Loading video...

Video Failed to Load

Go Home

okay another demo fo Gemma4 + Falcon Perception for automated video segmentation & tracking, no human prompts needed the idea: you feed Gemma4 a few sampled frames and ask it to describe what it sees. those descriptions get passed to Falcon Perception which segments and tracks them across the...

dahou_yasser's profile picture

Yasser Dahou

1,011 subscribers

52,338 views • 4 months ago •via X (Twitter)

LIVE

Anya Rossi• Live Now

Private livecam show

0 Comments

No comments available

Comments from the original post will appear here

Related Videos

Next mlx-vlm release will ship with continuous batching support on the server 🚀 What's coming: → Continuous batching — new requests join the active batch immediately, no waiting. Mixed image + text batches supported → OpenAI-compatible API — field-for-field match with mlx-lm, reasoning/content split for thinking models, tag-aware streaming → Multi-turn tool calling — full tool use support across streaming and non-streaming, works with Gemma4 and other templates → Vision feature caching — cache image embeddings across turns. Gemma4: 228x speedup, Qwen3.5: 23x on cache hit All running locally on Apple Silicon. Check our this demo running 4 concurrent requests (mixed image + text) to gemma-4-26B-A4B-IT by Google Gemma in bf16 using Pi + MLX-VLM server on my M3 Ultra. One of the requests ingests a 8K resolution image!
0:52

Next mlx-vlm release will ship with continuous batching support on the server 🚀 What's coming: → Continuous batching — new requests join the active batch immediately, no waiting. Mixed image + text batches supported → OpenAI-compatible API — field-for-field match with mlx-lm, reasoning/content split for thinking models, tag-aware streaming → Multi-turn tool calling — full tool use support across streaming and non-streaming, works with Gemma4 and other templates → Vision feature caching — cache image embeddings across turns. Gemma4: 228x speedup, Qwen3.5: 23x on cache hit All running locally on Apple Silicon. Check our this demo running 4 concurrent requests (mixed image + text) to gemma-4-26B-A4B-IT by Google Gemma in bf16 using Pi + MLX-VLM server on my M3 Ultra. One of the requests ingests a 8K resolution image!

Prince Canuma

82,349 views • 4 months ago

Most AI video tools still feel like traditional editors with a few AI features added in. You still end up adjusting timelines, fixing frames, and spending time editing. But Hailuo AI felt different to me. I gave it a simple prompt and one image, and it generated a full video on its own. No complicated workflow, no manual editing just idea to video in minutes. The text-to-video results are surprisingly good, image-to-video transitions look smooth, and the overall output feels cinematic. What I liked most is the speed. You can quickly test ideas without getting stuck in editing. Try it here: #Hailuo
0:23

Most AI video tools still feel like traditional editors with a few AI features added in. You still end up adjusting timelines, fixing frames, and spending time editing. But Hailuo AI felt different to me. I gave it a simple prompt and one image, and it generated a full video on its own. No complicated workflow, no manual editing just idea to video in minutes. The text-to-video results are surprisingly good, image-to-video transitions look smooth, and the overall output feels cinematic. What I liked most is the speed. You can quickly test ideas without getting stuck in editing. Try it here: #Hailuo

Markandey Sharma

32,326 views • 4 months ago

“I can’t understand why we let someone else rule our land.” Those words are from a song by The Proclaimers called “Cap In Hand,” and they describe what will ultimately happen if we don’t pass the SAVE America Act. We can get this done, but we have to be willing to work for it—and we have to make filibustering senators “earn” their right to filibuster by showing up and speaking on the Senate floor. Death to the Zombie Filibuster™️! Long live the SAVE America Act, and with it popular sovereignty in America.
2:48

“I can’t understand why we let someone else rule our land.” Those words are from a song by The Proclaimers called “Cap In Hand,” and they describe what will ultimately happen if we don’t pass the SAVE America Act. We can get this done, but we have to be willing to work for it—and we have to make filibustering senators “earn” their right to filibuster by showing up and speaking on the Senate floor. Death to the Zombie Filibuster™️! Long live the SAVE America Act, and with it popular sovereignty in America.

Mike Lee

77,337 views • 1 month ago

If you can think it, you can make it. That used to be a cliché. Now it's literally how video editing works. A year ago, making motion graphics meant learning After Effects, spending hours on keyframes, or hiring a designer. Today, you describe what you want in one sentence and it appears on your timeline. They're code-based, fully editable, exportable with transparency. You can tweak every parameter by hand or ask the agent to build you a custom control panel for the exact thing you want to adjust. The barrier to creating professional video content has disappeared. You don't need a specialized degree. You don't need expensive software. You need an idea. This is what we're building at ChatCut. Full demo 👇
1:28

If you can think it, you can make it. That used to be a cliché. Now it's literally how video editing works. A year ago, making motion graphics meant learning After Effects, spending hours on keyframes, or hiring a designer. Today, you describe what you want in one sentence and it appears on your timeline. They're code-based, fully editable, exportable with transparency. You can tweak every parameter by hand or ask the agent to build you a custom control panel for the exact thing you want to adjust. The barrier to creating professional video content has disappeared. You don't need a specialized degree. You don't need expensive software. You need an idea. This is what we're building at ChatCut. Full demo 👇

