Loading video...

Video Failed to Load

Go Home

Experimenting with a multi-level cache for high-frequency effects based on my previous work for SPWI surface GI (left). Specular effects like caustics scale in 2D on surfaces. Also researching solutions to angular aliasing and flickering of depth-upscaled probes (right). No TAA.

CodyBennett's profile picture

Cody Bennett

4,118 subscribers

40,090 views • 6 days 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

Six years later this is still one of my favorite outline effects especially for hard surface models like CAD & architecture. And now it works with three.js' WebGPURenderer thanks to a contribution from "cmhhelgeson" on Github!
0:16

Six years later this is still one of my favorite outline effects especially for hard surface models like CAD & architecture. And now it works with three.js' WebGPURenderer thanks to a contribution from "cmhhelgeson" on Github!

Garrett Johnson

23,282 views • 3 months ago

WeatherEdit: Controllable Weather Editing with 4D Gaussian Field Contributions: 1. Based on our analysis of weather editing characteristics, we introduce WeatherEdit, a comprehensive and efficient framework for realistic and controllable weather generation. Compared with existing methods that focus on either background editing or static weather effects, a progressive 2D-to-4D transformation process in WeatherEdit enhances adaptability across a wider range of scenarios. 2. We introduce an all-in-one adapter to enable a diffusion model for multi-weather (snowy, rainy, and fog) synthesis, along with a Temporal-View attention to ensure consistent editing across multi-frame and multi-view. 3. We design a 4D Gaussian field for weather particle modeling, enabling plausible simulation of raindrops, snowflakes, and fog with controllable severity. 4. We demonstrate WeatherEdit’s effectiveness in generating realistic, consistent, and controllable weather effects in 3D driving scenes, showcasing its applicability to real-world scenarios.
0:13

WeatherEdit: Controllable Weather Editing with 4D Gaussian Field Contributions: 1. Based on our analysis of weather editing characteristics, we introduce WeatherEdit, a comprehensive and efficient framework for realistic and controllable weather generation. Compared with existing methods that focus on either background editing or static weather effects, a progressive 2D-to-4D transformation process in WeatherEdit enhances adaptability across a wider range of scenarios. 2. We introduce an all-in-one adapter to enable a diffusion model for multi-weather (snowy, rainy, and fog) synthesis, along with a Temporal-View attention to ensure consistent editing across multi-frame and multi-view. 3. We design a 4D Gaussian field for weather particle modeling, enabling plausible simulation of raindrops, snowflakes, and fog with controllable severity. 4. We demonstrate WeatherEdit’s effectiveness in generating realistic, consistent, and controllable weather effects in 3D driving scenes, showcasing its applicability to real-world scenarios.

MrNeRF

10,691 views • 1 year ago

Visual Preset #01 Ink-Brush Cinematic 3D: A high-end cinematic 3D style where expressive ink-brush effects become the primary visual language for fast-paced anime action. Lately I've noticed that I've been experimenting with different visual presets across my videos and I'd like to explore that direction even further. Going forward, I'll be sharing some of these style experiments. The video below was generated using only a character sheet, a single-line scene description and the visual preset shown below. Created with Seedance 2.0 on Try ArtCraft Seedance 2.0 Prompt: A mesmerizing display of @[character]'s masterful swordsmanship. High-end cinematic 3D realism fused with expressive ink-brush action. High-sakuga anime choreography, sweeping sumi-e brush strokes, flowing ink splashes, dynamic calligraphic energy and graphic black ink trails define every movement. Extreme perspective, dramatic foreshortening, cinematic tracking shots, volumetric lighting, heavy atmospheric haze and explosive ink bursts replace conventional visual effects, while realistic materials and feature-film rendering preserve depth, weight and scale.
0:15

Visual Preset #01 Ink-Brush Cinematic 3D: A high-end cinematic 3D style where expressive ink-brush effects become the primary visual language for fast-paced anime action. Lately I've noticed that I've been experimenting with different visual presets across my videos and I'd like to explore that direction even further. Going forward, I'll be sharing some of these style experiments. The video below was generated using only a character sheet, a single-line scene description and the visual preset shown below. Created with Seedance 2.0 on Try ArtCraft Seedance 2.0 Prompt: A mesmerizing display of @[character]'s masterful swordsmanship. High-end cinematic 3D realism fused with expressive ink-brush action. High-sakuga anime choreography, sweeping sumi-e brush strokes, flowing ink splashes, dynamic calligraphic energy and graphic black ink trails define every movement. Extreme perspective, dramatic foreshortening, cinematic tracking shots, volumetric lighting, heavy atmospheric haze and explosive ink bursts replace conventional visual effects, while realistic materials and feature-film rendering preserve depth, weight and scale.

