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.

I'm obsessed with pushing local small models to their limits. Qwen3.5:0.8b doing real-time video captioning on a Mac Studio M2 Ultra, streaming descriptions as the video plays. Under 1s per frame — 269 frames captured & described from a 3m49s video. Pause anywhere and read the captions, it describes... show more

stevibe

27,081 subscribers

106,441 views • 5 months ago •via X (Twitter)

Science & Technology

Anya Rossi• Live Now

Private livecam show

0 Comments

No comments available

Comments from the original post will appear here

Related Videos

Real-time video captioning in your browser with @LiquidAI's LFM2-VL model on WebGPU. Sending every frame to a server was never going to be the answer. Imagine the bandwidth, latency and cost. Local inference. No server costs. Infinitely scalable. This is the way.

Real-time video captioning in your browser with @LiquidAI's LFM2-VL model on WebGPU. Sending every frame to a server was never going to be the answer. Imagine the bandwidth, latency and cost. Local inference. No server costs. Infinitely scalable. This is the way.

Xenova

48,798 views • 5 months ago

Qwen3.5:9b reasoning head-to-head: Mac Studio M2 Ultra 64GB: 43.08 tok/s Mac Mini M4 16GB: 13.07 tok/s Qwen

Qwen3.5:9b reasoning head-to-head: Mac Studio M2 Ultra 64GB: 43.08 tok/s Mac Mini M4 16GB: 13.07 tok/s Qwen

stevibe

243,408 views • 5 months ago

Update: Qwen3.5:9b head-to-head (MLX, Apple Silicon optimized) Mac Studio M2 Ultra: 89.74 tok/s Mac Mini M4: 20.82 tok/s MLX basically doubles the speed on both machines.

Update: Qwen3.5:9b head-to-head (MLX, Apple Silicon optimized) Mac Studio M2 Ultra: 89.74 tok/s Mac Mini M4: 20.82 tok/s MLX basically doubles the speed on both machines.

stevibe

134,256 views • 5 months ago

A breakthrough in real-time video generation. As a research preview developed with NVIDIA and shared at NVIDIAGTC this week, we trained a new real-time video model running on Vera Rubin. HD videos generate instantly, with time-to-first-frame under 100ms. Unlocking an entirely new creative paradigm and bolstering the foundations of our General World Model, GWM-1. Real-time generation opens a fundamentally different design space for video models and world simulation. We're investing in co-designing our models alongside advances in hardware to keep pushing this frontier.

A breakthrough in real-time video generation. As a research preview developed with NVIDIA and shared at NVIDIAGTC this week, we trained a new real-time video model running on Vera Rubin. HD videos generate instantly, with time-to-first-frame under 100ms. Unlocking an entirely new creative paradigm and bolstering the foundations of our General World Model, GWM-1. Real-time generation opens a fundamentally different design space for video models and world simulation. We're investing in co-designing our models alongside advances in hardware to keep pushing this frontier.

Runway

1,164,026 views • 5 months ago

SOMEONE VIBE CODED A VIDEO STREAM THAT IS SECRETLY 100% TEXT SO IT CANT BE BLOCKED it plays 360p video at 30fps, but theres no actual video on the page. every frame is just colored text characters being repainted on a canvas to the browser its not media at all, its javascript updating some text its called asciline, and here's the trick: > the server decodes the real video and streams it as binary packed text over websockets > the browser paints thousands of colored block characters fast enough to look like 360p > ad blockers and autoplay blockers cant catch it because theres no video element to catch > it streams in kilobytes since its just strings, so it runs on trash internet since the video is literally text, you can apply css glows to it, let people copy paste a moving frame, or feed it straight to a local llm however, an unblockable stream is also an unblockable ad as well

SOMEONE VIBE CODED A VIDEO STREAM THAT IS SECRETLY 100% TEXT SO IT CANT BE BLOCKED it plays 360p video at 30fps, but theres no actual video on the page. every frame is just colored text characters being repainted on a canvas to the browser its not media at all, its javascript updating some text its called asciline, and here's the trick: > the server decodes the real video and streams it as binary packed text over websockets > the browser paints thousands of colored block characters fast enough to look like 360p > ad blockers and autoplay blockers cant catch it because theres no video element to catch > it streams in kilobytes since its just strings, so it runs on trash internet since the video is literally text, you can apply css glows to it, let people copy paste a moving frame, or feed it straight to a local llm however, an unblockable stream is also an unblockable ad as well

Om Patel

1,859,197 views • 2 months ago

Our research team just released Flex-Forcing: a video generation method that lets a single model switch between generation methods at inference time. Right now there are two main approaches to video generation. Bidirectional diffusion models attend to every frame at once, holding structure well at the cost of speed. Autoregressive models generate frame by frame, so they stream fast and scale to long clips, but accumulate error and drift over time. Flex-Forcing trains a single model to do both, letting you choose from the range at inference based on your compute budget.