ChatCut

218,958 views • 4 months ago

OpenClaw just got a free image generator built directly inside it. And it's powered by Google's fastest model ever. Here's what you can do now: → Describe a thumbnail and OpenClaw generates it instantly. → Post it straight to Discord or Telegram automatically. → Create a dedicated channel just for generating visuals on demand. → Build ads, social content, and landing page graphics without switching apps. → Set it up with one API key and a single skill file. No clicking between tools. No switching apps. No waiting. You describe it. OpenClaw builds it. Done. Most people are still bouncing between 5 different image tools every day. The 1% just automated the whole thing inside their AI agent. Want the skill file and exact setup prompt? Drop a comment below. 👇
12:25

OpenClaw just got a free image generator built directly inside it. And it's powered by Google's fastest model ever. Here's what you can do now: → Describe a thumbnail and OpenClaw generates it instantly. → Post it straight to Discord or Telegram automatically. → Create a dedicated channel just for generating visuals on demand. → Build ads, social content, and landing page graphics without switching apps. → Set it up with one API key and a single skill file. No clicking between tools. No switching apps. No waiting. You describe it. OpenClaw builds it. Done. Most people are still bouncing between 5 different image tools every day. The 1% just automated the whole thing inside their AI agent. Want the skill file and exact setup prompt? Drop a comment below. 👇

Julian Goldie SEO

11,862 views • 6 months ago

Do our Puranas really contradict each other? 🤔 Or is this just another half-baked propaganda spread by critics of Sanatan Dharma? In this video, it is claimed that because different Puranas describe different creators, at least three of them must be false. But the truth is that you cannot understand PhD-level Vedic knowledge with a kindergarten-level mindset! The Rigveda declares: "Ekam Sat Vipra Bahudha Vadanti" — Truth is One; the wise describe it in many ways. The Supreme Reality is one, though it is worshipped and explained through different forms and perspectives. Unless you understand the concepts of Kalpa Bheda (differences across cosmic cycles) and the idea of a multiverse, these descriptions may appear contradictory. In reality, they represent different contexts within the vast framework of Vedic cosmology. Watch the video till the end to learn the complete explanation and the reasoning behind these apparent differences. 🌌✨ For more scriptural and logical responses to common misconceptions, follow us now and share this video so the truth reaches everyone who has been misled.
2:53

Do our Puranas really contradict each other? 🤔 Or is this just another half-baked propaganda spread by critics of Sanatan Dharma? In this video, it is claimed that because different Puranas describe different creators, at least three of them must be false. But the truth is that you cannot understand PhD-level Vedic knowledge with a kindergarten-level mindset! The Rigveda declares: "Ekam Sat Vipra Bahudha Vadanti" — Truth is One; the wise describe it in many ways. The Supreme Reality is one, though it is worshipped and explained through different forms and perspectives. Unless you understand the concepts of Kalpa Bheda (differences across cosmic cycles) and the idea of a multiverse, these descriptions may appear contradictory. In reality, they represent different contexts within the vast framework of Vedic cosmology. Watch the video till the end to learn the complete explanation and the reasoning behind these apparent differences. 🌌✨ For more scriptural and logical responses to common misconceptions, follow us now and share this video so the truth reaches everyone who has been misled.

श्री

10,161 views • 1 month ago

Mini tutorial for Google DeepMind 's VEO 2 Prompt Buddy. (Works well with other AI video tools as well). I’ve created my VEO 2 Prompt Buddy Custom GPT to help me go beyond my own brain and prompts. It can describe your images, enhance your prompts, be creative and more. It will provide you 4 different sections to choose from. I’ve also added a camera combinations section inspired by the Hailuo I2V-01-Director model for testing purposes. It’s currently in BETA mode in my GPT because, sometimes, multiple camera movements don’t work optimally in VEO 2 right now. Output sections: 1. PRO – Provides detailed, professional, and cinematically accurate results. 2. OUT OF THE BOX – Creates something more creative and unexpected from your idea. 3. ULTRA DETAILED – Expands your idea with extreme precision and even more detail. 4. CAMERA COMBINATIONS – Generates a version with cinematic camera transitions. This hot sauce is FREE for a limited time! Hope you have fun with this one. Link down below: --- All results are text-to-video
1:21

Mini tutorial for Google DeepMind 's VEO 2 Prompt Buddy. (Works well with other AI video tools as well). I’ve created my VEO 2 Prompt Buddy Custom GPT to help me go beyond my own brain and prompts. It can describe your images, enhance your prompts, be creative and more. It will provide you 4 different sections to choose from. I’ve also added a camera combinations section inspired by the Hailuo I2V-01-Director model for testing purposes. It’s currently in BETA mode in my GPT because, sometimes, multiple camera movements don’t work optimally in VEO 2 right now. Output sections: 1. PRO – Provides detailed, professional, and cinematically accurate results. 2. OUT OF THE BOX – Creates something more creative and unexpected from your idea. 3. ULTRA DETAILED – Expands your idea with extreme precision and even more detail. 4. CAMERA COMBINATIONS – Generates a version with cinematic camera transitions. This hot sauce is FREE for a limited time! Hope you have fun with this one. Link down below: --- All results are text-to-video

