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 built software to cut docker image sizes by more than 80%. It uses dynamic analysis and and has a LOT more room for improvement. The demo here shows an ETL workflow docker image, cut down from 1.28GB to 193MB (-85%)! If large docker image sizes are becoming a... show more

omkaar

4,371 subscribers

353,176 views • 1 year ago •via X (Twitter)

Education Science & Technology News & Politics

Anya Rossi• Live Now

Private livecam show

10 Comments

omkaar1 year ago

Read more:

harambe_musk🍌1 year ago

This tool could come in handy for exploring layers of docker that can be shrinked:

omkaar1 year ago

dive is awesome!

vivekb.eth | w.ai1 year ago

this is a common issue we face when running production containers on GPUs, we don’t even have a model in the image just a lot of dependencies. cool stuff would love to check it out

omkaar1 year ago

thats the biggest problem with ML dependencies, they are extensible and bloated + dependencies creates a hoard of unused package. static analysis can only get you so far. would love to chat more if this is a pressing problem. Here is a quick explainer -

Adam M. Smith (史亚当)1 year ago

The idea of watching for which parts of the FS are actually touched during execution reminds me of @playablequotes /

Vasco Lameiras1 year ago

Interesting, thanks for sharing. I usually settle with distroless

omkaar1 year ago

distroless is the next best thing, my tool is more aggressive than distroless

Bobai Kato1 year ago

Simply reducing the size of the Docker image and confirming that it “still works fine” is not sufficient to guarantee that the image is free of defects. Could you provide more details on the additional testing methods you have used to validate its functionality and stability?

omkaar1 year ago

good question, the ETL tool minimal image did precisely the same transformations as the larger image. For images with broader scope, tests on the application are needed to validate whether functionality is working. luckily, most orgs have tests like this. did you have any other ideas?

Related Videos

If it doesn't have tools access, it's not an agent! In Episode 3 of AI Guide to the Galaxy, Jim Clark shows how Docker's MCP Toolkit and Gateway help you build agents that are actually ready for production. Watch here:

If it doesn't have tools access, it's not an agent! In Episode 3 of AI Guide to the Galaxy, Jim Clark shows how Docker's MCP Toolkit and Gateway help you build agents that are actually ready for production. Watch here:

Docker

50,602 views • 10 months ago

Everyone can build a chatbot. Not everyone can ship a secure, production-grade agent. In Episode 3 of AI Guide to the Galaxy, we break down how to enable your AI agents for the real-world deployments with Docker’s MCP Toolkit and Gateway. Watch here:

Everyone can build a chatbot. Not everyone can ship a secure, production-grade agent. In Episode 3 of AI Guide to the Galaxy, we break down how to enable your AI agents for the real-world deployments with Docker’s MCP Toolkit and Gateway. Watch here:

Docker

48,500 views • 10 months ago

Quick Image Upscale Tutorial 🧬 If you are doing images in Midjourney and want to really push more pixels out of them, this is for you. Here I start with an 1456 x 816px image End up with a 13104 x 7344px image This is great if you need more pixels 🧪

Quick Image Upscale Tutorial 🧬 If you are doing images in Midjourney and want to really push more pixels out of them, this is for you. Here I start with an 1456 x 816px image End up with a 13104 x 7344px image This is great if you need more pixels 🧪

Linus ✦ Ekenstam

159,644 views • 3 years ago

P-Image-Upscale is the fastest and cheapest image upscaler in the world: supporting outputs up to 128 MP under 1 seconds. A couple of weeks ago we released p-image-upscale, and it now works better than ever. It’s the fastest image upscaler in the world, supporting outputs up to 128 MP while keeping pricing simple and predictable: - $0.005/image for 1–4 MP - $0.01/image for 4–8 MP - $0.02/image for 8–16 MP - $0.04/image for 16–32 MP - $0.06/image for 32–64 MP - $0.12/image for 64–128 MP That means you can go from low-res input to production-ready output with extreme speed, while preserving detail and keeping costs easy to understand. Available on: - Pruna AI | - each::labs | - inference.sh | - Replicate | - Runware | - Segmind | - WaveSpeedAI | - Wiro | If you’re building workflows where image quality, price, latency, and scale all matter, p-image-upscale is built for you.