Kōda

43,274 views • 3 months ago

The Complete Godot VFX Handbook is our next project! We’ll cover 2D, 3D, UI, and Particle System VFX, with hands-on projects and techniques for creating stunning effects in Godot. However, we need 1k wishlists to start developing it!Check it out and add it to your wishlist if you’d like to see this book become a reality! 🔗 #indiedev #godot
0:15

The Complete Godot VFX Handbook is our next project! We’ll cover 2D, 3D, UI, and Particle System VFX, with hands-on projects and techniques for creating stunning effects in Godot. However, we need 1k wishlists to start developing it!Check it out and add it to your wishlist if you’d like to see this book become a reality! 🔗 #indiedev #godot

Shaders Bible Series

75,116 views • 1 month ago

Most people have no idea that these photos exist from the surface of Venus. Climate change is real. It’s happening on Venus and it’s happening on Earth. Did you know that Venus may have been habitable at one point in time? It’s known as Earth’s sister planet. Below are some of the only images ever taken from the surface of Venus. One day Earth may look similar to this. There is no argument that climate change is real. It happened on Venus and it’s constantly happening on Earth. Human interaction most definitely has some effects on the planet’s climate change, but so too do non-human factors — even to a greater degree. My hope is that, not only can humans work to slow down the process, but also one day come up with solutions to reverse it.
0:19

Most people have no idea that these photos exist from the surface of Venus. Climate change is real. It’s happening on Venus and it’s happening on Earth. Did you know that Venus may have been habitable at one point in time? It’s known as Earth’s sister planet. Below are some of the only images ever taken from the surface of Venus. One day Earth may look similar to this. There is no argument that climate change is real. It happened on Venus and it’s constantly happening on Earth. Human interaction most definitely has some effects on the planet’s climate change, but so too do non-human factors — even to a greater degree. My hope is that, not only can humans work to slow down the process, but also one day come up with solutions to reverse it.

Ed Krassenstein

1,591,184 views • 2 years ago

One small step for Hivemind, one giant leap for mission autonomy. Hivemind is going to space! Shield AI and Sedaro are partnering to send Hivemind to space, bringing the same resilient, edge-based autonomy that is redefining warfare to satellites delivering critical infrastructure and national defense capabilities. Our strategic partnership establishes Hivemind Pilot as Sedaro’s preferred autonomy software for on-orbit demonstrations, while also enabling the use of the Sedaro Platform for developing, testing, and demonstrating Hivemind in space-relevant scenarios. “Combining Hivemind Pilot with Sedaro’s high-fidelity models and simulation environment will unlock multi-agent cognitive teaming for space applications and new mission capabilities for our customers.” - Christian Gutierrez, Vice President of Hivemind Solutions. Hivemind is multi-domain mission autonomy. We've done autonomy for air, ground, and surface applications this year, and we couldn't be more excited to go to space. If your systems need autonomy, we'd love to work with you and help give your systems the best autonomy in the business. Read the full press release: Autonomy for the world. The greatest victory requires no war.
0:15

One small step for Hivemind, one giant leap for mission autonomy. Hivemind is going to space! Shield AI and Sedaro are partnering to send Hivemind to space, bringing the same resilient, edge-based autonomy that is redefining warfare to satellites delivering critical infrastructure and national defense capabilities. Our strategic partnership establishes Hivemind Pilot as Sedaro’s preferred autonomy software for on-orbit demonstrations, while also enabling the use of the Sedaro Platform for developing, testing, and demonstrating Hivemind in space-relevant scenarios. “Combining Hivemind Pilot with Sedaro’s high-fidelity models and simulation environment will unlock multi-agent cognitive teaming for space applications and new mission capabilities for our customers.” - Christian Gutierrez, Vice President of Hivemind Solutions. Hivemind is multi-domain mission autonomy. We've done autonomy for air, ground, and surface applications this year, and we couldn't be more excited to go to space. If your systems need autonomy, we'd love to work with you and help give your systems the best autonomy in the business. Read the full press release: Autonomy for the world. The greatest victory requires no war.