Allar Haltsonen

58,509 views • 1 year ago

Really happy to see the interest around our “Hands-on with Gemini” video. In our developer blog yesterday, we broke down how Gemini was used to create it. We gave Gemini sequences of different modalities — image and text in this case — and had it respond by predicting what might come next. Devs can try similar things when access to Pro opens on 12/13 🚀. The knitting demo used Ultra⚡ All the user prompts and outputs in the video are real, shortened for brevity. The video illustrates what the multimodal user experiences built with Gemini could look like. We made it to inspire developers. When you’re building an app, you can get similar results (there’s always some variability with LLMs) by prompting Gemini with an instruction that allows the user to "configure" the behavior of the model, like inputting “you are an expert in science …” before a user can engage in the same kind of back and forth dialogue. Here’s a clip of what this looks like in AI Studio with Gemini Pro. We’ve come a long way since Flamingo 🦩 & PALI, looking forward to seeing what people build with it!
2:24

Really happy to see the interest around our “Hands-on with Gemini” video. In our developer blog yesterday, we broke down how Gemini was used to create it. We gave Gemini sequences of different modalities — image and text in this case — and had it respond by predicting what might come next. Devs can try similar things when access to Pro opens on 12/13 🚀. The knitting demo used Ultra⚡ All the user prompts and outputs in the video are real, shortened for brevity. The video illustrates what the multimodal user experiences built with Gemini could look like. We made it to inspire developers. When you’re building an app, you can get similar results (there’s always some variability with LLMs) by prompting Gemini with an instruction that allows the user to "configure" the behavior of the model, like inputting “you are an expert in science …” before a user can engage in the same kind of back and forth dialogue. Here’s a clip of what this looks like in AI Studio with Gemini Pro. We’ve come a long way since Flamingo 🦩 & PALI, looking forward to seeing what people build with it!

Oriol Vinyals

180,996 views • 2 years ago

Last week, we shared a refresher on our favorite Flow features. Today, we're doing a deep dive into prompting with a few tips on how to craft the optimal prompt, create your dream video, and #FindYourFlow 🕺 Clearly identify your characters or objects and describe their movements 🎥 Frame your shot with compositional terms like "wide shot" or “close-up" and direct the camera with instructions like "tracking shot" or "aerial view" 💡The lighting and environment set the entire mood. Instead of "a room" use more descriptive terms like "a dusty attic filled with forgotten treasures, a single beam of afternoon light cutting through a grimy window" 🧸 Flow is not limited to realistic visuals. You can experiment with a wide array of animation styles like "stop motion" or "knitted animation" 🗣️ With Veo 3, you can include sound effects and dialogue in your prompt. Describe the ambient sounds happening throughout the scene and generate dialogue by including it in your prompt. For more Flow tips, see here ➡️ 🚨Pro tip: this is another great bookmark 😉
0:26

Last week, we shared a refresher on our favorite Flow features. Today, we're doing a deep dive into prompting with a few tips on how to craft the optimal prompt, create your dream video, and #FindYourFlow 🕺 Clearly identify your characters or objects and describe their movements 🎥 Frame your shot with compositional terms like "wide shot" or “close-up" and direct the camera with instructions like "tracking shot" or "aerial view" 💡The lighting and environment set the entire mood. Instead of "a room" use more descriptive terms like "a dusty attic filled with forgotten treasures, a single beam of afternoon light cutting through a grimy window" 🧸 Flow is not limited to realistic visuals. You can experiment with a wide array of animation styles like "stop motion" or "knitted animation" 🗣️ With Veo 3, you can include sound effects and dialogue in your prompt. Describe the ambient sounds happening throughout the scene and generate dialogue by including it in your prompt. For more Flow tips, see here ➡️ 🚨Pro tip: this is another great bookmark 😉

Google Labs

33,904 views • 1 year ago

We all know the pain of choosing the right model. You have an idea, testing it one model at a time in ten different tabs, using a spreadsheet to track usable outputs. Stop guessing and start building with Creator Central by Atlas Cloud. Model Explorer runs the same prompt across models and puts the results side by side, up to 10 at once. Start from a curated set or assemble your own. With Wan3.0 coming shortly, we're dropping the price of Seedance 2.5 so creators can compare and choose the right model for their projects. Create something now with our Generator templated workflow: Pick what you like and access the exact prompt and settings behind that result, ready for you to edit. #AIVideo #GenerativeAI #VideoProduction #Seedance #Wan3
1:02

We all know the pain of choosing the right model. You have an idea, testing it one model at a time in ten different tabs, using a spreadsheet to track usable outputs. Stop guessing and start building with Creator Central by Atlas Cloud. Model Explorer runs the same prompt across models and puts the results side by side, up to 10 at once. Start from a curated set or assemble your own. With Wan3.0 coming shortly, we're dropping the price of Seedance 2.5 so creators can compare and choose the right model for their projects. Create something now with our Generator templated workflow: Pick what you like and access the exact prompt and settings behind that result, ready for you to edit. #AIVideo #GenerativeAI #VideoProduction #Seedance #Wan3