P-Image-Upscale is the fastest and cheapest image upscaler in the world: supporting outputs up to 128 MP under 1 seconds. A couple of weeks ago we released p-image-upscale, and it now works better than ever. It’s the fastest image upscaler in the world, supporting outputs up to 128 MP while keeping pricing simple and predictable: - $0.005/image for 1–4 MP - $0.01/image for 4–8 MP - $0.02/image for 8–16 MP - $0.04/image for 16–32 MP - $0.06/image for 32–64 MP - $0.12/image for 64–128 MP That means you can go from low-res input to production-ready output with extreme speed, while preserving detail and keeping costs easy to understand. Available on: - Pruna AI | - each::labs | - inference.sh | - Replicate | - Runware | - Segmind | - WaveSpeedAI | - Wiro | If you’re building workflows where image quality, price, latency, and scale all matter, p-image-upscale is built for you.

Pruna AI

23,723 views • 1 month ago

🔥 PROMPT SHARE 🔥 I've gotten many DMs asking how I prompt for dynamic movement using Luma AI Ray3. The trick is to prompt like you are writing a vivid scene in a screenplay. Sometimes, even explaining both visuals and sound can really bring an image to life. Happy prompting. Sometimes, simple keyword prompting works. Depends on the image and level of dynamic motion you want.

🔥 PROMPT SHARE 🔥 I've gotten many DMs asking how I prompt for dynamic movement using Luma AI Ray3. The trick is to prompt like you are writing a vivid scene in a screenplay. Sometimes, even explaining both visuals and sound can really bring an image to life. Happy prompting. Sometimes, simple keyword prompting works. Depends on the image and level of dynamic motion you want.

Dave Clark

12,173 views • 9 months ago

It's June 16. My birthday. So I built a Mini version of Kubernetes. You give it a container image → it schedules it, pulls it, runs it, and tracks its lifecycle. What it does: - Submit jobs via a REST API - Move jobs through SUBMITTED → RUNNABLE → RUNNING → SUCCEEDED / FAILED - Pull container images and run real Docker containers - Persist state in Postgres - Use BullMQ + Valkey for scheduling and execution Building this from scratch gave me a much deeper appreciation for how orchestration systems work behind the scenes. There's still plenty to improve, but seeing the control loop model in action was incredibly satisfying. Stack: Node.js · TypeScript · Express · BullMQ · Valkey · Postgres · Docker · Drizzle Not a bad way to spend a birthday. 🎂

It's June 16. My birthday. So I built a Mini version of Kubernetes. You give it a container image → it schedules it, pulls it, runs it, and tracks its lifecycle. What it does: - Submit jobs via a REST API - Move jobs through SUBMITTED → RUNNABLE → RUNNING → SUCCEEDED / FAILED - Pull container images and run real Docker containers - Persist state in Postgres - Use BullMQ + Valkey for scheduling and execution Building this from scratch gave me a much deeper appreciation for how orchestration systems work behind the scenes. There's still plenty to improve, but seeing the control loop model in action was incredibly satisfying. Stack: Node.js · TypeScript · Express · BullMQ · Valkey · Postgres · Docker · Drizzle Not a bad way to spend a birthday. 🎂

Pankaj Kumar

12,458 views • 1 month ago

How can we help *any* image-input policy generalize better? 👉 Meet PEEK 🤖 — a framework that uses VLMs to decide *where* to look and *what* to do, so downstream policies — from ACT, 3D-DA, or even π₀ — generalize more effectively! 🧵

How can we help any image-input policy generalize better? 👉 Meet PEEK 🤖 — a framework that uses VLMs to decide where to look and what to do, so downstream policies — from ACT, 3D-DA, or even π₀ — generalize more effectively! 🧵

Jesse Zhang @ RSS 2026 ✈️

13,754 views • 9 months ago

Cognition’s Scott Wu: “If anything, there’s going to be a lot more software engineers. Everyone talks about Jevons paradox and how it relates to AI; I think there’s nowhere that it’s more true than software, because we really never seem to run out of demand for more code and more software.”

