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.

WanVideo sliding context window test on the 480p 14B T2V -model, worked surprisingly well but incredibly slow (50 mins on a 5090, 513 frames at 832x480)

Jukka Seppänen

6,451 subscribers

13,085 views • 1 year ago •via X (Twitter)

Anya Rossi• Live Now

Private livecam show

0 Comments

No comments available

Comments from the original post will appear here

Related Videos

WanVideo has been lot of fun, currently playing with applying sliding context windowing similar to AnimateDiff, the 1.3B model is especially suitable for this due to it's speed, here's 1025 frames in one go, under 5GB VRAM used, but took about 10 mins on a 5090 with 30 steps.

WanVideo has been lot of fun, currently playing with applying sliding context windowing similar to AnimateDiff, the 1.3B model is especially suitable for this due to it's speed, here's 1025 frames in one go, under 5GB VRAM used, but took about 10 mins on a 5090 with 30 steps.

Jukka Seppänen

13,808 views • 1 year ago

Today, I went out during a snowstorm to test Tesla's FSD V14.2.1. The car adapted in real time to the poor road conditions and performed incredibly well. My Model Y even counter-steered at one moment to offset some sliding. Mind blown! FSD handles snowy conditions very well.

Today, I went out during a snowstorm to test Tesla's FSD V14.2.1. The car adapted in real time to the poor road conditions and performed incredibly well. My Model Y even counter-steered at one moment to offset some sliding. Mind blown! FSD handles snowy conditions very well.

Sawyer Merritt

134,520 views • 8 months ago

Everyone's comparing the DGX Spark to a 5090 and calling it slow. I think that's the wrong comparison. I ran Qwen3.6 35B-A3B FP8 with the full 262K context window enabled (~96GB RAM) — something gaming GPUs can't really do. Results: 🟢No context: 51.3 tok/s, TTFT 110ms 🟣200K prefill: 34.6 tok/s, TTFT 85s (~2,341 tok/s prefill) Prefill is way faster than a Mac. And 35 tok/s deep into 200K context, on a model this strong, is genuinely usable. The Spark plays a different game.

Everyone's comparing the DGX Spark to a 5090 and calling it slow. I think that's the wrong comparison. I ran Qwen3.6 35B-A3B FP8 with the full 262K context window enabled (~96GB RAM) — something gaming GPUs can't really do. Results: 🟢No context: 51.3 tok/s, TTFT 110ms 🟣200K prefill: 34.6 tok/s, TTFT 85s (~2,341 tok/s prefill) Prefill is way faster than a Mac. And 35 tok/s deep into 200K context, on a model this strong, is genuinely usable. The Spark plays a different game.

stevibe

34,141 views • 3 months ago

A 5-second 480P video, generated entirely on a Mac, in 30 seconds. No CUDA, no cloud, 3.9 GiB of memory. FastMetal brings the FastWan-QAD family to Apple Silicon. The DiT, the DMD sampler, and the decoder all run on Metal through MLX, INT8 by default. Three models: 1.3B at 480P, 5B at 720P, 14B for quality. 📷 Blog: 📷 Code: 📷 Models:

A 5-second 480P video, generated entirely on a Mac, in 30 seconds. No CUDA, no cloud, 3.9 GiB of memory. FastMetal brings the FastWan-QAD family to Apple Silicon. The DiT, the DMD sampler, and the decoder all run on Metal through MLX, INT8 by default. Three models: 1.3B at 480P, 5B at 720P, 14B for quality. 📷 Blog: 📷 Code: 📷 Models:

Hao AI Lab

16,924 views • 5 days ago

Translating a full-length news article on iPhone 17 Pro with TranslateGemma from Google DeepMind The model works incredibly well on long text Running fully on-device with MLX

Translating a full-length news article on iPhone 17 Pro with TranslateGemma from Google DeepMind The model works incredibly well on long text Running fully on-device with MLX

Adrien Grondin

15,493 views • 6 months ago

Introducing Large Video Planner (LVP-14B) — a robot foundation model that actually generalizes. LVP is built on video gen, not VLA. As my final work at Massachusetts Institute of Technology (MIT), LVP has all its eval tasks proposed by third parties as a maximum stress test, but it excels!🤗

