A GGUF version of 1.4v 10eros checkpoint original model made for myself to check if i was able to run it locally on my 5060ti 16gb and runs great on 32gb ram ddr5 6000mhz.- Intended for I2V - works best.
I simply share it so everyone else can run it too.. ALL CREDIT GOES TO TENSTRIP . Please give him a thumbs up if this one works for you.
I Do not own any right sin this model and do not provide any support to it.
Workflow used is the DMD v5 of TENSTRIP ONE: Check his main model for guides. I only provide a modified FaceID workflow of HIS WORKFLOW for this GGUF at high quality 0 compression Image. Overall time 20 minutes Upscale included 0 Image compression. for 10seconds high quality output. 2-3 minutes previews VAE first pass.
Tested on 25 seconds clips and works just fine.
Description
FAQ
Comments (12)
Can you share your workflow? I can never get none-distilled LTX models to get outputs
Its in the sidebar the adjusted GGUF workflow, same as 10eros workflow with faceid. takes TIME but makes good quality first VAE in like 4 minutes to see if its going OK. before upscale. Make sure to read original tenstrip MODEL 10eros for his DMD required LORA. and 0.8 -1.0.
If only LTX 2.3 could run on 8 GB of Vram...Sigh a man can dream...A man can dream.
It can work at 8Gb vram fine...but need a lot RAM like >48GB
@flo11ok874 Nope - I'm running a 3060ti & 32gb. With an INT8 checkpoint (CP), I'm getting preview gens in 90seconds and upscaled versions in a further 3mins (there's a good workflow on this site that delivers 4 previews which you can select from). I've just started using a recent INT4 CP and i2v preview gens are 45s, with no obvious quality loss in upscaled versions.
I can run it with a GTX 1060 6GB. ltx-2.3-22b-dev-Q5_K_M
you should make sure your cuda version and pytorch versions are correct for your card, too.
Does anyone know that ltx, sulphur or eros can run with 14.7 gbs of Vram?
I see that this comments section is very active, I know that it can be done but I don't know how to do it, if someone can offer me help with the model name, simple instructions and workflow, I will be very grateful, I don't ask for much, I just want to make videos without scaling.
@dupe_900 You can do offload ram if you have 32gb ram or more. Always can try that, sageattention enabled too is a great resource for that too. Takes more time but can be done. I've used a lot this gguf q6k and gives quite well results as longs its not weird fast phase motions. overall all works great.
@eddmoe I have 15 Vram and 12.7 RAM, I don't have 32 :(
I don't think it will work, I would only need 1.55gb of Vram to run this ckpt.
What can I do?
The skydiver in the first video is quite a character: he’s got four balls, with a tiny vagina right between them :-)
Yes, was a quite bad prompted one, also the lora DR34M or something was making it more complicated for a test one