Our research team just released Flex-Forcing: a video generation method that lets a single model switch between generation methods at inference time. Right now there are two main approaches to video generation. Bidirectional diffusion models attend to every frame at once, holding structure well at the cost of speed. Autoregressive models generate frame by frame, so they stream fast and scale to long clips, but accumulate error and drift over time. Flex-Forcing trains a single model to do both, letting you choose from the range at inference based on your compute budget.

NVIDIA AI

31,412 views • 1 month ago

Runway has unveiled a massive breakthrough in AI video generation at NVIDIA's GTC event, showcasing a new model capable of generating high-definition video in genuine real-time. The new model boasts a "time-to-first-frame" of under 100 milliseconds. This means HD video begins generating and playing almost the instant a command is given. This research preview was co-developed with NVIDIA and runs on their newly announced Vera Rubin architecture. This real-time capability is a foundational step for Runway's General World Model (GWM-1). It pushes the technology beyond simple video generation and into the realm of live, interactive environments.

Runway has unveiled a massive breakthrough in AI video generation at NVIDIA's GTC event, showcasing a new model capable of generating high-definition video in genuine real-time. The new model boasts a "time-to-first-frame" of under 100 milliseconds. This means HD video begins generating and playing almost the instant a command is given. This research preview was co-developed with NVIDIA and runs on their newly announced Vera Rubin architecture. This real-time capability is a foundational step for Runway's General World Model (GWM-1). It pushes the technology beyond simple video generation and into the realm of live, interactive environments.

Wes Roth

82,506 views • 5 months ago

This is how you build the next generation of AI applications. You need an idea, a good model architecture, and a great harness. That's what MUZIM did. It's an app that runs local multimodal and vision models offline to parse and index your images, videos, and documents frame by frame. • Super fast on-device processing • On-device video & image understanding • You can also bring your API keys for heavy reasoning • Your data stays where it is / nothing goes to the cloud You can do a couple of things with this: 1. Vibe searching for whatever you remember in photos or raw video clips. The app will jump straight to the exact frame and timestamp of the video. This is search as it should be. 2. Use local-running agents to perform tasks on your files. I show a couple of examples in my video. You can check it out here: Thanks to the MUZIM by OpenSoul team for showing me the tool and partnering with me on this post.

This is how you build the next generation of AI applications. You need an idea, a good model architecture, and a great harness. That's what MUZIM did. It's an app that runs local multimodal and vision models offline to parse and index your images, videos, and documents frame by frame. • Super fast on-device processing • On-device video & image understanding • You can also bring your API keys for heavy reasoning • Your data stays where it is / nothing goes to the cloud You can do a couple of things with this: 1. Vibe searching for whatever you remember in photos or raw video clips. The app will jump straight to the exact frame and timestamp of the video. This is search as it should be. 2. Use local-running agents to perform tasks on your files. I show a couple of examples in my video. You can check it out here: Thanks to the MUZIM by OpenSoul team for showing me the tool and partnering with me on this post.

Santiago

64,932 views • 8 days ago

🥳We’re releasing StreamDiffusionV2 for the live-stream community—from individual creators with one GPU to enterprise platforms with many. StreamDiffusionV2 is our follow-up to StreamDiffusion: #StreamDiffusion powered real products, but temporal consistency still bugged us. We’re fixing that for the community: 🤔Efficiency ≠ live-readiness: many “real-time” video diffusion models are fast, but live streaming demands per-frame deadlines with low jitter. StreamDiffusionV2 balances temporal stability with strict per-frame deadlines, achieves fast interaction, and is ready for your AI Youtuber Stream Live👇 📒Project page: 💻Code:

🥳We’re releasing StreamDiffusionV2 for the live-stream community—from individual creators with one GPU to enterprise platforms with many. StreamDiffusionV2 is our follow-up to StreamDiffusion: #StreamDiffusion powered real products, but temporal consistency still bugged us. We’re fixing that for the community: 🤔Efficiency ≠ live-readiness: many “real-time” video diffusion models are fast, but live streaming demands per-frame deadlines with low jitter. StreamDiffusionV2 balances temporal stability with strict per-frame deadlines, achieves fast interaction, and is ready for your AI Youtuber Stream Live👇 📒Project page: 💻Code:

Chenfeng_X

20,424 views • 10 months ago

local LLMs on Mac are getting good. but the workflow around models is still messy. you have models in: •⁠ ⁠LM Studio •⁠ ⁠Ollama •⁠ ⁠Hugging Face cache •⁠ ⁠random folders •⁠ ⁠manual downloads and every runtime expects something slightly different. we built ModelHub to clean up this layer. it’s a native macOS menu bar app for discovering, downloading, and managing local LLMs. not a replacement for Ollama, oMLX or LM Studio. it sits beside them. Ollama = run models MLX = Apple Silicon acceleration LM Studio = local model workspace Hugging Face = model ecosystem ModelHub = model discovery + management layer we’re live on Product Hunt today. would love feedback from people running local models on Mac.

local LLMs on Mac are getting good. but the workflow around models is still messy. you have models in: •⁠ ⁠LM Studio •⁠ ⁠Ollama •⁠ ⁠Hugging Face cache •⁠ ⁠random folders •⁠ ⁠manual downloads and every runtime expects something slightly different. we built ModelHub to clean up this layer. it’s a native macOS menu bar app for discovering, downloading, and managing local LLMs. not a replacement for Ollama, oMLX or LM Studio. it sits beside them. Ollama = run models MLX = Apple Silicon acceleration LM Studio = local model workspace Hugging Face = model ecosystem ModelHub = model discovery + management layer we’re live on Product Hunt today. would love feedback from people running local models on Mac.

sabesh 📟

31,497 views • 3 months ago

GPT-4V + TTS = AI Sports narrator 🪄⚽️ Passed every frame of a football video to gpt-4-vision-preview, and with some simple prompting asked to generate a narration No edits, this is as it came out from the model (aka can be SO MUCH BETTER)

GPT-4V + TTS = AI Sports narrator 🪄⚽️ Passed every frame of a football video to gpt-4-vision-preview, and with some simple prompting asked to generate a narration No edits, this is as it came out from the model (aka can be SO MUCH BETTER)

Gonzalo

2,665,480 views • 2 years ago

Most people think a 0.8B model is a toy. 3rd test. This time: OCR I fed Qwen 3.5 0.8B three images: a Kindle page, my terrible handwriting x 2. Mostly correct. Running locally. From a model smaller than your browser tab.

Most people think a 0.8B model is a toy. 3rd test. This time: OCR I fed Qwen 3.5 0.8B three images: a Kindle page, my terrible handwriting x 2. Mostly correct. Running locally. From a model smaller than your browser tab.

stevibe

125,292 views • 5 months ago

EVERYONE IS COPYING AI PROMPTS, BUT ALMOST NOBODY IS COPYING THE SYSTEM BEHIND THEM That is exactly why a small number of creators are generating millions of views while everyone else keeps wondering why the same prompts never work twice. 27 videos → 175,000 subscribers → Millions of views per upload ust a system and the patience to set it up once. The channel is about backyard bunker timelapses. Which sounds absurdly niche. And that's exactly why it works. It's the one content format where: the viewer literally cannot stop watching because they need to see it finished > ChatGPT generates the full construction blueprint, every phase, every prompt, in perfect sequence > Google Whisk creates the opening frame in 9:16 before a single video is generated > Grok animates each phase from the last frame of the previous clip, seamless continuity > the frame extractor is the secret, last frame of clip 1 becomes first frame of clip 2, every time > CapCut stitches it all together, and the raw construction sounds stay in because they're what hooks people

EVERYONE IS COPYING AI PROMPTS, BUT ALMOST NOBODY IS COPYING THE SYSTEM BEHIND THEM That is exactly why a small number of creators are generating millions of views while everyone else keeps wondering why the same prompts never work twice. 27 videos → 175,000 subscribers → Millions of views per upload ust a system and the patience to set it up once. The channel is about backyard bunker timelapses. Which sounds absurdly niche. And that's exactly why it works. It's the one content format where: the viewer literally cannot stop watching because they need to see it finished > ChatGPT generates the full construction blueprint, every phase, every prompt, in perfect sequence > Google Whisk creates the opening frame in 9:16 before a single video is generated > Grok animates each phase from the last frame of the previous clip, seamless continuity > the frame extractor is the secret, last frame of clip 1 becomes first frame of clip 2, every time > CapCut stitches it all together, and the raw construction sounds stay in because they're what hooks people