Atlas Cloud

640,724 views • 24 days ago

SAM-2 is apache license - you can indeed just build crazy apps with them without fear. This weekend I ported SAM-2 models to run optimally on Apple Silicon. I have also quantized all the SAM models to 16bit, 8bit, and 4bit variants (collection below) Repo is open source and pip installable. Comes with all you'll need for interactive video segmentation - - bidirectional propagation, - positive and negative clicks, - bounding boxes, - streaming masks back to user, - lightweight fast-api server that you can extend I basically asked Codex to port over the original Facebook sam2 codebase to mlx and optimize it for mlx. And it just worked. This demo for example is a 4bit base-plus model (~100MB download) To speed up, you can also do spatial or temporal downsampling. That speeds inference on my old macbook by ~3x and the hit in parity is negligible.
0:53

SAM-2 is apache license - you can indeed just build crazy apps with them without fear. This weekend I ported SAM-2 models to run optimally on Apple Silicon. I have also quantized all the SAM models to 16bit, 8bit, and 4bit variants (collection below) Repo is open source and pip installable. Comes with all you'll need for interactive video segmentation - - bidirectional propagation, - positive and negative clicks, - bounding boxes, - streaming masks back to user, - lightweight fast-api server that you can extend I basically asked Codex to port over the original Facebook sam2 codebase to mlx and optimize it for mlx. And it just worked. This demo for example is a 4bit base-plus model (~100MB download) To speed up, you can also do spatial or temporal downsampling. That speeds inference on my old macbook by ~3x and the hit in parity is negligible.

AVB

17,428 views • 3 months ago

🎥 Today we’re premiering Meta Movie Gen: the most advanced media foundation models to-date. Developed by AI research teams at Meta, Movie Gen delivers state-of-the-art results across a range of capabilities. We’re excited for the potential of this line of research to usher in entirely new possibilities for casual creators and creative professionals alike. More details and examples of what Movie Gen can do ➡️ 🛠️ Movie Gen models and capabilities Movie Gen Video: 30B parameter transformer model that can generate high-quality and high-definition images and videos from a single text prompt. Movie Gen Audio: A 13B parameter transformer model that can take a video input along with optional text prompts for controllability to generate high-fidelity audio synced to the video. It can generate ambient sound, instrumental background music and foley sound — delivering state-of-the-art results in audio quality, video-to-audio alignment and text-to-audio alignment. Precise video editing: Using a generated or existing video and accompanying text instructions as an input it can perform localized edits such as adding, removing or replacing elements — or global changes like background or style changes. Personalized videos: Using an image of a person and a text prompt, the model can generate a video with state-of-the-art results on character preservation and natural movement in video. We’re continuing to work closely with creative professionals from across the field to integrate their feedback as we work towards a potential release. We look forward to sharing more on this work and the creative possibilities it will enable in the future.
0:30

🎥 Today we’re premiering Meta Movie Gen: the most advanced media foundation models to-date. Developed by AI research teams at Meta, Movie Gen delivers state-of-the-art results across a range of capabilities. We’re excited for the potential of this line of research to usher in entirely new possibilities for casual creators and creative professionals alike. More details and examples of what Movie Gen can do ➡️ 🛠️ Movie Gen models and capabilities Movie Gen Video: 30B parameter transformer model that can generate high-quality and high-definition images and videos from a single text prompt. Movie Gen Audio: A 13B parameter transformer model that can take a video input along with optional text prompts for controllability to generate high-fidelity audio synced to the video. It can generate ambient sound, instrumental background music and foley sound — delivering state-of-the-art results in audio quality, video-to-audio alignment and text-to-audio alignment. Precise video editing: Using a generated or existing video and accompanying text instructions as an input it can perform localized edits such as adding, removing or replacing elements — or global changes like background or style changes. Personalized videos: Using an image of a person and a text prompt, the model can generate a video with state-of-the-art results on character preservation and natural movement in video. We’re continuing to work closely with creative professionals from across the field to integrate their feedback as we work towards a potential release. We look forward to sharing more on this work and the creative possibilities it will enable in the future.

AI at Meta

2,266,357 views • 1 year ago

AI + robotics research is starting to pick up steam. We can now instruct Spot using natural language using Language-guided Skill Coordination (LSC). A user provides a natural language instruction: "Bring me the chocolates box, cereal box, and pill bottle, and put them on the bedroom table" and the robot navigates to the location of the target objects and places them on the room table. Current versions of Spot rely on a fixed-set vocabulary that cannot generalize to diverse instructions. In this demo, the researchers presented a method that uses large language models to receive a free-form natural language instruction for object rearrangement, which it then executes. It combines: -A voice-to-text model that processes the instructions into text -An LLM that converts natural language instruction to call a library of skills -A perception module that provides ground truth locations and visual object detection And what your left with is a glimpse into what our future will hold: robots executing tasks from human voice prompts.
0:30

AI + robotics research is starting to pick up steam. We can now instruct Spot using natural language using Language-guided Skill Coordination (LSC). A user provides a natural language instruction: "Bring me the chocolates box, cereal box, and pill bottle, and put them on the bedroom table" and the robot navigates to the location of the target objects and places them on the room table. Current versions of Spot rely on a fixed-set vocabulary that cannot generalize to diverse instructions. In this demo, the researchers presented a method that uses large language models to receive a free-form natural language instruction for object rearrangement, which it then executes. It combines: -A voice-to-text model that processes the instructions into text -An LLM that converts natural language instruction to call a library of skills -A perception module that provides ground truth locations and visual object detection And what your left with is a glimpse into what our future will hold: robots executing tasks from human voice prompts.