Shield AI

33,072 views • 10 months ago

Building gigawatt scale datacenters in the oceans may be even easier than I had previously thought. Normally for OTEC cold deep sea water and warm surface waters are needed. Because that gigawatt is generating a lot of heat as an end product of compute, along with some additional solar thermal collection, we won’t need warm surface water at all, allowing these datacenters to go almost anywhere with enough depth. Additionally, that gigawatt of heat may be hotter than surface seawater normally used for OTEC, allowing for much shallower piping (200m instead of 1000m) while also being more efficient. The availability of unlimited cold water for the condenser is the key factor we can’t replicate on land. This brings the potential energy generation for ocean based compute up from 10 terawatts to potentially thousands of terawatts, as the locations are not limited to areas with the warmest surface waters. The excess energy needed for hundreds or thousands of people to also live on these structures is minimal compared to that needed for compute.
0:10

Building gigawatt scale datacenters in the oceans may be even easier than I had previously thought. Normally for OTEC cold deep sea water and warm surface waters are needed. Because that gigawatt is generating a lot of heat as an end product of compute, along with some additional solar thermal collection, we won’t need warm surface water at all, allowing these datacenters to go almost anywhere with enough depth. Additionally, that gigawatt of heat may be hotter than surface seawater normally used for OTEC, allowing for much shallower piping (200m instead of 1000m) while also being more efficient. The availability of unlimited cold water for the condenser is the key factor we can’t replicate on land. This brings the potential energy generation for ocean based compute up from 10 terawatts to potentially thousands of terawatts, as the locations are not limited to areas with the warmest surface waters. The excess energy needed for hundreds or thousands of people to also live on these structures is minimal compared to that needed for compute.

Ben Silone

79,214 views • 3 months ago

I'll always root for a team that open-sources its best work, and Robbyant just did it properly. Robbyant, Ant Group's embodied-AI company, released LingBot-Vision, a vision foundation model for robots, and the part I love is the data. They trained it on 161M images, filtered down from 2B raw ones and mostly pulled straight from the open web, with no human labels, no edge detectors, no depth sensors anywhere in the loop. It learns the exact edges of objects from raw pixels. That's roughly a tenth of the data DINOv3 saw, and under a third of the training. And it shows in the results. On depth, working out how far away things are, the 1B model edges out a 7B on NYU-Depth. It also powers LingBot-Depth 2.0, which reads the surfaces cameras usually choke on, glass and mirrors, and halves indoor depth error. LingBot-Vision is fully open. Weights from the 1.1B flagship down to a tiny 21M version, code, and the paper. This is the timeline I want more of. Robbyant
0:10

I'll always root for a team that open-sources its best work, and Robbyant just did it properly. Robbyant, Ant Group's embodied-AI company, released LingBot-Vision, a vision foundation model for robots, and the part I love is the data. They trained it on 161M images, filtered down from 2B raw ones and mostly pulled straight from the open web, with no human labels, no edge detectors, no depth sensors anywhere in the loop. It learns the exact edges of objects from raw pixels. That's roughly a tenth of the data DINOv3 saw, and under a third of the training. And it shows in the results. On depth, working out how far away things are, the 1B model edges out a 7B on NYU-Depth. It also powers LingBot-Depth 2.0, which reads the surfaces cameras usually choke on, glass and mirrors, and halves indoor depth error. LingBot-Vision is fully open. Weights from the 1.1B flagship down to a tiny 21M version, code, and the paper. This is the timeline I want more of. Robbyant

Chubby♨️

48,249 views • 2 months ago

for this video wanted to see how far i could get with claude code no premiere or after effects allowed this would've taken 15 min using those traditional tools instead i prompted the entire edit and described all effects through natural language it wound up being an incredibly terrible user experience that took me 4 hours and burned through a ton of tokens 😂 shot in ibiza on a $50 camera i got on amazon by my guy Wut
0:17

for this video wanted to see how far i could get with claude code no premiere or after effects allowed this would've taken 15 min using those traditional tools instead i prompted the entire edit and described all effects through natural language it wound up being an incredibly terrible user experience that took me 4 hours and burned through a ton of tokens 😂 shot in ibiza on a $50 camera i got on amazon by my guy Wut

figge

18,392 views • 4 months ago

Some updates on the multiview vistadream pipeline with Rerun! Rerun came in extremely useful here, as being able to visualize depths at each stage of the pipeline allowed me to debug some nasty bugs. Since the last time, I was only working with a single image input. I've added in VGGT as my multiview pose + depth estimator. It works REALLY well for getting camera poses, but the depths are not that great. To try and fix that, I estimated depth maps from MoGeV2 for each of the views, and scale+shift aligned them so that they would match up to the confident sections of VGGT's depth predictions. You can see in the video just how much sharper the visualized 2d depth maps are! The biggest issue continues to be the multiview consistency 🫠 That's up next, along with actually training the Gaussian splat. Lots of work went into actually understanding inputs+outputs for VGGT. I had some funky bugs where the confidence values would all collapse to true I'm also really excited for this pipeline to use Difix3D+ Nvidia instead of Flux Inpainting, it seems like a better suited for a multiview pipeline.
0:14

