices. It combines WakeNet for wake-word detection, MultiNet for offline speech commands, and an Audio Front End with echo cancellation, noise suppression, and voice activity detection. ESP32-S3 is the recommended platform, with the latest models also targeting ESP32-P4.

Want to add offline voice AI to your ESP32? Start with ESP-Skainet. It is an open-source speech AI framework for building always-on voice assistants that run entirely on ESP32 devices. It combines WakeNet for wake-word detection, MultiNet for offline speech commands, and an Audio Front End with echo cancellation, noise suppression, and voice activity detection. ESP32-S3 is the recommended platform, with the latest models also targeting ESP32-P4.

Praveen Kumar Verma

15,209 views • 3 months ago

AI BASKETBALL ANALYSIS. A FULL COMPUTER VISION SYSTEM. BUILT ON YOLO, OPENCV, AND PYTHON. Take any NBA broadcast. Any camera angle. Any resolution. Feed it into the system. YOLO finds every player and the ball. Frame by frame. No manual annotation. No pre-labeled data.The model just sees the court and understands it. Zero-shot classification looks at jersey pixels and decides who plays for whom. Two teams separated in milliseconds. Without ever being told the team names. But here's where it gets wild. Court keypoint detection identifies the geometry of the playing surface. Homography transforms the broadcast camera into a top-down tactical map. Real meters. Real coordinates. From a flat 2D video. Now every player has a position on a real court. Every movement becomes a data point. Every pass becomes a vector. Speed. Distance. Possession time. Pass networks. All extracted from nothing but pixels. No GPS. No chips in the ball. No million-dollar Second Spectrum setup. A Python script and a GPU. The full pipeline is open source. The tutorial walks through every line of code. From detection to transformation to analytics output. Teams used to pay six figures for this data. Now you build it in an afternoon. Computer vision didn't just enter basketball. It democratized it.

AI BASKETBALL ANALYSIS. A FULL COMPUTER VISION SYSTEM. BUILT ON YOLO, OPENCV, AND PYTHON. Take any NBA broadcast. Any camera angle. Any resolution. Feed it into the system. YOLO finds every player and the ball. Frame by frame. No manual annotation. No pre-labeled data.The model just sees the court and understands it. Zero-shot classification looks at jersey pixels and decides who plays for whom. Two teams separated in milliseconds. Without ever being told the team names. But here's where it gets wild. Court keypoint detection identifies the geometry of the playing surface. Homography transforms the broadcast camera into a top-down tactical map. Real meters. Real coordinates. From a flat 2D video. Now every player has a position on a real court. Every movement becomes a data point. Every pass becomes a vector. Speed. Distance. Possession time. Pass networks. All extracted from nothing but pixels. No GPS. No chips in the ball. No million-dollar Second Spectrum setup. A Python script and a GPU. The full pipeline is open source. The tutorial walks through every line of code. From detection to transformation to analytics output. Teams used to pay six figures for this data. Now you build it in an afternoon. Computer vision didn't just enter basketball. It democratized it.

zostaff

231,762 views • 6 months ago

We know it took a little time but we are so glad that you’re here for it. Amen is a testament to never giving up and an affirmation of Love being the glue in all times, not just our brightest and darkest ones. We can never do this without any of you ❤️

We know it took a little time but we are so glad that you’re here for it. Amen is a testament to never giving up and an affirmation of Love being the glue in all times, not just our brightest and darkest ones. We can never do this without any of you ❤️

The Heavy

25,724 views • 3 years ago

reminder that this video is 100% AI and it was way cheaper than any real UGC video out there this is an example from my V3 AI UGC system: > the visuals are now more realistic than ever (notice how the natural physics actually look right now) > it's extremely cheap. literally much cheaper than launching a real UGC program for organic > consistent high-quality voice throughout the entire video > infinite-length videos that stay cost effective no matter how long they run AI UGC can match the same exact realism as a real video if you use the right tools and systems..

reminder that this video is 100% AI and it was way cheaper than any real UGC video out there this is an example from my V3 AI UGC system: > the visuals are now more realistic than ever (notice how the natural physics actually look right now) > it's extremely cheap. literally much cheaper than launching a real UGC program for organic > consistent high-quality voice throughout the entire video > infinite-length videos that stay cost effective no matter how long they run AI UGC can match the same exact realism as a real video if you use the right tools and systems..

Miko

36,141 views • 3 months ago

ROSClaw Integration Update Read the text in the video carefully ⊙ Robot is present both physically can in Nvidia Isaac Sim simulation. ⊙ The agent’s initial goal set to explore any object it finds in its spatial view. ⊙ Think of it as if the physical robot is seeing through the simulation. ⊙ ROSClaw vision uses Gemini Live for spatial narration ⊙ An object (a playground slide) is identified and the ROSClaw (OpenClaw) agent takes over. ⊙ The agent initiates action. Attempts to walk towards object. ⊙ None of it is pre-planned. The agent (robot) is acting on its own

ROSClaw Integration Update Read the text in the video carefully ⊙ Robot is present both physically can in Nvidia Isaac Sim simulation. ⊙ The agent’s initial goal set to explore any object it finds in its spatial view. ⊙ Think of it as if the physical robot is seeing through the simulation. ⊙ ROSClaw vision uses Gemini Live for spatial narration ⊙ An object (a playground slide) is identified and the ROSClaw (OpenClaw) agent takes over. ⊙ The agent initiates action. Attempts to walk towards object. ⊙ None of it is pre-planned. The agent (robot) is acting on its own

Irvin (in Japan 🇯🇵)

23,374 views • 7 months ago

An 8-year-old turned an $8 microcontroller into a fully offline, voice-controlled AI storyteller. No Wi-Fi. No cloud. No API. Using an ESP32-S3, INMP 441 mic, i2s speaker and Max 98357A amplifier, Laksh built a bot that generates stories, creates animations, reads them aloud and responds to voice commands: play, pause, resume and next story. He built it by forking the brilliant open-source TinyStories LLM for ESP32 created by Slava S.. It had text. Laksh added voice, interaction, controls and animation. Git repo: Current drawback: bad English TTS on the ESP32. So that is what we are building next.

An 8-year-old turned an $8 microcontroller into a fully offline, voice-controlled AI storyteller. No Wi-Fi. No cloud. No API. Using an ESP32-S3, INMP 441 mic, i2s speaker and Max 98357A amplifier, Laksh built a bot that generates stories, creates animations, reads them aloud and responds to voice commands: play, pause, resume and next story. He built it by forking the brilliant open-source TinyStories LLM for ESP32 created by Slava S.. It had text. Laksh added voice, interaction, controls and animation. Git repo: Current drawback: bad English TTS on the ESP32. So that is what we are building next.

Capt Venkat 🇮🇳

167,029 views • 3 months ago