AI Breakfast

33,725 views • 3 years ago

"I've spent $30K on a launch video, now you can make one with code." Here's my new episode with Bin Liu and Jake Moran, where they shared their 5-step playbook for using HyperFrames to make professional launch videos in Codex and Claude Code: 1. Gather video assets and create a frame.md 2. Create a storyboard.md with the scenes 3. Pull animations from the HyperFrames repo 4. Generate and review static frames 5. Build the video with HyperFrames Studio You can also start simple by using the built-in /website-to-video skill to one shot a launch video from any URL. HyperFrames is 100% free and I love how it democratizes video creation for everyone. 📌 Watch now: Thanks to our sponsors: Wispr Flow: 4x faster than typing with your voice Riverside: All-in-one AI studio for podcasts and video
0:43

"I've spent $30K on a launch video, now you can make one with code." Here's my new episode with Bin Liu and Jake Moran, where they shared their 5-step playbook for using HyperFrames to make professional launch videos in Codex and Claude Code: 1. Gather video assets and create a frame.md 2. Create a storyboard.md with the scenes 3. Pull animations from the HyperFrames repo 4. Generate and review static frames 5. Build the video with HyperFrames Studio You can also start simple by using the built-in /website-to-video skill to one shot a launch video from any URL. HyperFrames is 100% free and I love how it democratizes video creation for everyone. 📌 Watch now: Thanks to our sponsors: Wispr Flow: 4x faster than typing with your voice Riverside: All-in-one AI studio for podcasts and video

Peter Yang

70,122 views • 2 months ago

x402 payments have arrived on Wurk!🤖 Agents can now interact with humans, asking for help and rewarding them. Imagine an agent getting stuck because it can’t find an answer or lacks the right tools. It can now create a microjob on Wurk, describe what it needs, and humans can respond with solutions. 💡 Use cases: - Blurred image clarification - Solving CAPTCHAs - Interpreting humor - Visiting protected pages - Design feedback - Training or testing tasks - And much more Beyond microjobs, agents can also open social jobs and custom jobs, like meme contests or creative challenges. ⚙️What is x402? x402 is an emerging standard for machine-to-machine payments, developed by Coinbase Developer Platform🛡️ and brought to Solana by PayAI Network | x402 Facilitator Are you a developer and ready to build? DM us! Most agents run in the background and we will never see them. To visualize it and make the concept easy to understand, we built a simple demo agent using Wurk’s infrastructure to get human assistance. Check it out👇
2:26

x402 payments have arrived on Wurk!🤖 Agents can now interact with humans, asking for help and rewarding them. Imagine an agent getting stuck because it can’t find an answer or lacks the right tools. It can now create a microjob on Wurk, describe what it needs, and humans can respond with solutions. 💡 Use cases: - Blurred image clarification - Solving CAPTCHAs - Interpreting humor - Visiting protected pages - Design feedback - Training or testing tasks - And much more Beyond microjobs, agents can also open social jobs and custom jobs, like meme contests or creative challenges. ⚙️What is x402? x402 is an emerging standard for machine-to-machine payments, developed by Coinbase Developer Platform🛡️ and brought to Solana by PayAI Network | x402 Facilitator Are you a developer and ready to build? DM us! Most agents run in the background and we will never see them. To visualize it and make the concept easy to understand, we built a simple demo agent using Wurk’s infrastructure to get human assistance. Check it out👇

WURK.FUN

147,294 views • 10 months ago

We made a thing! Very happy to announce sqlcoder-pro and the Defog Alignment Platform. Available to use immediately without a wait-list, weights will be open-sourced very soon. The video does a quick show and tell comparison against ChatGPT (with gpt-4o). Read on for more details! TLDR 💪 equal (or better) performance on text-to-SQL as the most capable Claude-3.5 or GPT-4 models 🤝 You can use it today on a free plan/free trial, without a waitlist 🪽 self-hostable on a single RTX4090, with 2 second median generation times for SQL queries 🔁 exactly the same output every time, give the same prompt 👨🏻‍🏫 teachable and steerable: show the model what you want it to do 🛞 debuggable – you can understand WTF is going on inside the model, instead of treating it like a black box Let's dig into each of these one-by-one! Performance SQLCoder-8b-pro significantly exceeds the performance of our previous sqlcoder-8b model on Postgres text-to-SQL (from 88.2% to 90.2% accuracy - gpt-4o is at 87.6%, for reference). It is also better at following instructions. This was done via self-merges, hand crafted fine-tuning data, and adapting the training data to fit our tokenizer. Cost You can host this on the model on a single $3,500 RTX4090, and support ~5 requests/second via VLLM. If you're looking to host on the cloud instead, you can run it on a single L4 GPU that costs $300/mo on GCP Repeatability We have a dense 8b model with no MoE shenanigans. For the same prompt with temperature=0, you'll always get the same answer – which is critical in BI. Teachable In our alignment and feedback modes, you can give the model feedback on how it answered certain questions, and it will automatically adapt to the feedback. Debuggable You can use logprobs and attention scores to determine where, exactly is the model paying attention to inside a prompt + what it's getting confused by when generating outputs. Available today You can use Defog on the cloud today by going to docs[dot]defog[dot]ai, and getting an API key. Excited to hear what you think!
2:33