Some updates on the multiview vistadream pipeline with Rerun! Rerun came in extremely useful here, as being able to visualize depths at each stage of the pipeline allowed me to debug some nasty bugs. Since the last time, I was only working with a single image input. I've added in VGGT as my multiview pose + depth estimator. It works REALLY well for getting camera poses, but the depths are not that great. To try and fix that, I estimated depth maps from MoGeV2 for each of the views, and scale+shift aligned them so that they would match up to the confident sections of VGGT's depth predictions. You can see in the video just how much sharper the visualized 2d depth maps are! The biggest issue continues to be the multiview consistency 🫠 That's up next, along with actually training the Gaussian splat. Lots of work went into actually understanding inputs+outputs for VGGT. I had some funky bugs where the confidence values would all collapse to true I'm also really excited for this pipeline to use Difix3D+ Nvidia instead of Flux Inpainting, it seems like a better suited for a multiview pipeline.

Pablo Vela

29,904 views • 1 year ago

Physiology of PEEP Alveolar Recruitment and Stabilization Recruitment: PEEP opens collapsed alveoli, increasing the surface area for gas exchange. Stabilization: By maintaining alveoli open, PEEP prevents the cyclic opening and closing of alveoli, reducing shear stress and the risk of ventilator-induced lung injury (VILI). Improvement in Oxygenation V/Q Matching: PEEP improves ventilation-perfusion matching by redirecting blood flow to well-ventilated alveoli, reducing intrapulmonary shunting. Redistribution of Edema: In conditions like ARDS, PEEP can redistribute alveolar edema, improving compliance and gas exchange. Effects on Compliance Static Compliance: PEEP can increase static compliance by recruiting alveoli, but excessive PEEP may overdistended alveoli, decreasing compliance. Dynamic Compliance: PEEP may also affect dynamic compliance by altering airway resistance. Hemodynamic Implications Venous Return: Increased intrathoracic pressure reduces venous return, potentially decreasing cardiac output. Afterload: PEEP may increase left ventricular afterload by increasing transpulmonary pressure. Right Ventricular Function: High PEEP may cause right ventricular dilation and dysfunction, especially in the presence of pulmonary hypertension. Effects on Intracranial Pressure (ICP) PEEP may increase ICP by reducing venous outflow from the brain, a critical consideration in neurocritical care. Clinical Application and Monitoring ARDS: PEEP/FiO2 Tables: Utilizing evidence-based tables to titrate PEEP based on FiO2 requirements. Recruitment Maneuvers: Often used in conjunction with PEEP to assess recruitability. Monitoring with Esophageal Manometry: To assess transpulmonary pressure and individualize PEEP settings. Obstructive Lung Disease: Careful application of PEEP to prevent air trapping and intrinsic PEEP (auto-PEEP). Heart Failure and Fluid Status: Echocardiographic Monitoring: To assess the impact of PEEP on cardiac function and filling pressures. Pulmonary Artery Catheterization: May be used to monitor the effects of PEEP on pulmonary artery pressures and cardiac output. Protective Lung Ventilation in Surgery: Utilizing PEEP to prevent atelectasis and postoperative pulmonary complications. Weaning Process Gradual Reduction: Monitoring respiratory mechanics, work of breathing, and gas exchange. Spontaneous Breathing Trials (SBT): Assessing the ability to tolerate lower PEEP levels. Conclusion PEEP is a complex and vital component of mechanical ventilation, with multifaceted effects on respiratory mechanics, gas exchange, hemodynamics, and even neurodynamics. Its application requires a nuanced understanding of underlying pathophysiology, continuous monitoring with advanced tools, and individualized titration to optimize patient outcomes. The integration of PEEP into a comprehensive respiratory care strategy exemplifies the complexity and precision required in critical care medicine.
0:15