Cognition’s Scott Wu: “If anything, there’s going to be a lot more software engineers. Everyone talks about Jevons paradox and how it relates to AI; I think there’s nowhere that it’s more true than software, because we really never seem to run out of demand for more code and more software.”

Stripe

15,177 views • 10 months ago

Just tested Kling AI 2.5 Turbo with Start–End Frame, and the improvement is real, higher fidelity, cleaner consistency from start to finish, and motion that feels more controlled and cinematic. Definitely a solid step forward for AI video. For filmmakers, this update pushes the workflow closer to real production. Base image Freepik (now Magnific)

Just tested Kling AI 2.5 Turbo with Start–End Frame, and the improvement is real, higher fidelity, cleaner consistency from start to finish, and motion that feels more controlled and cinematic. Definitely a solid step forward for AI video. For filmmakers, this update pushes the workflow closer to real production. Base image Freepik (now Magnific)

Dinda Prasetyo

26,526 views • 8 months ago

We’re excited to announce the release and open-source of HunyuanImage 3.0 — the largest and most powerful open-source text-to-image model to date, with over 80 billion total parameters, of which 13 billion are activated per token during inference.The effect is completely comparable to the industry’s flagship closed-source model.🚀🚀🚀 HunyuanImage 3.0 originates from our internally developed native multimodal large language model, with fine-tuning and post-training focused on text-to-image generation. This unique foundation gives the model a powerful set of capabilities: ✅Reason with world knowledge ✅Understand complex, thousand-word prompts ✅Generate precise text within images Different from traditional DiT architecture image generation models, HunyuanImage 3.0’s MoE architecture uses a Transfusion-based approach to deeply couple Diffusion and LLM training for a single, powerful system. Built on Hunyuan-A13B, HunyuanImage 3.0 was trained on a massive dataset: 5 billion image-text pairs, video frames, interleaved image-text data, and 6 trillion tokens of text corpora. This hybrid training across multimodal generation, understanding, and LLM capabilities allows the model to seamlessly integrate multiple tasks. Whether you're an illustrator, designer, or creator, this is built to slash your workflow from hours to minutes. HunyuanImage 3.0 can generate intricate text, detailed comics, expressive emojis, and lively, engaging illustrations for educational content. The current release focuses solely on text-to-image generation and future updates will include image-to-image, image editing, multi-turn interaction, and more. 👉🏻Try it now: 🔗GitHub: 🤗Hugging Face:

We’re excited to announce the release and open-source of HunyuanImage 3.0 — the largest and most powerful open-source text-to-image model to date, with over 80 billion total parameters, of which 13 billion are activated per token during inference.The effect is completely comparable to the industry’s flagship closed-source model.🚀🚀🚀 HunyuanImage 3.0 originates from our internally developed native multimodal large language model, with fine-tuning and post-training focused on text-to-image generation. This unique foundation gives the model a powerful set of capabilities: ✅Reason with world knowledge ✅Understand complex, thousand-word prompts ✅Generate precise text within images Different from traditional DiT architecture image generation models, HunyuanImage 3.0’s MoE architecture uses a Transfusion-based approach to deeply couple Diffusion and LLM training for a single, powerful system. Built on Hunyuan-A13B, HunyuanImage 3.0 was trained on a massive dataset: 5 billion image-text pairs, video frames, interleaved image-text data, and 6 trillion tokens of text corpora. This hybrid training across multimodal generation, understanding, and LLM capabilities allows the model to seamlessly integrate multiple tasks. Whether you're an illustrator, designer, or creator, this is built to slash your workflow from hours to minutes. HunyuanImage 3.0 can generate intricate text, detailed comics, expressive emojis, and lively, engaging illustrations for educational content. The current release focuses solely on text-to-image generation and future updates will include image-to-image, image editing, multi-turn interaction, and more. 👉🏻Try it now: 🔗GitHub: 🤗Hugging Face:

Tencent Hy

412,658 views • 9 months ago

We are thrilled to announce "Layered Diffusion Brushes": a real-time training-free image editor powered by diffusion models. 🎨✨ This is new work from my PhD student Peyman Gholami Peyman Gholami. Explore the interactive demo and check out more videos at:

