TenStrip

/

10Eros-Max

Like 964

Properly configured, a two staged workflow will always be more memory efficient than a single staged one for the same final resolution and duration, because the hard part is rendered at a lower resolution, while the high res refiner is much shorter.

The clip extension also can be adapted. But in any case, Beta7 should still work fine on a single stage pipeline with proper sampling/scheduling . Due to how it is designed, VSA sparsity will still be more or less required for better quality tho

Thanks for the response. It seems your workflow requires loken video tools. Comfy removed it from there download manager. Wonder why. There is also the problem with keeping the image consistent when using upscalers for long extended videos. I will try VSA with a single pass. Hopefully it works out. If not, there's always sulfurs upcoming model supposedly being released by the end of this week. Thanks anyway.

Also, I tried many upscalers in the past with the resolution being set to the same as my single pass for the final output. It always ran out of vram when going with 20 seconds and over. With a single pass, I can accomplish this. Unless yours doesn't use a latent upscaler, I don't see how it can use less vram. Or maybe its just a me problem and I'm messing up somewhere.

the upscaler itself is quite memory hungry, a memory spike there can happen on large latents.
Base H3 was designed with a 15s max. Never intended to go over that. It can work but its not the desired use case. At those lenghts you will probably be better served by splitting your scene in several shorter takes

Thx for the heads up on the node. Will have to check the manager registry site. I will check it up, but meanwhile you can download it from github directly

Just checked the Comfyui Manager Registry and my node pack is up and active. No removed whatsoever.

https://registry.comfy.org/es/nodes/lokken-videotools

It should be up and running. Both the web and the api reports the node as active and downloadable. Might be a problem with your Comfy Manager cache?

thanks very much for all your advice and the link. Unfortunately I won't using this model seeing as I find the audio to be just too poor in comparison to version 5. By the way, with VSA, the video quality was good for single pass so thanks for suggesting that.

Its too bad that this only works well with an upscaler workflow. Not much good for people who like to use workflows that extend on clips. I also find these upscaler workflows are just not good for people with low vram. Ill stick to beta 5 for now. Hopefully your next version won't require this in order to not have poor image quality.

I mean more like, it only works well and easily right now. There may be many or a simple sampling fix to get it working otherwise but I haven't looked into it. AI assistant suggested a latent normalizing or flattening node for example.

the upscaler itself is quite memory hungry, a memory spike there can happen on large latents.
Base H3 was designed with a 15s max. Never intended to go over that. It can work but its not the desired use case. At those lenghts you will probably be better served by splitting your scene in several shorter takes

Thx for the heads up on the node. Will have to check the manager registry site. I will check it up, but meanwhile you can download it from github directly

i find H3 does easily 18 to 20 seconds. I never use it for less. The real breakdown only seems to come at 22 or so. And beta6 also does ok there. Which is good, because 15secs is just too little. (32gb+96gb)

And yes, i'm aware that months ago 5seconds was the limit of models. Although i have wan2.2+SVI videos of 1 - 2 minutes . That was so easy to use. Also ltx2.3 which did 40seconds, with it's usual weaknesses.

I agree. I've been generating 20 second clips since day one for Minimax and never had a problem. 40 seconds for LTX as well before making the switch to H3.

Hey Andyx1976, I would love to know how you were able to create 1 to 2 minute clips with wan. I use SVI all the time but I can't go over 30 seconds or it becomes saturated. I'm still using the original SVI workflow for that. Has there been a breakthrough/fix for this that I am not aware of? How does your video look at the 1 to 2 minute mark compared to the beginning? If its no longer an issue, I don't suppose you could provide a link to the workflow your using to get around this? Appreciate any info or links you can provide.