I am sort of new to this and by this I mean using ComfyUI. Can I use ComfyUi and run the full model of Hunyuan on a rented server with the equipment to run Hunyuan?
absolutely.. if you rent a large enough model you can run the full meal deal. The versions we're using here are optimized for 16gb-24gb (this version in this video uses settings and further quantized models).. if you rented a GPU with 48+ VRAM you could run the full thing and be competing with SORA
@@cognibuild I use to be a programmer back in the day and I also messed with Unreal Engine and ComfyUI reminds me of blueprints. Is there a way that you can do image to video using ComfyUi with Hunyuan?
@@TheAiKonnekt Not yet with Hunyuan but we've been using LTX and getting some really awesome stuff. You can find a video on that and I have installers on the patreon
I bet you could get a better result if you use the low vram workflow in the examples folder. I can't confirm for sure since I have 16 GB, but I tried purposely lowering the vram requirements as much as possible by using the blockswap node and setting double to 20 and single to 40, the fp8 scaled model and sage attention. I never saw memory usage go beyond 6 GB when the sampler was running at 720x480 at 85 frames. When it got to the decode step, it went just barely under 8 GB. Since you have a 8 GB card, you should be able to do this as well. It took about 4 minutes on a 4080.
![[Bad] - YouTube ВСЁ? (победа VK и прощание с подписчиками)](http://i.ytimg.com/vi/pQ1bCOW5270/mqdefault.jpg)
I love your channel, you bring a lot of value, in this new discovered space, everyday is newer and better. hope you keep making great videos.
thank you for the compliment and good to have you around!
Thank you, bro! My dishes are actually going to be done now, or not. How do I add a lora into the work flow so the video has a specific character?
Hi, thanks for you tutorial. Is it possible to use LORA to get consistent image for low vram Hunyuan?
did you figure that out yet?
@@propelproject no. In the other video I heard that there is no custom Lora yet for the Hunyuan.
Thanks for the video
Would this work for my macbook pro? M4 Max 64gb unified mem? Or any other solution that would work for me?
it depends on your GPU.. do you know hgow much vram you have?
@@cognibuild I have a 40-core GPU. The 64gb unified memory serves as both system RAM and VRAM for macs.
I am sort of new to this and by this I mean using ComfyUI. Can I use ComfyUi and run the full model of Hunyuan on a rented server with the equipment to run Hunyuan?
absolutely.. if you rent a large enough model you can run the full meal deal. The versions we're using here are optimized for 16gb-24gb (this version in this video uses settings and further quantized models).. if you rented a GPU with 48+ VRAM you could run the full thing and be competing with SORA
@@cognibuild I use to be a programmer back in the day and I also messed with Unreal Engine and ComfyUI reminds me of blueprints. Is there a way that you can do image to video using ComfyUi with Hunyuan?
@@TheAiKonnekt Not yet with Hunyuan but we've been using LTX and getting some really awesome stuff. You can find a video on that and I have installers on the patreon
@@cognibuild I will definitely take a look at it. Man, thanks for all that you do.
My Test 1: 13 minutes for a 2 second video at 320x240 14 FPS, using 8Gig Ram card.
thats actrually really great! way better than I expected
come shar it on the Discord
I bet you could get a better result if you use the low vram workflow in the examples folder. I can't confirm for sure since I have 16 GB, but I tried purposely lowering the vram requirements as much as possible by using the blockswap node and setting double to 20 and single to 40, the fp8 scaled model and sage attention. I never saw memory usage go beyond 6 GB when the sampler was running at 720x480 at 85 frames. When it got to the decode step, it went just barely under 8 GB. Since you have a 8 GB card, you should be able to do this as well. It took about 4 minutes on a 4080.
FP 8 diffuser dropped since this came out, maybe that one would perform better?
Thanks
Bro got 12 gb can it run
Yes. I ran it with only 8Gig but only at 320x240 and it took 13 minutes for 2 seconds. You will have a better outcome if you have 12 Gig.