We are thrilled to announce "Layered Diffusion Brushes": a real-time training-free image editor powered by diffusion models. 🎨✨ This is new work from my PhD student Peyman Gholami Peyman Gholami. Explore the interactive demo and check out more videos at:

Robert Xiao

14,334 views • 2 years ago

Phoenix GenAI’s new generative AI model Image-to-3D is now online. Create high fidelity 3D models importable to Unreal Engine 5 or Unity simply by sending an original image into PhoenixLLM. Better yet, generate the source image using Phoenix GenAI’s Flux and feed it into Image-to-3D. This release marks yet another upgrade of GenAI’s arsenal of capabilities, getting it ready for multi-workflow GenAI agents, in which users will be able to combine text-to-image, image-to-prompt, text-to-video, text-to-3D, and image-to-3D into complex multi-step workflows with simple commands via PhoenixLLM. Image-to-3D is yet another addition to Phoenix’s Vertical AI Solutions for gaming, content, and metaverse. Users are able to use it as a Phoenix-native alternative to SkyNet AI Marketplace’s Tripo Integration earlier this year. #Phoenix $PHB

Phoenix GenAI’s new generative AI model Image-to-3D is now online. Create high fidelity 3D models importable to Unreal Engine 5 or Unity simply by sending an original image into PhoenixLLM. Better yet, generate the source image using Phoenix GenAI’s Flux and feed it into Image-to-3D. This release marks yet another upgrade of GenAI’s arsenal of capabilities, getting it ready for multi-workflow GenAI agents, in which users will be able to combine text-to-image, image-to-prompt, text-to-video, text-to-3D, and image-to-3D into complex multi-step workflows with simple commands via PhoenixLLM. Image-to-3D is yet another addition to Phoenix’s Vertical AI Solutions for gaming, content, and metaverse. Users are able to use it as a Phoenix-native alternative to SkyNet AI Marketplace’s Tripo Integration earlier this year. #Phoenix $PHB

Phoenix AI

30,853 views • 1 year ago

Hollywood is part of L.A.’s identity. The entertainment industry should be able to thrive in the city that built it. That’s why I cut filming fees, reduced unnecessary production costs, and reopened city buildings for filming to help keep jobs and production in Los Angeles. There’s still a lot more work to do, but we’re finally seeing an uptick in production in L.A.

Hollywood is part of L.A.’s identity. The entertainment industry should be able to thrive in the city that built it. That’s why I cut filming fees, reduced unnecessary production costs, and reopened city buildings for filming to help keep jobs and production in Los Angeles. There’s still a lot more work to do, but we’re finally seeing an uptick in production in L.A.

Karen Bass

151,136 views • 1 month ago

omni flash + gpt image 2 ads are PRINTING like crazy right now. i found a stupidly simple way to reverse engineer viral ads and pump out winners in minutes. i just made a step-by-step guide that breaks down the entire workflow. rt + reply 'workflow' and I’ll send it over (must be following so i can dm)

omni flash + gpt image 2 ads are PRINTING like crazy right now. i found a stupidly simple way to reverse engineer viral ads and pump out winners in minutes. i just made a step-by-step guide that breaks down the entire workflow. rt + reply 'workflow' and I’ll send it over (must be following so i can dm)

0x ROAS

632,863 views • 8 days ago

⚡ What Freepik has achieved is insane How the hell did it go from being just a simple stock image site to becoming the all-in-one AI suite? Honestly, this should be case study material in entrepreneurship classes someday 😂 It’s wild what they’ve pulled off by stitching together every tool on the market (Magnific, all the video generators, everything from Flux, inpainting, outpainting, etc.) into one seamless workflow: from transforming any image with Flux’s latest to animating it, upscaling it, and turning it into a full-on video. The whole thing flows smoother than sliding down a waterslide.