Shadow Nick

35,606 views • 2 months ago

the real UGC industry isn't dying, it already died created this AI UGC in just a few minutes using my brand new system, from first prompt to final video > product stays in frame exactly where it should > few different people in same one video > can be adapted to any script, character, and environment > voice sounds like a real person, not a robot reading a script > movements are smooth and controlable in every shot > no time limit - make it as long as you need with this level of AI UGC, you can create and test 50x more ads than with real creators while saving $300-$500+ per video

MAX

18,229 views • 4 months ago

Restyling games is much harder than generating a video. Games can’t unintentionally “reset” at every frame: • Your character should stay the same when the camera moves • Damage, lighting, and objects must persist • A room should look the same when you look away and look back And it all has to run in real-time. Reverie v0.1.0 is our first stab at a real-time generative model for playable worlds. Below is a comparison with ours (left) with MirageLSD v2 (right). It’s early and far from perfect but watch what stays consistent as the video continues. 🧵 Prompts below (GPT-5.2).

Restyling games is much harder than generating a video. Games can’t unintentionally “reset” at every frame: • Your character should stay the same when the camera moves • Damage, lighting, and objects must persist • A room should look the same when you look away and look back And it all has to run in real-time. Reverie v0.1.0 is our first stab at a real-time generative model for playable worlds. Below is a comparison with ours (left) with MirageLSD v2 (right). It’s early and far from perfect but watch what stays consistent as the video continues. 🧵 Prompts below (GPT-5.2).

Sharon Lee

30,737 views • 8 months ago

A full-length car commercial made on invideo using Sora 2. AI video is moving fast, the progress is right there for anyone to see. Slop's easier now as well, but great work is easier too. What we make of it is upto us with human creativity still doing the heavy lifting.

A full-length car commercial made on invideo using Sora 2. AI video is moving fast, the progress is right there for anyone to see. Slop's easier now as well, but great work is easier too. What we make of it is upto us with human creativity still doing the heavy lifting.

Sanket Shah

225,401 views • 10 months ago

Pause this video anywhere. Mid-gesture, mid-smirk, mid-word. Every frame you land on was generated in real time, none of it existed a second before you watched it. The tell isn't the lips. Everyone fakes lips now. It's the hands and the face doing what hands and faces do when someone actually means what they're saying. Dead eyes and missing arms are what pre-rendered avatars look like. This is what live ones look like. AI humans that listen, react, gesture, and answer instantly. On your website, around the clock. Come argue with one →

Pause this video anywhere. Mid-gesture, mid-smirk, mid-word. Every frame you land on was generated in real time, none of it existed a second before you watched it. The tell isn't the lips. Everyone fakes lips now. It's the hands and the face doing what hands and faces do when someone actually means what they're saying. Dead eyes and missing arms are what pre-rendered avatars look like. This is what live ones look like. AI humans that listen, react, gesture, and answer instantly. On your website, around the clock. Come argue with one →

Ojin

48,813 views • 28 days ago

AI video generation is moving incredibly fast, and it’s exciting to see tools reaching a level where high-quality results are becoming much more accessible. I’ve been exploring Seedance 2.5 on Morphic and it’s impressive how well it handles motion, prompt understanding, visual consistency, and cinematic quality. It’s the kind of update that gives creators more freedom to experiment without spending hours refining every scene. The pace of innovation in AI video is unreal, and I’m looking forward to pushing this model with even more creative ideas. Created using Seedance 2.5 on Morphic. 🎬✨ What would you create with it first?

AI video generation is moving incredibly fast, and it’s exciting to see tools reaching a level where high-quality results are becoming much more accessible. I’ve been exploring Seedance 2.5 on Morphic and it’s impressive how well it handles motion, prompt understanding, visual consistency, and cinematic quality. It’s the kind of update that gives creators more freedom to experiment without spending hours refining every scene. The pace of innovation in AI video is unreal, and I’m looking forward to pushing this model with even more creative ideas. Created using Seedance 2.5 on Morphic. 🎬✨ What would you create with it first?

Marry Evan

71,490 views • 18 days ago