Introducing Large Video Planner (LVP-14B) — a robot foundation model that actually generalizes. LVP is built on video gen, not VLA. As my final work at Massachusetts Institute of Technology (MIT), LVP has all its eval tasks proposed by third parties as a maximum stress test, but it excels!🤗

Boyuan Chen

100,357 views • 7 months ago

Reform UK's Nigel Farage is optimistic on the party's chances against PM Keir Starmer, saying 'We're doing incredibly well in many, many traditional Labour areas'. Polls predict Reform UK could make huge gains in the election, a key test for Starmer's sliding popularity 🎥🇬🇧

Reform UK's Nigel Farage is optimistic on the party's chances against PM Keir Starmer, saying 'We're doing incredibly well in many, many traditional Labour areas'. Polls predict Reform UK could make huge gains in the election, a key test for Starmer's sliding popularity 🎥🇬🇧

AFP News Agency

10,377 views • 3 months ago

TIPPERARY GOAL - 64 mins Darragh Stakelum at the end of a well-worked move to raise a green flag for the hosts #rtegaa #allianzleagues 📺Tipperary GAA v Galway GAA Official on RTÉ2 and RTÉ Player

TIPPERARY GOAL - 64 mins Darragh Stakelum at the end of a well-worked move to raise a green flag for the hosts #rtegaa #allianzleagues 📺Tipperary GAA v Galway GAA Official on RTÉ2 and RTÉ Player

The Sunday Game

20,835 views • 7 months ago

ran Manaforge Omega on the alpha to test out the new raid frames. I don't think they're quite there yet, but it's a good first step

ran Manaforge Omega on the alpha to test out the new raid frames. I don't think they're quite there yet, but it's a good first step

Naomi 🐾

185,881 views • 10 months ago

Qwen3.5 runs quite well in mlx-lm. Awesome that we have a frontier-level hybrid model. The context gets longer but the inference speed and memory use barely change. Here's the Q4 generating a space invaders game on an M3 Ultra. Generated 4,120 tokens at 37.6 tok/s.

Qwen3.5 runs quite well in mlx-lm. Awesome that we have a frontier-level hybrid model. The context gets longer but the inference speed and memory use barely change. Here's the Q4 generating a space invaders game on an M3 Ultra. Generated 4,120 tokens at 37.6 tok/s.

Awni Hannun

48,673 views • 6 months ago

Idefics3-Llama is out! 💥 It's a multimodal model based on Llama 3.1 that accepts arbitrary number of interleaved images with text with a huge context window (10k tokens!) 😍 Link to demo and model in the next one 😏

Idefics3-Llama is out! 💥 It's a multimodal model based on Llama 3.1 that accepts arbitrary number of interleaved images with text with a huge context window (10k tokens!) 😍 Link to demo and model in the next one 😏

merve

28,017 views • 2 years ago

CHINA JUST DROPPED AN AI CODING MODEL WITH A 1M CONTEXT WINDOW. And I connected it to Claude Code to see what it could actually do. Meet GLM-X Preview On paper, a few things immediately stood out: → 1M context window → Agentic coding capabilities → Works inside Claude Code → Designed for large-scale coding and reasoning workflows But specs don't matter much if the model can't deliver in practice. So I gave it a real-world task. THE TEST One prompt: > Build a modern AI lead generation dashboard using React and Tailwind CSS. Requirements: → Dark mode → Analytics dashboard → Lead table → Email outreach section → Responsive design → Production-ready component structure Instead of generating a few snippets, it planned the architecture, generated the dashboard components, created the Tailwind configuration, and walked through the implementation requirements. What impressed me most wasn't the code itself. It was how well it maintained context throughout the workflow. That's where a 1M context window starts becoming useful. Less time re-explaining requirements. Less context loss. More room for complex projects. The AI coding race is getting very interesting. And it's no longer just GPT, Claude, and Gemini competing for attention. Results from my test below 👇