⚡ What Freepik has achieved is insane How the hell did it go from being just a simple stock image site to becoming the all-in-one AI suite? Honestly, this should be case study material in entrepreneurship classes someday 😂 It’s wild what they’ve pulled off by stitching together every tool on the market (Magnific, all the video generators, everything from Flux, inpainting, outpainting, etc.) into one seamless workflow: from transforming any image with Flux’s latest to animating it, upscaling it, and turning it into a full-on video. The whole thing flows smoother than sliding down a waterslide.

Javi Lopez ⛩️

61,201 views • 1 year ago

Jensen Huang says, In the future, we will be able to talk to a protein We'll ask it: "what are you? are you soluble? how do you react to heat or chemicals?" This is how AI can talk to an image today, and how we've moved from only processing data to actually understanding it

Jensen Huang says, In the future, we will be able to talk to a protein We'll ask it: "what are you? are you soluble? how do you react to heat or chemicals?" This is how AI can talk to an image today, and how we've moved from only processing data to actually understanding it

Haider.

149,649 views • 8 months ago

RT to help someone in need of free services for image based abuse! If you or someone you know is a victim of image based sexual abuse go to Foundation Ra for free take down services, legal advice, therapy, law enforcement and so much more. I created Foundation Ra because these are the resources I wish I had when I became a victim of image based abuse and received no help or support from anywhere instead everyone wanted to profit off of me and kick me while I was already down. So I made sure I created a non profit with free and effective resources for anyone going through this today.

RT to help someone in need of free services for image based abuse! If you or someone you know is a victim of image based sexual abuse go to Foundation Ra for free take down services, legal advice, therapy, law enforcement and so much more. I created Foundation Ra because these are the resources I wish I had when I became a victim of image based abuse and received no help or support from anywhere instead everyone wanted to profit off of me and kick me while I was already down. So I made sure I created a non profit with free and effective resources for anyone going through this today.

Uldouz

16,659 views • 1 year ago

As I promised yesterday, I'll briefly explain LoRA training and share a workflow I made so you can do it quickly. First, let me answer a very common question: 'Why train LoRAs when we have such advanced models?' Even though we have incredibly advanced models now (like NBP), we still can't always get them to do specific things we want. Simplest example: the spritesheet LoRA I made the other day. I generated 1000 images with Nano Banana and only 100 were what I wanted. The LoRA I trained using those 100 images gives me nearly 100% consistent results. Second point is cost and speed. With LoRA, we can cut costs by 4-5x. And while doing that, we're generating 4-5x faster. How many images do you need for a good LoRA? This depends on your LoRA's complexity. For example, when I training the spritesheet LoRA, even though I used 100 images, I didn't include buildings in the training data, so this LoRA doesn't work for buildings. So think about your LoRA's use cases and add examples for as many use cases as possible to improve quality. What are paired images and how to train LoRAs for image-editing? When training LoRAs for image editing on fal, we call each edit example paired images - one with _start suffix, one with _end suffix. For example, if you're training a background remove LoRA, the unedited original photo will be your '_start' image. The image with background removed will be the '_end' image. Simply put: images we want to edit or use as reference get _start, target images we want to achieve get '_end'. Important: save both images with the same name. Like image332_start.jpg and image332_end.jpg. This way the system knows which images pair together. What about training LoRAs for models with multiple image inputs? Same logic. We still use _start and _end suffixes, but with one difference. Since there are multiple input images, we can number them: _start, _start1, _start2. Example: start images, 1st image = Woman portrait (image35_start.jpg) 2nd image = Glasses photo (image35_start1.jpg) 3rd image = Hat photo (image35_start2.jpg) Output image = portrait of woman wearing glasses and hat (image35_end.jpg) Can we do more detailed captioning? Yes. Similarly, you can improve training quality by creating a txt file for each set with the caption inside. Example: create image35.txt and write: 'Recreate the image by putting the glasses from the second image and the hat from the third image on the woman in the first image.' What are Steps? How many should I use? What's Learning Rate? Steps determines how many times the model sees and processes your training data (your images). Each step, the model learns a bit more. But as steps increase, so does the risk of overfitting. So there's no real default. But for a simpler LoRA with 20 paired images, 1000 steps is ideal. Here's a metaphor for the Steps and Learning Rate relationship: Imagine you have a balloon. Our goal is to inflate it to the optimal size. Steps = How many times we blow into the balloon Learning rate = How hard we blow each time If we blow too softly, we need to blow many more times. If we blow too hard, we risk popping it quickly and can't reach optimal size. Of course training won't explode, but it won't work as intended because it wasn't trained optimally. Training's done, now what? Once training's complete, you'll have a safetensors file. Every model you train on fal has a LoRA inference endpoint. In that inference, add your safetensors file link to the LoRA url input, and you can use your LoRA. Thanks for the read! The workflow in the video: If I forgot anything, let me know in the replies.

