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 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

6,725 subscribers

14,778 views • 1 year ago •via X (Twitter)

Arts Science & Technology

Anya Rossi• Live Now

Private livecam show

10 Comments

Reverent Elusarca1 year ago

Thanks for your hard work! Also would love to see every benchmark result with 5090. I believe you had 4090 before so how is the performance so far

Jukka Seppänen1 year ago

I actually have both, I couldn't justify upgrading but 2nd rig for all this stuff I couldn't resist, so I splurged. It's considerably faster, around ~40% or so with all optimizations. Slightly difficult to use still as you have to compile many things from source. Runs hot though

Page to Pixel Publishing2 years ago

The Art of Flight is a homage to 80s/90s arcade action shmups with a fresh twist on the genre. Pilot multiple ships at the same time to take on oncoming waves of enemies in this fast paced space shooter. Wishlist on Steam today!

OverPowered1 year ago

A whole minute video in 10 minutes is insane

zazoum1 year ago

Is it like stitching timed prompts?Is it implemented on the repo, and if yes do you have a workflow in the examples to experiment with?

Jukka Seppänen1 year ago

Currently only single prompt with sliding context windows, but I plan to test multiple ones if I can figure all that out.

GJohGJ ©1 year ago

Wait what? 1025 frames? on what framerate?

Jukka Seppänen1 year ago

The model is supposed to output 16fps, though people seem to debate that. This video was saved with 16fps.

Neural_Ghost1 year ago

Wow

Gorilla Rogue A.I.1 year ago

Whoa! @cocktailpeanut check this out 👀

Related Videos

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 • 8 months ago

THIS $1,300 CARD HAS THE SAME 32GB OF VRAM AS A $2,000+ RTX 5090 - FOR LOCAL AI, THAT'S THE SPEC THAT DECIDES EVERYTHING. that clip is the gigabyte radeon ai pro r9700 ai top, 32gb of vram, built for local ai. the pitch is simple: the same memory capacity as an rtx 5090, for roughly a third of what a 5090 actually sells for. here's why that matters, and where it doesn't. for ai, vram is the gate. a model either fits in memory or it doesn't. 32gb loads models a 16gb card can't touch, full stop. so for running local llms and image models, this card does the one thing that counts, cheap. honest line: this is not a 5090 for gaming. the 5090 is far faster at raster and ray tracing. they only match on how much vram they carry. if you want frames, buy the 5090. if you want to fit big models locally, you just found the same 32gb for a third of the money. no $2,000+ price tag, no cloud gpu rental, no model too big for 16gb. the uncomfortable part isn't that amd is cheaper. it's that for local ai you were about to overpay for gaming power you'd never use. follow RetroChainer for more on paying for the spec that matters, and bookmark this before your next gpu.

THIS $1,300 CARD HAS THE SAME 32GB OF VRAM AS A $2,000+ RTX 5090 - FOR LOCAL AI, THAT'S THE SPEC THAT DECIDES EVERYTHING. that clip is the gigabyte radeon ai pro r9700 ai top, 32gb of vram, built for local ai. the pitch is simple: the same memory capacity as an rtx 5090, for roughly a third of what a 5090 actually sells for. here's why that matters, and where it doesn't. for ai, vram is the gate. a model either fits in memory or it doesn't. 32gb loads models a 16gb card can't touch, full stop. so for running local llms and image models, this card does the one thing that counts, cheap. honest line: this is not a 5090 for gaming. the 5090 is far faster at raster and ray tracing. they only match on how much vram they carry. if you want frames, buy the 5090. if you want to fit big models locally, you just found the same 32gb for a third of the money. no $2,000+ price tag, no cloud gpu rental, no model too big for 16gb. the uncomfortable part isn't that amd is cheaper. it's that for local ai you were about to overpay for gaming power you'd never use. follow RetroChainer for more on paying for the spec that matters, and bookmark this before your next gpu.

RetroChainer

38,844 views • 1 month ago

MiniMax H3 🧐 Comfy Tested -Minimum requirements for 480p video on this model is a 3060 with 12GB vram, 32GB of system ram and a good nvme SSD. -generating a 5 second (124 frames) 480p (864x480) video on this system and it took a bit less than 9 minutes end to end (20 steps)

MiniMax H3 🧐 Comfy Tested -Minimum requirements for 480p video on this model is a 3060 with 12GB vram, 32GB of system ram and a good nvme SSD. -generating a 5 second (124 frames) 480p (864x480) video on this system and it took a bit less than 9 minutes end to end (20 steps)

Stable Diffusion Tutorials

102,325 views • 1 month ago

anything under deepseek v4 flash basically doesn't exist to me. with <200GB of vram, i'm getting ~800K context @ 200 tokens per second. It's not quite Cerebras fast, but damn. It is a joy to use, and its basically somewhere in the opus 4.7/4.8 range. It's incredibly good at browser control too. this model is a beast. w/ qwen 3.8 27B being multimodal though, it might just be worth switching to, especially if you want to run a lot of parallel agents will 300toks/s be possible? Pretty addicting let me tell you. Imagine a computer use model this fast. watch below, this is basically human speed.