Physiology of PEEP Alveolar Recruitment and Stabilization Recruitment: PEEP opens collapsed alveoli, increasing the surface area for gas exchange. Stabilization: By maintaining alveoli open, PEEP prevents the cyclic opening and closing of alveoli, reducing shear stress and the risk of ventilator-induced lung injury (VILI). Improvement in Oxygenation V/Q Matching: PEEP improves ventilation-perfusion matching by redirecting blood flow to well-ventilated alveoli, reducing intrapulmonary shunting. Redistribution of Edema: In conditions like ARDS, PEEP can redistribute alveolar edema, improving compliance and gas exchange. Effects on Compliance Static Compliance: PEEP can increase static compliance by recruiting alveoli, but excessive PEEP may overdistended alveoli, decreasing compliance. Dynamic Compliance: PEEP may also affect dynamic compliance by altering airway resistance. Hemodynamic Implications Venous Return: Increased intrathoracic pressure reduces venous return, potentially decreasing cardiac output. Afterload: PEEP may increase left ventricular afterload by increasing transpulmonary pressure. Right Ventricular Function: High PEEP may cause right ventricular dilation and dysfunction, especially in the presence of pulmonary hypertension. Effects on Intracranial Pressure (ICP) PEEP may increase ICP by reducing venous outflow from the brain, a critical consideration in neurocritical care. Clinical Application and Monitoring ARDS: PEEP/FiO2 Tables: Utilizing evidence-based tables to titrate PEEP based on FiO2 requirements. Recruitment Maneuvers: Often used in conjunction with PEEP to assess recruitability. Monitoring with Esophageal Manometry: To assess transpulmonary pressure and individualize PEEP settings. Obstructive Lung Disease: Careful application of PEEP to prevent air trapping and intrinsic PEEP (auto-PEEP). Heart Failure and Fluid Status: Echocardiographic Monitoring: To assess the impact of PEEP on cardiac function and filling pressures. Pulmonary Artery Catheterization: May be used to monitor the effects of PEEP on pulmonary artery pressures and cardiac output. Protective Lung Ventilation in Surgery: Utilizing PEEP to prevent atelectasis and postoperative pulmonary complications. Weaning Process Gradual Reduction: Monitoring respiratory mechanics, work of breathing, and gas exchange. Spontaneous Breathing Trials (SBT): Assessing the ability to tolerate lower PEEP levels. Conclusion PEEP is a complex and vital component of mechanical ventilation, with multifaceted effects on respiratory mechanics, gas exchange, hemodynamics, and even neurodynamics. Its application requires a nuanced understanding of underlying pathophysiology, continuous monitoring with advanced tools, and individualized titration to optimize patient outcomes. The integration of PEEP into a comprehensive respiratory care strategy exemplifies the complexity and precision required in critical care medicine.

𝗥𝗲𝘀𝘂𝘀𝗠𝗲𝗱

84,714 views • 3 years ago

Throughout my journey in developing multimodal models, I’ve always wanted a framework that lets me plug & play modality encoders/decoders on top of an auto-regressive LLM. I want to prototype fast, try new architectures, and have my demo files scale effortlessly — with full support for parallelism and optimization. Not just to hack⚙️, but also to scale🚀. So finally we built it for ourselves. LMMs-Engine: a lean, efficient framework built to train unified multimodal model at scale. From Qwen LLM, VLM, LLaVA-OV, and WanVideo, to unified models like Qwen-Omni and BAGEL — plus Linear-Attn GDN and research prototypes like RAE and SiT - all under one modular system that seamlessly integrates diverse datasets and optimization strategies. Powered by FSDP2 multi-dim parallelism, Ulysses sequence parallel, Flash-Attention, Liger Kernels, and Native Sparse Attention (also with bonus support for the Muon optimizer for all models).
0:19

Throughout my journey in developing multimodal models, I’ve always wanted a framework that lets me plug & play modality encoders/decoders on top of an auto-regressive LLM. I want to prototype fast, try new architectures, and have my demo files scale effortlessly — with full support for parallelism and optimization. Not just to hack⚙️, but also to scale🚀. So finally we built it for ourselves. LMMs-Engine: a lean, efficient framework built to train unified multimodal model at scale. From Qwen LLM, VLM, LLaVA-OV, and WanVideo, to unified models like Qwen-Omni and BAGEL — plus Linear-Attn GDN and research prototypes like RAE and SiT - all under one modular system that seamlessly integrates diverse datasets and optimization strategies. Powered by FSDP2 multi-dim parallelism, Ulysses sequence parallel, Flash-Attention, Liger Kernels, and Native Sparse Attention (also with bonus support for the Muon optimizer for all models).