CHINA JUST DROPPED AN AI CODING MODEL WITH A 1M CONTEXT WINDOW. And I connected it to Claude Code to see what it could actually do. Meet GLM-X Preview On paper, a few things immediately stood out: → 1M context window → Agentic coding capabilities → Works inside Claude Code → Designed for large-scale coding and reasoning workflows But specs don't matter much if the model can't deliver in practice. So I gave it a real-world task. THE TEST One prompt: > Build a modern AI lead generation dashboard using React and Tailwind CSS. Requirements: → Dark mode → Analytics dashboard → Lead table → Email outreach section → Responsive design → Production-ready component structure Instead of generating a few snippets, it planned the architecture, generated the dashboard components, created the Tailwind configuration, and walked through the implementation requirements. What impressed me most wasn't the code itself. It was how well it maintained context throughout the workflow. That's where a 1M context window starts becoming useful. Less time re-explaining requirements. Less context loss. More room for complex projects. The AI coding race is getting very interesting. And it's no longer just GPT, Claude, and Gemini competing for attention. Results from my test below 👇

Md Riyazuddin

31,199 views • 2 months ago

'You just see Alex Pretti on the ground, we're missing a broader context' — Michael Shellenberger speaks to Tucker Carlson 'These are incredibly well-organized efforts to disrupt law enforcement operations'

'You just see Alex Pretti on the ground, we're missing a broader context' — Michael Shellenberger speaks to Tucker Carlson 'These are incredibly well-organized efforts to disrupt law enforcement operations'

RT

17,240 views • 6 months ago

Wan2.2 is now natively supported in ComfyUI on Day 0! 🔹 A next-gen video model with MoE (Mixture of Experts) architecture with dual noise experts, under Apache 2.0 license! - Cinematic-level Aesthetic Control - Large-scale Complex Motion - Precise Semantic Compliance 📚 Versions available: - Wan2.2-TI2V-5B: FP16 - Wan2.2-I2V-14B: FP16/FP8 - Wan2.2-T2V-14B: FP16/FP8 💻 Down to 8GB VRAM requirement for the 5B version with ComfyUI auto-offloading.

Wan2.2 is now natively supported in ComfyUI on Day 0! 🔹 A next-gen video model with MoE (Mixture of Experts) architecture with dual noise experts, under Apache 2.0 license! - Cinematic-level Aesthetic Control - Large-scale Complex Motion - Precise Semantic Compliance 📚 Versions available: - Wan2.2-TI2V-5B: FP16 - Wan2.2-I2V-14B: FP16/FP8 - Wan2.2-T2V-14B: FP16/FP8 💻 Down to 8GB VRAM requirement for the 5B version with ComfyUI auto-offloading.

ComfyUI

84,512 views • 1 year ago

My thoughts on Grok 3 after 24hrs: - it’s *really* good for code - context window is HUGE - utilizes context extremely well - great at instruction following (agents!) - delightful coworker personality Here’s a 5min demo of how I’ll be using it in my code workflow going forward.

My thoughts on Grok 3 after 24hrs: - it’s really good for code - context window is HUGE - utilizes context extremely well - great at instruction following (agents!) - delightful coworker personality Here’s a 5min demo of how I’ll be using it in my code workflow going forward.

Mckay Wrigley

2,897,647 views • 1 year ago

Seedance 2.5, the world's best AI video model, is here! Use it with up to 50 reference images at once and generate consistent 30s clips. Available now on LetzAI for as low as $0.19 (480p) and $0.38 (720p) per second.

Seedance 2.5, the world's best AI video model, is here! Use it with up to 50 reference images at once and generate consistent 30s clips. Available now on LetzAI for as low as $0.19 (480p) and $0.38 (720p) per second.

LetzAI

375,442 views • 18 days ago

Never bet against Cursor. Subagents solve so many issues between different models it's kind of insane, especially with Composer 1. Composer 1 can quickly gather context for a model like GPT-5.2 Codex, which otherwise could take on the order of 10 mins just to search files. Definitely one of the most underrated models right now. Now you get the benefits of it's speed, but can handoff to a smarter model at implementation. The video I attached is just a simple example, but you can see how long GPT 5.2 takes to gather context on a very simple problem. For harder problems, the search time grows significantly.