We made a thing! Very happy to announce sqlcoder-pro and the Defog Alignment Platform. Available to use immediately without a wait-list, weights will be open-sourced very soon. The video does a quick show and tell comparison against ChatGPT (with gpt-4o). Read on for more details! TLDR 💪 equal (or better) performance on text-to-SQL as the most capable Claude-3.5 or GPT-4 models 🤝 You can use it today on a free plan/free trial, without a waitlist 🪽 self-hostable on a single RTX4090, with 2 second median generation times for SQL queries 🔁 exactly the same output every time, give the same prompt 👨🏻‍🏫 teachable and steerable: show the model what you want it to do 🛞 debuggable – you can understand WTF is going on inside the model, instead of treating it like a black box Let's dig into each of these one-by-one! Performance SQLCoder-8b-pro significantly exceeds the performance of our previous sqlcoder-8b model on Postgres text-to-SQL (from 88.2% to 90.2% accuracy - gpt-4o is at 87.6%, for reference). It is also better at following instructions. This was done via self-merges, hand crafted fine-tuning data, and adapting the training data to fit our tokenizer. Cost You can host this on the model on a single $3,500 RTX4090, and support ~5 requests/second via VLLM. If you're looking to host on the cloud instead, you can run it on a single L4 GPU that costs $300/mo on GCP Repeatability We have a dense 8b model with no MoE shenanigans. For the same prompt with temperature=0, you'll always get the same answer – which is critical in BI. Teachable In our alignment and feedback modes, you can give the model feedback on how it answered certain questions, and it will automatically adapt to the feedback. Debuggable You can use logprobs and attention scores to determine where, exactly is the model paying attention to inside a prompt + what it's getting confused by when generating outputs. Available today You can use Defog on the cloud today by going to docs[dot]defog[dot]ai, and getting an API key. Excited to hear what you think!

Rishabh Srivastava

13,469 views • 2 years ago

GPT Image 2 - Character Reference Sheet Prompt You can create a character reference sheet in any style you want for use in your Seedance 2.0 videos. All you need to do is either provide a reference image or describe your character. You can also input random values to generate completely random characters. This reference sheet provides a clear, structured, and highly readable visual breakdown of the character, helping the model maintain strong identity consistency across different outputs. The layout ensures the model focuses on essential visual information rather than stylistic noise, while the inclusion of expressions, head angles, and performance cues improves the model’s ability to generate believable acting and varied poses without losing character integrity. All clips in the demonstration video were generated using the reference sheet image and the same Seedance 2.0 prompt. You can find the prompts in the replies. 👇 GPT Image 2 and Seedance 2.0 are both available on mitte.ai
1:15

GPT Image 2 - Character Reference Sheet Prompt You can create a character reference sheet in any style you want for use in your Seedance 2.0 videos. All you need to do is either provide a reference image or describe your character. You can also input random values to generate completely random characters. This reference sheet provides a clear, structured, and highly readable visual breakdown of the character, helping the model maintain strong identity consistency across different outputs. The layout ensures the model focuses on essential visual information rather than stylistic noise, while the inclusion of expressions, head angles, and performance cues improves the model’s ability to generate believable acting and varied poses without losing character integrity. All clips in the demonstration video were generated using the reference sheet image and the same Seedance 2.0 prompt. You can find the prompts in the replies. 👇 GPT Image 2 and Seedance 2.0 are both available on mitte.ai

Kōda

164,464 views • 4 months ago

AI ads have reached a point where it outperforms traditional UGC on both cost AND quality you can create everything and anything with it and the only thing stopping you is your creativity realistic UGC videos, ai animations or mix of both together (this one performed suprisingly well for us last month) for over a year now we've been using it across all my and our clients' offers, 100% legal, without deepfaking anyone or risk of being sued i cracked the code to create videos that look real, stop the scroll, convert like crazy, and are easy to automate i'm finally sharing my FULL system with you, here's what you're getting: - all the prompts im using - full step by step playbook from zero to edited ad - how i create them in less than 15mins and $2 - how to automate full pipeline and scale even more RT + reply 'ADS' and i'll send you the step-by-step system (must follow so i can dm)
1:21

AI ads have reached a point where it outperforms traditional UGC on both cost AND quality you can create everything and anything with it and the only thing stopping you is your creativity realistic UGC videos, ai animations or mix of both together (this one performed suprisingly well for us last month) for over a year now we've been using it across all my and our clients' offers, 100% legal, without deepfaking anyone or risk of being sued i cracked the code to create videos that look real, stop the scroll, convert like crazy, and are easy to automate i'm finally sharing my FULL system with you, here's what you're getting: - all the prompts im using - full step by step playbook from zero to edited ad - how i create them in less than 15mins and $2 - how to automate full pipeline and scale even more RT + reply 'ADS' and i'll send you the step-by-step system (must follow so i can dm)