Brian Li

54,840 views • 11 months ago

Green Crow Dev and I are making an addon for #GodotEngine that lets you control every aspect of rendering and post-processing in one place. For example, you'll be able to adjust rendering settings that are usually hidden in the Project Settings, like shadow resolution and anti-aliasing, control the sun, tweak WorldEnvironment settings, add and customize lens flare and other effects, as well as handle post-processing in general (color correction, chromatic aberration, film grain). You'll also be able to change the mood of a scene entirely with just a couple of clicks by selecting a preset (e.g. cinematic, horror, noir, etc.). Would this be something you'd be interested in?
0:14

Green Crow Dev and I are making an addon for #GodotEngine that lets you control every aspect of rendering and post-processing in one place. For example, you'll be able to adjust rendering settings that are usually hidden in the Project Settings, like shadow resolution and anti-aliasing, control the sun, tweak WorldEnvironment settings, add and customize lens flare and other effects, as well as handle post-processing in general (color correction, chromatic aberration, film grain). You'll also be able to change the mood of a scene entirely with just a couple of clicks by selecting a preset (e.g. cinematic, horror, noir, etc.). Would this be something you'd be interested in?

FR3NKD

22,369 views • 11 months ago

We’re excited to announce our integration with SKALE, a high-performance, zero-gas blockchain purpose-built for speed, scale, and security. This partnership strengthens our infrastructure as we continue building transparent, trust-based systems for decentralized science. We’re excited about what this unlocks for researchers, contributors, and the future of data integrity in DeSci. 👀 Look out for more on how we’re using SKALE in the AxonDAO ecosystem.
0:15

We’re excited to announce our integration with SKALE, a high-performance, zero-gas blockchain purpose-built for speed, scale, and security. This partnership strengthens our infrastructure as we continue building transparent, trust-based systems for decentralized science. We’re excited about what this unlocks for researchers, contributors, and the future of data integrity in DeSci. 👀 Look out for more on how we’re using SKALE in the AxonDAO ecosystem.

AxonDAO

33,926 views • 1 year ago

2 years of experience & 20+ clients… I FINALLY finished it. The most in-depth mastermind on SMM for Web3 projects. 5 modules packed with resources, research, systems, and templates. Companies pay $3500 / mo to get this knowledge… And I’m gonna teach it for FREE to a select group that’s ready to level up. Hosting this right after X Coach AI drops next week. Who wants in?
0:19

2 years of experience & 20+ clients… I FINALLY finished it. The most in-depth mastermind on SMM for Web3 projects. 5 modules packed with resources, research, systems, and templates. Companies pay $3500 / mo to get this knowledge… And I’m gonna teach it for FREE to a select group that’s ready to level up. Hosting this right after X Coach AI drops next week. Who wants in?

Ares🌱

27,217 views • 1 year ago

Hi my friends 🌞 let’s make this Sunday more interesting with latex style🔥 Freya Allan, Lexi Marvel, My Model, Sabrina Carpenter 🔥 👉🏻Subscribe for more content ⚡️ Nano Banana 2 & MiniMax H3 via Hailuo AI Prompt: { "member": "Freya Allan", "subject": { "identity": "facial structure, complexion and proportions of Freya Allan", "hair": "long dark wavy hair over the shoulders", "expression": "head tilted down toward right shoulder, eyes looking up at the lens", "pose": "leaning forward, torso tilted right, right hand on upper right thigh, left arm on metal railing behind her, left knee raised on a step, right leg cropped at the bottom edge" }, "wardrobe": { "silhouette": "short form-fitting high-neck halter dress, gathered at bust, open sides, ruched seams, mid-thigh hem", "material": "glossy latex", "color": "neon pink", "footwear": "only the top edge of a neon pink heel visible on the left foot, rest cropped", "accessories": "opaque black tights, thin hoop earrings partly hidden by hair" }, "scene": { "location": "dilapidated industrial stairwell", "atmosphere": "hazy dusty air catching the light", "background": "peeling paint and rusted wall on the left with small graffiti 'Keor' near a red dragon drawing, metal staircase with diamond-plate treads and weathered railing on the right", "composition": "tall vertical frame, slightly low angle looking up, subject centered and filling the height", "lighting": "dramatic high-contrast warm light from upper left, angular rays on the wall and face, lower right staircase in deep shadow", "camera": "50mm, f/2.8", "grade": "high-contrast, warm highlights, rich shadows" }, "compiled_prompt": "Portrait of Freya Allan, one continuous full-frame exposure. Long dark wavy hair over her shoulders. Head tilted down toward her right shoulder, eyes looking up at the lens. Leaning forward, torso tilted right, right hand on her thigh, left arm on a metal railing. Left knee raised on a step, right leg cropped at the bottom. Short form-fitting glossy neon-pink latex high-neck halter dress with open sides and ruched seams, mid-thigh hem. Opaque black tights, thin hoop earrings. Only the top of a neon-pink heel visible. Dilapidated industrial stairwell, hazy dusty air. Textured peeling wall with small 'Keor' graffiti near a red dragon drawing on the left, metal staircase on the right. Tall vertical low-angle composition. Warm directional light from upper left with specular highlights on the latex, deep shadows on the stairs. 50mm f/2.8, dramatic high-contrast grade." } ```
0:10