Never bet against Cursor. Subagents solve so many issues between different models it's kind of insane, especially with Composer 1. Composer 1 can quickly gather context for a model like GPT-5.2 Codex, which otherwise could take on the order of 10 mins just to search files. Definitely one of the most underrated models right now. Now you get the benefits of it's speed, but can handoff to a smarter model at implementation. The video I attached is just a simple example, but you can see how long GPT 5.2 takes to gather context on a very simple problem. For harder problems, the search time grows significantly.

ostyn

19,708 views • 7 months ago

SITUATION ANALYSIS: Religious texts may work as AI alignment tools. What Tim Hwang says Institute for a Christian Machine Intelligence (ICMI) is testing: • Put Psalms in the model context window • Run the model on moral decision benchmarks • Measure changes on virtue-based evals • Test Christian reasoning as a steering method • Try images like the Annunciation in multimodal models • Watch how the model uses religious symbols in reasoning • Compare theological alignment against secular approaches The weird part: Tim says the Psalms result improved Virtue Bench in a statistically significant way.

SITUATION ANALYSIS: Religious texts may work as AI alignment tools. What Tim Hwang says Institute for a Christian Machine Intelligence (ICMI) is testing: • Put Psalms in the model context window • Run the model on moral decision benchmarks • Measure changes on virtue-based evals • Test Christian reasoning as a steering method • Try images like the Annunciation in multimodal models • Watch how the model uses religious symbols in reasoning • Compare theological alignment against secular approaches The weird part: Tim says the Psalms result improved Virtue Bench in a statistically significant way.

MTS

31,144 views • 3 months ago

Google presents Still-Moving Customized Video Generation without Customized Video Data Customizing text-to-image (T2I) models has seen tremendous progress recently, particularly in areas such as personalization, stylization, and conditional generation. However, expanding this progress to video generation is still in its infancy, primarily due to the lack of customized video data. In this work, we introduce Still-Moving, a novel generic framework for customizing a text-to-video (T2V) model, without requiring any customized video data. The framework applies to the prominent T2V design where the video model is built over a text-to-image (T2I) model (e.g., via inflation). We assume access to a customized version of the T2I model, trained only on still image data (e.g., using DreamBooth or StyleDrop). Naively plugging in the weights of the customized T2I model into the T2V model often leads to significant artifacts or insufficient adherence to the customization data. To overcome this issue, we train lightweight Spatial Adapters that adjust the features produced by the injected T2I layers. Importantly, our adapters are trained on "frozen videos" (i.e., repeated images), constructed from image samples generated by the customized T2I model. This training is facilitated by a novel Motion Adapter module, which allows us to train on such static videos while preserving the motion prior of the video model. At test time, we remove the Motion Adapter modules and leave in only the trained Spatial Adapters. This restores the motion prior of the T2V model while adhering to the spatial prior of the customized T2I model. We demonstrate the effectiveness of our approach on diverse tasks including personalized, stylized, and conditional generation. In all evaluated scenarios, our method seamlessly integrates the spatial prior of the customized T2I model with a motion prior supplied by the T2V model.

Google presents Still-Moving Customized Video Generation without Customized Video Data Customizing text-to-image (T2I) models has seen tremendous progress recently, particularly in areas such as personalization, stylization, and conditional generation. However, expanding this progress to video generation is still in its infancy, primarily due to the lack of customized video data. In this work, we introduce Still-Moving, a novel generic framework for customizing a text-to-video (T2V) model, without requiring any customized video data. The framework applies to the prominent T2V design where the video model is built over a text-to-image (T2I) model (e.g., via inflation). We assume access to a customized version of the T2I model, trained only on still image data (e.g., using DreamBooth or StyleDrop). Naively plugging in the weights of the customized T2I model into the T2V model often leads to significant artifacts or insufficient adherence to the customization data. To overcome this issue, we train lightweight Spatial Adapters that adjust the features produced by the injected T2I layers. Importantly, our adapters are trained on "frozen videos" (i.e., repeated images), constructed from image samples generated by the customized T2I model. This training is facilitated by a novel Motion Adapter module, which allows us to train on such static videos while preserving the motion prior of the video model. At test time, we remove the Motion Adapter modules and leave in only the trained Spatial Adapters. This restores the motion prior of the T2V model while adhering to the spatial prior of the customized T2I model. We demonstrate the effectiveness of our approach on diverse tasks including personalized, stylized, and conditional generation. In all evaluated scenarios, our method seamlessly integrates the spatial prior of the customized T2I model with a motion prior supplied by the T2V model.

AK

40,485 views • 2 years ago