MAX

22,574 views • 1 month ago

VEO 2 by Google DeepMind : MY CHEAT SHEET Alright, so after 500h-ish spent on VEO and giving birth to both "Kitsune" and "Banished", tons of people asked for a making-of. Instead, I decided to give you what I actually know of VEO 2 to this day. Please share! it's made to be spread around! 1/ If you're not using a LLM (Gemini, ChatGPT, whatever), you're doing it wrong. VEO 2 currently has a sweet spot when it comes to prompt length: too short is poor, too long drops information, action, description etc. I did a lot of back and forth to find my sweet spot, but once I got in a place I thought felt right, I used a LLM to help me keep my structure, length, and help me draft actions. I would then spent an extensive amount of time tweaking, iterating, removing words, changing order, adding others, but the draft would come from a LLM and a conversation I built and trained to understand what my structure looked like, what was a success, or a failure. I would also share the prompts working well for further reference, and sharing the failures also for further reference. This would ensure my LLM conversation became a true companion. 2/ Structure, structure, structure Structure is important. Each recipe is different but same as any GenAI text-to something, it looks like the "higher on the prompt has more weight" rule applies. So, in my case I would start by describing the aesthetics I am looking for, time of day, colors, mood, then move to camera, subject, action, and all the rest. Once again, you might have a different experience but what is important is to stick to whatever structure you have as you move forward. Keeping it organized also makes it easier to edit later. 3/ Only describe what you see in the frame If you have a character you want to keep consistent, but you want a close-up on the face for example, your reflex will be to describe the character from head to toe and then mention you want a close-up...It's not that simple. If I tell VEO I want a face close-up but then proceed to describe the character's feet, the close-up mention will be dropped by VEO... Once again, the LLM can help you in this by giving it the instruction to only describe what is in the frame. 4/ Patience Well, it can get costly to be patient, but even if you repeat the same structure, sometimes changing one word can still throw the entire thing out and totally change the aesthetics of your scene. It is by nature extremely consistent if you conserve most words, but sometimes it happens. In those situations, trace your steps back and try to figure out which words are triggering a larger change. 5/ Documenting When I started "Kitsune" (and did the same for all others), the first thing I did was start a Figjam file so I could save the successful prompts and come back to them for future reference. Why Figjam? So I could also upload 1 to 4 generations from this prompt, and browse through them in the future. 6/ VEO is the Midjourney of video Currently, no text-to-video tool (Minimax being the closest behind) gave me a feeling I could provide strong art directions and actually get them. I have been a designer for nearly 20 years, and art direction to me has been one of the strongest foundations of most of my work. Dark, light, happy, sad, colorful or not, it doesn't matter as long as you have a point of view and please...have a point of view. Recently watched a great video about the slow death of art direction in film (link in comments) and oh boy, did VEO 2 deliver on giving me the feeling I was listened. Try starting your prompts with different kinds of medium (watercolor for example), the mood you are trying to achieve, the kind of lighting you want, the dust in the rays of light, etc... which gets me to the next one 7/ You can direct your colors in VEO It's as simple as mentioning the hues you want to have in the final result, in which quantity, and where. When I direct shots, I am constantly describing colors for two reasons: 1. Well, having a point of view and 2. reaching better consistency through text-to-video. If I have a strong and consistent mood but my character is slightly different because of text-to-video, the impact won't be dramatic because a strong art direction helps a lot with consistency. 8/ Describe your life away Some people asked me how I achieved a good consistency between shots knowing it's only text-to-video and the answer is simple: I describe my characters, their unique traits, their clothing, their haircut, etc..anything which could help someone visually impaired have a very precise mental representation of the subject. 9/ But don't describe too much either... It would be magical if you could stuff 3000 words in the window and have exactly what you asked for, right? Well, it turns out VEO is amazing with its prompt adherence, but there is always a moment where it starts dropping animations or visual elements when your prompt stretches for a tad too long. This actually happens way before the character limit allowed by VEO is reached, so don't overdo it, it's no use and will play against the results. For info, 200-250 words seems like a sweet spot! 10/ Natural movements but... VEO is great with natural movements and this is also one of the reasons why I used it so extensively: people walking don't walk in slow-motion. That being said, don't try to be too ambitious on some of the expected movements: multiple camera movements won't work, full 360 revolutions around a subject won't work, anime-style crazy camera movements won't work, etc... what it can do is already great, but there are still some limitations...
1:56

