Name: Run vision AI models in the browser
Uploaded: 2025-11-14T17:21:49.902Z
Duration: 3 min 45 s
Channel: Roboflow
Description: Learn how to get a visual AI model to run in the browser in under 5 minutes. With mobile web, you can get vision AI applications into the hands of millions of users. Big opportunities ahead for physical AI.

Roboflow

49,047 followers

Learn how to get a visual AI model to run in the browser in under 5 minutes. With mobile web, you can get vision AI applications into the hands of millions of users. Big opportunities ahead for physical AI.

Transcript

Do a fun interactive demo where where you'll be able to interact with Microsoft Coco on on your device. So getting started with influence Jazz is, is pretty simple. And this is assuming that you already have a model trains and within the the Roflow ecosystem, you can pull it in from universe or you can pull in your own models. And so once you have a model that you want to bring in, all you have to do is grab the identifier, the version from the model and and your publishable key. And then this is what the integration looks like. So you can pull it in. Yeah, MPM package, you can run it as a a script source tag. And once you instantiate the infer engine, then you just pass your your data images and into the infer engine and and run inference and and get the predictions right away, quote UN quote live coding demo. So this is actually a pre recorded video that I took a couple of days ago. So we'll hop into it. And you can see, so I'm, I'm actually vibe coding here. I think vibe coding is amazing because we're, we're creating a bunch of tools that abstract away just a lot of complexity and inviting more people into the traditional software engineering space. And this is very similar to the goal at robo flow, right, of democratizing visioning and giving people the ability to create computer vision apps, even if they're not and they'll experts, right. So I, I one shot, I, I full transparency, did kind of work on that prompt a little bit, but we're creating a web app here that's going to be integrated with my webcam. And you can see out of the gate, I'm pulling in that Microsoft Coco model from the universe. And so I haven't had to train or provide any data. And now I'm running it on on my webcam. You can see us doing a great job of identifying me as a person. But you probably also notice that it wasn't identifying the iPhone. And this is because of kind of the data that is within the Microsoft Coco data set. It's kind of an older data set. So here we're hopping into a new tool, a new product that we're really excited about at Robo Flow called Rapid. And this is kind of the beginning of what we're thinking of is CV2 dot O right where where AI is is really steering and and humans are just assisting as opposed to traditionally it's been the other way around. So we have a lot of the AI assisted labeling tools in here. You you saw previously that I was generating the predictions for the phone. I can also draw my own custom boxes on here. And then we collect a couple data points. You'll notice that this is video 1st and you're also not really required. To collect that much data, I'm going to collect two more data points here. It was able to run box prompting on that based off of the previous prompts that I had provided it and then immediately you're able to test the model that you built and and run run the model in the browser on whatever the data set is that you provided so in this kids video and you can see like we're actually doing a pretty good job of of of identifying the phone in in that video so I deployed the model and now I'm going to go ahead and. Pull in this Modell's E and the version number and as as this is kind of. Flowing out, one thing you probably notice is that the iPhone at its side wasn't actually getting picked up. And so this is an example of where you would enter into that iterative process, go back to the video, add a label or two at that at that kind of side angle of the phone and it would capture it moving forward. But this is just to show you pretty much how fast you can build and deploy a model on Robo Flow Rapid and then pull it into your client side. Application O We've reached the end of my main talking points and at this point I'd like to invite you to take a couple of minutes. We have a scavenger hunt interactive demo.

To view or add a comment, sign in

More Relevant Posts

Chamith Dilshan

Generative AI Educator | Streamlining Digital Workflows | AI Tools & Productivity Tutorials | Digital Content Creator & Strategist | BBA (Hons) Management & Finance, Marketing Focus at University of Ruhuna (Reading)
2w Edited
Report this post
Are you still using a browser that just shows you websites? It’s time for an upgrade. The new Perplexity AI Comet Browser isn't just a new competitor to Google Chrome; it's a completely new way to use the internet. It transforms your browser from a simple tool into an AI-powered agent that can do the work for you. In this video, I break down the amazing new features that you absolutely need to know. We'll go hands-on to see if Comet is really the future of browsing, or just hype. ⭐ Mind-Blowing Features We Cover: The AI Agent: Watch me give Comet a task (like "Find me the best deals on a new laptop") and see it browse, compare, and fill out forms on its own. Cross-Tab Intelligence: This is the craziest part. I open 10 different tabs and ask the AI one question. It reads all 10 tabs at once to give me a single, summarised answer. Multi-Model Magic: Don't like the answer? Comet lets you switch between different AI models (like GPT-5, Claude, and more) on the fly. Instant Summaries: We test the "Summarize" button on long articles and even entire YouTube videos to see how much time it saves. Productivity Hub: See how it connects to your Gmail and Calendar to draft emails and manage your schedule. By the end of this video, you'll know if it's time to ditch Chrome and switch to the Perplexity Comet browser. #PerplexityAI #CometBrowser #AIBrowser #JustOneClick