anything under deepseek v4 flash basically doesn't exist to me. with <200GB of vram, i'm getting ~800K context @ 200 tokens per second. It's not quite Cerebras fast, but damn. It is a joy to use, and its basically somewhere in the opus 4.7/4.8 range. It's incredibly good at browser control too. this model is a beast. w/ qwen 3.8 27B being multimodal though, it might just be worth switching to, especially if you want to run a lot of parallel agents will 300toks/s be possible? Pretty addicting let me tell you. Imagine a computer use model this fast. watch below, this is basically human speed.

am.will

72,729 views • 1 month ago

I trained this LTX LTX 2.3 LoRA of George Costanza at home on my 5090 in about a day with AI Toolkit. I generated this 30 second video with ComfyUI on my 5090 in 6 minutes. Open source is, always has been, and always will be, the future of generative AI. (SOUND ON)

I trained this LTX LTX 2.3 LoRA of George Costanza at home on my 5090 in about a day with AI Toolkit. I generated this 30 second video with ComfyUI on my 5090 in 6 minutes. Open source is, always has been, and always will be, the future of generative AI. (SOUND ON)

Ostris

395,312 views • 6 months ago

I had a lot of fun playing with the steam frame, especially Half-Life Alyx Standalone but you want to know something insane... I got Half-Life Alyx running Standalone on the Quest 3... Still needs a lot of optimization and a lot of work but watch out later this week for more info. This is all done with the same AI modding tools that I've been using. It's insane. Still requires you to own Half-Life Alyx and it requires the files from the steam frame branch of the game.

I had a lot of fun playing with the steam frame, especially Half-Life Alyx Standalone but you want to know something insane... I got Half-Life Alyx running Standalone on the Quest 3... Still needs a lot of optimization and a lot of work but watch out later this week for more info. This is all done with the same AI modding tools that I've been using. It's insane. Still requires you to own Half-Life Alyx and it requires the files from the steam frame branch of the game.

BMFVR

21,378 views • 15 days ago

30 sec MiniMax H3 clip generated locally w/ a 4-step fused ref2va (omni) model in Maestro. (Coming soon in Maestro v2.3) This is not a multi-clip sliding window. With sliding windows you can make very long seamless clips; as the last several latent frames of audio & video are sent to each consecutive clip, but they lose context of anything out of frame. That can be partially solved with omni references. But if you want longer clips that retain full context, this is that. Thanks to TechHalla for his epic prompts! This one is slightly modified. Maestro also can take any prompt, such as ones designed for Seedance 2.5 & rewrite them for whatever model you have selected. 30s clips in H3 are still very experimental. But still impressive for a local highly distilled, low step model.

30 sec MiniMax H3 clip generated locally w/ a 4-step fused ref2va (omni) model in Maestro. (Coming soon in Maestro v2.3) This is not a multi-clip sliding window. With sliding windows you can make very long seamless clips; as the last several latent frames of audio & video are sent to each consecutive clip, but they lose context of anything out of frame. That can be partially solved with omni references. But if you want longer clips that retain full context, this is that. Thanks to TechHalla for his epic prompts! This one is slightly modified. Maestro also can take any prompt, such as ones designed for Seedance 2.5 & rewrite them for whatever model you have selected. 30s clips in H3 are still very experimental. But still impressive for a local highly distilled, low step model.

Blaine Brown

22,795 views • 9 days ago

Work on this Gwen seq has been a lot of fun. It's not about how to animate badass spidey. rather, it's about how to transition creatively from one shot to the next. My leader and I think about what if we use "Train" for transition for the next? #SpiderManAcrossTheSpiderVerse

Work on this Gwen seq has been a lot of fun. It's not about how to animate badass spidey. rather, it's about how to transition creatively from one shot to the next. My leader and I think about what if we use "Train" for transition for the next? #SpiderManAcrossTheSpiderVerse

Seng

44,906 views • 3 years ago

We used to buy more VRAM. Now we’re finding ways around the VRAM limit. This little DDR5 SO-DIMM setup is paired with an old NVIDIA Tesla P40 in the video. The P40 is an old datacenter GPU with 24GB of VRAM. By today’s standards, it’s not fast. But combine that with a machine carrying 128GB of system memory, and suddenly you have a very different local AI box. The GPU handles what it’s good at. The system memory gives the model somewhere to go when VRAM runs out. You obviously pay for that with bandwidth and inference speed. But this is the part I find interesting: Local AI is making “how much memory do I have?” almost as important as “how powerful is my GPU?” We’ve spent years chasing faster GPUs. Maybe the next wave of local AI hardware will be about finding smarter ways to use all the memory we already have. More VRAM is still better. But it’s no longer the only way to run a bigger model.