Hi my friends 🌞 let’s make this Sunday more interesting with latex style🔥 Freya Allan, Lexi Marvel, My Model, Sabrina Carpenter 🔥 👉🏻Subscribe for more content ⚡️ Nano Banana 2 & MiniMax H3 via Hailuo AI Prompt: { "member": "Freya Allan", "subject": { "identity": "facial structure, complexion and proportions of Freya Allan", "hair": "long dark wavy hair over the shoulders", "expression": "head tilted down toward right shoulder, eyes looking up at the lens", "pose": "leaning forward, torso tilted right, right hand on upper right thigh, left arm on metal railing behind her, left knee raised on a step, right leg cropped at the bottom edge" }, "wardrobe": { "silhouette": "short form-fitting high-neck halter dress, gathered at bust, open sides, ruched seams, mid-thigh hem", "material": "glossy latex", "color": "neon pink", "footwear": "only the top edge of a neon pink heel visible on the left foot, rest cropped", "accessories": "opaque black tights, thin hoop earrings partly hidden by hair" }, "scene": { "location": "dilapidated industrial stairwell", "atmosphere": "hazy dusty air catching the light", "background": "peeling paint and rusted wall on the left with small graffiti 'Keor' near a red dragon drawing, metal staircase with diamond-plate treads and weathered railing on the right", "composition": "tall vertical frame, slightly low angle looking up, subject centered and filling the height", "lighting": "dramatic high-contrast warm light from upper left, angular rays on the wall and face, lower right staircase in deep shadow", "camera": "50mm, f/2.8", "grade": "high-contrast, warm highlights, rich shadows" }, "compiled_prompt": "Portrait of Freya Allan, one continuous full-frame exposure. Long dark wavy hair over her shoulders. Head tilted down toward her right shoulder, eyes looking up at the lens. Leaning forward, torso tilted right, right hand on her thigh, left arm on a metal railing. Left knee raised on a step, right leg cropped at the bottom. Short form-fitting glossy neon-pink latex high-neck halter dress with open sides and ruched seams, mid-thigh hem. Opaque black tights, thin hoop earrings. Only the top of a neon-pink heel visible. Dilapidated industrial stairwell, hazy dusty air. Textured peeling wall with small 'Keor' graffiti near a red dragon drawing on the left, metal staircase on the right. Tall vertical low-angle composition. Warm directional light from upper left with specular highlights on the latex, deep shadows on the stairs. 50mm f/2.8, dramatic high-contrast grade." } ```

KeorUnreal

32,670 views • 1 month ago

Day 11/90 of Inference Engineering How does vLLM work and how is it used in production? Before we discuss how vLLM works internally, it helps to understand what vLLM is. At a high level, vLLM is an inference engine that is designed to serve LLMs to thousands of concurrent users efficiently while managing scarce compute and memory. The goal for vLLM is to maximize throughput and minimize latency; optimizing for the best inference economics and experience for end users. With every request from the end user, it eventually ends up in the engine core, gets scheduled alongside other requests from other concurrent users, executes on the GPU, and updates the KV cache with the new key and value vectors, and streams the tokens back to the user. The Scheduler decides what requests should execute next while continuously batching requests together to maximize GPU utilization. Continuous batching is an inference optimization that allows new requests to join a running batch as other requests finish generating tokens. This helps with keeping the GPU utilization high instead of letting it sit idle waiting for an entire batch to complete generating. After the scheduler dispatches the selected batch to the Model Executor, the Model Executor prepares the tensors and metadata required for inference, retrieves each request’s block table from KV Cache Manager, launches the optimized transformer forward pass on the GPU, computes the logits, updates the KV cache with the new key and value vectors, and finally returns the results for sampling and streaming. The KV Cache Manager uses the PagedAttention memory layout to allocate fixed-size cache blocks on demand and maintains a Free Block Queue on the CPU that tracks which blocks in the GPU’s Paged KV Cache are currently free. When a request needs additional KV cache space, the KV Cache manager takes a free block from the queue and assigns it to that request, thus avoiding an expensive search through GPU memory for available cache blocks. All of these components form the core of vLLM’s inference engine. The Scheduler determines what requests are executed, the Model Executor determines how those requests are executed, the KV Cache Manager determines where each request’s KV cache lives using the PagedAttention Memory Layout. This architecture enables vLLM to serve thousands of concurrent requests with high throughput, low latency, and efficient GPU memory utilization. Heres a little animation that visualizes everything! - I've also completed the forward pass for my mnist.c project. I had a nice chat with shrey birmiwal, such a knowledgeable guy. Excited to learn more about vLLM and implement a tiny-vLLM one day.
0:14