As I promised yesterday, I'll briefly explain LoRA training and share a workflow I made so you can do it quickly. First, let me answer a very common question: 'Why train LoRAs when we have such advanced models?' Even though we have incredibly advanced models now (like NBP), we still can't always get them to do specific things we want. Simplest example: the spritesheet LoRA I made the other day. I generated 1000 images with Nano Banana and only 100 were what I wanted. The LoRA I trained using those 100 images gives me nearly 100% consistent results. Second point is cost and speed. With LoRA, we can cut costs by 4-5x. And while doing that, we're generating 4-5x faster. How many images do you need for a good LoRA? This depends on your LoRA's complexity. For example, when I training the spritesheet LoRA, even though I used 100 images, I didn't include buildings in the training data, so this LoRA doesn't work for buildings. So think about your LoRA's use cases and add examples for as many use cases as possible to improve quality. What are paired images and how to train LoRAs for image-editing? When training LoRAs for image editing on fal, we call each edit example paired images - one with _start suffix, one with _end suffix. For example, if you're training a background remove LoRA, the unedited original photo will be your '_start' image. The image with background removed will be the '_end' image. Simply put: images we want to edit or use as reference get _start, target images we want to achieve get '_end'. Important: save both images with the same name. Like image332_start.jpg and image332_end.jpg. This way the system knows which images pair together. What about training LoRAs for models with multiple image inputs? Same logic. We still use _start and _end suffixes, but with one difference. Since there are multiple input images, we can number them: _start, _start1, _start2. Example: start images, 1st image = Woman portrait (image35_start.jpg) 2nd image = Glasses photo (image35_start1.jpg) 3rd image = Hat photo (image35_start2.jpg) Output image = portrait of woman wearing glasses and hat (image35_end.jpg) Can we do more detailed captioning? Yes. Similarly, you can improve training quality by creating a txt file for each set with the caption inside. Example: create image35.txt and write: 'Recreate the image by putting the glasses from the second image and the hat from the third image on the woman in the first image.' What are Steps? How many should I use? What's Learning Rate? Steps determines how many times the model sees and processes your training data (your images). Each step, the model learns a bit more. But as steps increase, so does the risk of overfitting. So there's no real default. But for a simpler LoRA with 20 paired images, 1000 steps is ideal. Here's a metaphor for the Steps and Learning Rate relationship: Imagine you have a balloon. Our goal is to inflate it to the optimal size. Steps = How many times we blow into the balloon Learning rate = How hard we blow each time If we blow too softly, we need to blow many more times. If we blow too hard, we risk popping it quickly and can't reach optimal size. Of course training won't explode, but it won't work as intended because it wasn't trained optimally. Training's done, now what? Once training's complete, you'll have a safetensors file. Every model you train on fal has a LoRA inference endpoint. In that inference, add your safetensors file link to the LoRA url input, and you can use your LoRA. Thanks for the read! The workflow in the video: If I forgot anything, let me know in the replies.

ilker

15,133 views • 5 months ago

A node graph for the entire video pipeline, from brainstorm to final cut, built for scalable content production. Build a workflow once, run it forever, fork it into any project, and share it across your team with ctrl c + ctrl v. Every production agency we work with asked for Canvas, so here we go.

A node graph for the entire video pipeline, from brainstorm to final cut, built for scalable content production. Build a workflow once, run it forever, fork it into any project, and share it across your team with ctrl c + ctrl v. Every production agency we work with asked for Canvas, so here we go.

Higgsfield AI 🧩

149,849 views • 2 months ago