Perplexity AI Comet Browser | Amazing Features you need to know

https://www.youtube.com/
Like Comment
To view or add a comment, sign in
Consultium

9 followers
2w
Report this post
400MB. That's what stood between our client and real-time mobile AI. A production Android app trying to upload massive image files to S3 for model inference. Users waiting. Connections timing out. The promise of AI/ML meeting the harsh reality of mobile networks. We could have: ❌ Told them to wait for better network infrastructure ❌ Recommended expensive edge computing solutions ❌ Compromised on model accuracy Instead, we went with model quantization. The result? Real-time inference. Zero compromises. Happy users. The technical win was impressive, but the real victory? Proving that smart optimization beats brute force every time. This is what practical AI engineering looks like—not chasing benchmarks, but solving real problems for real users. #ai #ml #ProblemSolving #CloudSolutions #Innovation #PyTorch
Like Comment
To view or add a comment, sign in
airis

62 followers
1w
Report this post
Can Firefox's AI Window redefine browser interaction with user-controlled AI? 🚀 This innovation from #firefox could enhance #UserExperience. What are the potential benefits and challenges of integrating AI assistants into browsers?

firefoxAIWindow
Like Comment
To view or add a comment, sign in
Babajide Adelowo

Experienced IT Consulting professional Driving Brand Growth with Innovative Strategies & Data-Driven Insights | Web3, AI & Digital Automation Coach.
4w
Report this post
🌐 Exciting news in the tech world! The competition between web browsers is heating up with the help of AI technology 🚀 Learn how browsers are utilizing AI to enhance user experiences and drive innovation in the industry. Stay ahead of the game and check out the full article now! #AI #TechInnovation #WebBrowsers 🔥💻
Like Comment
To view or add a comment, sign in
Ishan Anand

AI Consultant & Educator
3w
Report this post
View Source democratized the web by making every site a learning opportunity. What if we did the same for AI? My CascadiaJS talk shows how transformers work using view source principles to debug AI right in your browser. Web devs curious about AI: this is your on ramp.

1 Comment
Like Comment
To view or add a comment, sign in
Balaprakash AVND ( CA/CS/ACCA/Masters)

20+ years of experience in Finance and Accounts, holding professional qualifications CA | CS | Masters degree in Commerce, Worked extensively in Mfg and Service industry| Vice President Finance|Guest faculty in B-Schools
1w
Report this post
AI-powered browser landscape is heating up intensely in 2025, focusing on intelligence layers and user productivity enhancements rather than traditional measures like speed or UI polish.Perplexity’s Comet, OpenAI’s Atlas significantly challenging Google's Chrome dominance enhanced with AI.
Like Comment
To view or add a comment, sign in
AI For Real

248 followers
4w
Report this post
"AI For Real And Browsers": What Are AI Browsers And How Do They Work? Web browsers are no longer passive instruments. Most of us think of a browser as a passive tool — we click, scroll, search, repeat. But the next wave of browsers doesn’t just show information from the Internet — it acts on it. Eg: AI browsers can not only search for the cheapest available flight but also book a ticket for you. https://lnkd.in/dMW7USHi -Breaking Down AI For Everyone- #AI #AgenticAI #AIBrowsers #OpenAI #Atlas #Comet #ArtificialIntelligence #FutureOfWork #Automation #AIForReal
Like Comment
To view or add a comment, sign in
Trinsic Technologies Inc.

1,794 followers
3w
Report this post
The danger of trusting AI for web searching. This technology still has a long way to go before it's ready for prime time. #AIPromises #AIBubble https://bit.ly/47D9IQP
Like Comment
To view or add a comment, sign in
Art Powell

AI Integration Strategist for SMBs | Turning MSPs into AI-Powered Service Models | 18 Years Scaling Tech + Theological Thought
3w
Report this post
The danger of trusting AI for web searching. This technology still has a long way to go before it's ready for prime time. #AIPromises #AIBubble https://bit.ly/47D9IQP
Like Comment
To view or add a comment, sign in
Stephan Theron

AI Engineer, Business Analyst, iGaming Specialist, Marketing & Strategic Advisor, Martech Solutions, Author
2w
Report this post
Optimising for the Surfaceless Web The article argues that as AI models continuously ingest and compress the web, the visible surface of pages and clicks gives way to a flatter, internalised representation of knowledge. Information that persists is chosen by selection pressures that favour usefulness
Like Comment
To view or add a comment, sign in

49,047 followers

View Profile Connect

Run vision AI models in the browser

Transcript

More Relevant Posts

Perplexity AI Comet Browser | Amazing Features you need to know

https://www.youtube.com/

Explore content categories