We used to buy more VRAM. Now we’re finding ways around the VRAM limit. This little DDR5 SO-DIMM setup is paired with an old NVIDIA Tesla P40 in the video. The P40 is an old datacenter GPU with 24GB of VRAM. By today’s standards, it’s not fast. But combine that with a machine carrying 128GB of system memory, and suddenly you have a very different local AI box. The GPU handles what it’s good at. The system memory gives the model somewhere to go when VRAM runs out. You obviously pay for that with bandwidth and inference speed. But this is the part I find interesting: Local AI is making “how much memory do I have?” almost as important as “how powerful is my GPU?” We’ve spent years chasing faster GPUs. Maybe the next wave of local AI hardware will be about finding smarter ways to use all the memory we already have. More VRAM is still better. But it’s no longer the only way to run a bigger model.

KyzoroX

35,554 views • 20 days ago

Sony on gaming. Here's the exact quote from the call "In gaming business, and moving away from a hardware centric business to, more to, community based engagement business. That has been increasing.” Here's a link to the actual earnings call with Sony's translation. 8 Mins 30 seconds in. I don't even know why this is being argued. It's a good thing.

Sony on gaming. Here's the exact quote from the call "In gaming business, and moving away from a hardware centric business to, more to, community based engagement business. That has been increasing.” Here's a link to the actual earnings call with Sony's translation. 8 Mins 30 seconds in. I don't even know why this is being argued. It's a good thing.

Destin

29,242 views • 1 year ago

"Will You Remember Us" It's been one hell of a ride with these guys for the past 6 years. They, especially one of 'em, mean a whole lot to me! But Requiem don't just mean a lot to only me. Show support to all the amazing artists featured! (in replies & @'s in the video)

"Will You Remember Us" It's been one hell of a ride with these guys for the past 6 years. They, especially one of 'em, mean a whole lot to me! But Requiem don't just mean a lot to only me. Show support to all the amazing artists featured! (in replies & @'s in the video)

The Korborium

18,841 views • 1 month ago

This is sad. I posted about this yesterday because she wrote a nice tribute to him but she has been acting as if she didn't like him in recent years. It's a reminder to not let politics interfere with close relationships, but for me it's hard with people who push it on me a lot.

This is sad. I posted about this yesterday because she wrote a nice tribute to him but she has been acting as if she didn't like him in recent years. It's a reminder to not let politics interfere with close relationships, but for me it's hard with people who push it on me a lot.

Melanie

79,832 views • 2 months ago

Since Covid, golf has become increasingly more mainstream. Scottie Scheffler was asked ahead of the Travelers Championship how he feels about the impact it’s had on the game: “Yeah, I think it's great for golf to become more mainstream. I think you see, especially since COVID, there's been a tremendous growth in the amount of people that are playing golf. “I think it's the greatest game in the world. You learn a lot about life. I met some of my best friends in the whole world playing golf. So for people getting involved in the sport I think is tremendous. “Like I said, getting to spend four hours with somebody outside hanging out playing a game is something that's pretty rare in the busyness I think that we live in today. So getting out taking some time with your buddies and going to play golf is one of the most fun things you can do. Especially when you start playing, you learn a lot about life and how to handle things and it's a challenging game as well, so overall I think it's just a great sport for anybody to get involved in.” PGA TOUR TravelersChamp

Since Covid, golf has become increasingly more mainstream. Scottie Scheffler was asked ahead of the Travelers Championship how he feels about the impact it’s had on the game: “Yeah, I think it's great for golf to become more mainstream. I think you see, especially since COVID, there's been a tremendous growth in the amount of people that are playing golf. “I think it's the greatest game in the world. You learn a lot about life. I met some of my best friends in the whole world playing golf. So for people getting involved in the sport I think is tremendous. “Like I said, getting to spend four hours with somebody outside hanging out playing a game is something that's pretty rare in the busyness I think that we live in today. So getting out taking some time with your buddies and going to play golf is one of the most fun things you can do. Especially when you start playing, you learn a lot about life and how to handle things and it's a challenging game as well, so overall I think it's just a great sport for anybody to get involved in.” PGA TOUR TravelersChamp

Flushing It

539,218 views • 3 months ago

JUST IN—President Trump speaks to the media outside the courtroom and confirms he was never entitled to a jury in the NY "fraud" case: "This witch hunt does not allow me a jury. A lot of people say no that can't be possible, someone didn't check a box. That has nothing to do with it. Under 63.12 you are not entitled to a jury. First time it's ever been used for a purpose like this. Never been used. They used it on me." "They're coordinating with Washington. 100%." Letitia James ran on "getting Trump" and the state of New York is about to rob the Trump family of a lifetime of work and achievement over, with no jury, over a "crime" with no victims.

JUST IN—President Trump speaks to the media outside the courtroom and confirms he was never entitled to a jury in the NY "fraud" case: "This witch hunt does not allow me a jury. A lot of people say no that can't be possible, someone didn't check a box. That has nothing to do with it. Under 63.12 you are not entitled to a jury. First time it's ever been used for a purpose like this. Never been used. They used it on me." "They're coordinating with Washington. 100%." Letitia James ran on "getting Trump" and the state of New York is about to rob the Trump family of a lifetime of work and achievement over, with no jury, over a "crime" with no victims.

Charlie Kirk

1,917,455 views • 3 years ago