Day 11/90 of Inference Engineering How does vLLM work and how is it used in production? Before we discuss how vLLM works internally, it helps to understand what vLLM is. At a high level, vLLM is an inference engine that is designed to serve LLMs to thousands of concurrent users efficiently while managing scarce compute and memory. The goal for vLLM is to maximize throughput and minimize latency; optimizing for the best inference economics and experience for end users. With every request from the end user, it eventually ends up in the engine core, gets scheduled alongside other requests from other concurrent users, executes on the GPU, and updates the KV cache with the new key and value vectors, and streams the tokens back to the user. The Scheduler decides what requests should execute next while continuously batching requests together to maximize GPU utilization. Continuous batching is an inference optimization that allows new requests to join a running batch as other requests finish generating tokens. This helps with keeping the GPU utilization high instead of letting it sit idle waiting for an entire batch to complete generating. After the scheduler dispatches the selected batch to the Model Executor, the Model Executor prepares the tensors and metadata required for inference, retrieves each request’s block table from KV Cache Manager, launches the optimized transformer forward pass on the GPU, computes the logits, updates the KV cache with the new key and value vectors, and finally returns the results for sampling and streaming. The KV Cache Manager uses the PagedAttention memory layout to allocate fixed-size cache blocks on demand and maintains a Free Block Queue on the CPU that tracks which blocks in the GPU’s Paged KV Cache are currently free. When a request needs additional KV cache space, the KV Cache manager takes a free block from the queue and assigns it to that request, thus avoiding an expensive search through GPU memory for available cache blocks. All of these components form the core of vLLM’s inference engine. The Scheduler determines what requests are executed, the Model Executor determines how those requests are executed, the KV Cache Manager determines where each request’s KV cache lives using the PagedAttention Memory Layout. This architecture enables vLLM to serve thousands of concurrent requests with high throughput, low latency, and efficient GPU memory utilization. Heres a little animation that visualizes everything! - I've also completed the forward pass for my mnist.c project. I had a nice chat with shrey birmiwal, such a knowledgeable guy. Excited to learn more about vLLM and implement a tiny-vLLM one day.

max fu

70,797 views • 2 months ago

Seedance 2 just dropped. This was my first test and it changed everything. Perfect lip sync. Realistic movement. Multi-shot support. Natural voice completely human, zero robotic tone. And the SFX? It nailed the sound effects automatically. Every detail synced perfectly. This is a game-changer. From now on, automation isn't theory it's real. And yes, it's happening on Crafft. With a high-fidelity model like this + the right prompting rules and skills, you can generate endless variations that work every single time. Crafft agents will handle the whole process. One model, infinite outputs. You can try Seedance 2 on CapCut right now. RT + comment "PROMPT" and I'll send you the exact setup (follow so I can DM)
0:13

Seedance 2 just dropped. This was my first test and it changed everything. Perfect lip sync. Realistic movement. Multi-shot support. Natural voice completely human, zero robotic tone. And the SFX? It nailed the sound effects automatically. Every detail synced perfectly. This is a game-changer. From now on, automation isn't theory it's real. And yes, it's happening on Crafft. With a high-fidelity model like this + the right prompting rules and skills, you can generate endless variations that work every single time. Crafft agents will handle the whole process. One model, infinite outputs. You can try Seedance 2 on CapCut right now. RT + comment "PROMPT" and I'll send you the exact setup (follow so I can DM)

Ahmad

31,558 views • 7 months ago