VEO 2 by Google DeepMind : MY CHEAT SHEET Alright, so after 500h-ish spent on VEO and giving birth to both "Kitsune" and "Banished", tons of people asked for a making-of. Instead, I decided to give you what I actually know of VEO 2 to this day. Please share! it's made to be spread around! 1/ If you're not using a LLM (Gemini, ChatGPT, whatever), you're doing it wrong. VEO 2 currently has a sweet spot when it comes to prompt length: too short is poor, too long drops information, action, description etc. I did a lot of back and forth to find my sweet spot, but once I got in a place I thought felt right, I used a LLM to help me keep my structure, length, and help me draft actions. I would then spent an extensive amount of time tweaking, iterating, removing words, changing order, adding others, but the draft would come from a LLM and a conversation I built and trained to understand what my structure looked like, what was a success, or a failure. I would also share the prompts working well for further reference, and sharing the failures also for further reference. This would ensure my LLM conversation became a true companion. 2/ Structure, structure, structure Structure is important. Each recipe is different but same as any GenAI text-to something, it looks like the "higher on the prompt has more weight" rule applies. So, in my case I would start by describing the aesthetics I am looking for, time of day, colors, mood, then move to camera, subject, action, and all the rest. Once again, you might have a different experience but what is important is to stick to whatever structure you have as you move forward. Keeping it organized also makes it easier to edit later. 3/ Only describe what you see in the frame If you have a character you want to keep consistent, but you want a close-up on the face for example, your reflex will be to describe the character from head to toe and then mention you want a close-up...It's not that simple. If I tell VEO I want a face close-up but then proceed to describe the character's feet, the close-up mention will be dropped by VEO... Once again, the LLM can help you in this by giving it the instruction to only describe what is in the frame. 4/ Patience Well, it can get costly to be patient, but even if you repeat the same structure, sometimes changing one word can still throw the entire thing out and totally change the aesthetics of your scene. It is by nature extremely consistent if you conserve most words, but sometimes it happens. In those situations, trace your steps back and try to figure out which words are triggering a larger change. 5/ Documenting When I started "Kitsune" (and did the same for all others), the first thing I did was start a Figjam file so I could save the successful prompts and come back to them for future reference. Why Figjam? So I could also upload 1 to 4 generations from this prompt, and browse through them in the future. 6/ VEO is the Midjourney of video Currently, no text-to-video tool (Minimax being the closest behind) gave me a feeling I could provide strong art directions and actually get them. I have been a designer for nearly 20 years, and art direction to me has been one of the strongest foundations of most of my work. Dark, light, happy, sad, colorful or not, it doesn't matter as long as you have a point of view and please...have a point of view. Recently watched a great video about the slow death of art direction in film (link in comments) and oh boy, did VEO 2 deliver on giving me the feeling I was listened. Try starting your prompts with different kinds of medium (watercolor for example), the mood you are trying to achieve, the kind of lighting you want, the dust in the rays of light, etc... which gets me to the next one 7/ You can direct your colors in VEO It's as simple as mentioning the hues you want to have in the final result, in which quantity, and where. When I direct shots, I am constantly describing colors for two reasons: 1. Well, having a point of view and 2. reaching better consistency through text-to-video. If I have a strong and consistent mood but my character is slightly different because of text-to-video, the impact won't be dramatic because a strong art direction helps a lot with consistency. 8/ Describe your life away Some people asked me how I achieved a good consistency between shots knowing it's only text-to-video and the answer is simple: I describe my characters, their unique traits, their clothing, their haircut, etc..anything which could help someone visually impaired have a very precise mental representation of the subject. 9/ But don't describe too much either... It would be magical if you could stuff 3000 words in the window and have exactly what you asked for, right? Well, it turns out VEO is amazing with its prompt adherence, but there is always a moment where it starts dropping animations or visual elements when your prompt stretches for a tad too long. This actually happens way before the character limit allowed by VEO is reached, so don't overdo it, it's no use and will play against the results. For info, 200-250 words seems like a sweet spot! 10/ Natural movements but... VEO is great with natural movements and this is also one of the reasons why I used it so extensively: people walking don't walk in slow-motion. That being said, don't try to be too ambitious on some of the expected movements: multiple camera movements won't work, full 360 revolutions around a subject won't work, anime-style crazy camera movements won't work, etc... what it can do is already great, but there are still some limitations...

Henry Daubrez 🌸💀

31,182 views • 1 year ago

The hallucinations of AI provide mental health break A new kind of AI is here. A real-time world model. This isn’t like anything else you’ve ever done on your computer. It creates frames one after another, which translates into a new kind of video generation engine. I’m using PixVerse’s R1, which is a category-creating technology and has just been upgraded to 720P HD video. R1 can create anything you want. I asked it to take me skiing in the Swiss Alps. Unlike other engines like Kling or Seedance it creates interactive videos. What makes this real time world model different than others like Kling or Seedance? It builds video frames one after another, so you can actually interact with the video while it plays (and it doesn’t take one to three minutes to generate like the others). More about what I learned by talking with Pixverse in this thread →>
0:46

The hallucinations of AI provide mental health break A new kind of AI is here. A real-time world model. This isn’t like anything else you’ve ever done on your computer. It creates frames one after another, which translates into a new kind of video generation engine. I’m using PixVerse’s R1, which is a category-creating technology and has just been upgraded to 720P HD video. R1 can create anything you want. I asked it to take me skiing in the Swiss Alps. Unlike other engines like Kling or Seedance it creates interactive videos. What makes this real time world model different than others like Kling or Seedance? It builds video frames one after another, so you can actually interact with the video while it plays (and it doesn’t take one to three minutes to generate like the others). More about what I learned by talking with Pixverse in this thread →>

Robert Scoble

12,937 views • 6 months ago