Hey, did I read correctly that with the new Flux Kohya training I can use the 48Gb preset with a 50
Hey, did I read correctly that with the new Flux Kohya training I can use the 48Gb preset with a 5090 ok?
JWAN 2.1 14B Image-to-Video 480P FP8 working fine now. Thanks!!
Fi hope results accurate
Flet me know
Fit fits into vram :d
1I thought that was case . Just testing a fresh install. It started ok but seems to have froze and not calculating it/s
Fit works. make sure that you vram usage is below than 2 gb
Fit uses like 28 gb
Falso these are when all images 1024x1024
Fif you have different res it impacts
1images are fine, Will check to free up memory. Just noticed I wasn't using onboard GPU for monitor.
1crippled it
Fye using too much vram
Fyou are using 48GB_GPU_28200MB_6.3_second_it_Tier_1.json right?
148GB_GPU_Quality_Tier_1_40000MB_9.3_Second_IT_Trains_T5_and_T5_Attention for Lora
1sorry, it's late. it's 40Gb needed for that.
Fit literally says 40 gb 
1I'll drop it to the Tier_3
1long day.
Jsomething is wrong with img to video 14b FP8. After 2-3 frames goes full pink image
J39 minutes to generate it with teacache and FP8
JDoing a second video. I'll post the result here
JWAN 2.1 14B Image-to-Video 480P FP8 is broken
JAnd the negative prompt box is always empty when you start the app
Jshould have the default or saved profile prompts, like previous versions
T@Dr. Furkan Gözükara I just come back and thank you for:
1. the Dreambooth training for 5090 is working ! (In windows)
2. the MAssed Cloud got 300G version for me to train my models !! U are a star ! @Dr. Furkan Gözükara
1. the Dreambooth training for 5090 is working ! (In windows)
2. the MAssed Cloud got 300G version for me to train my models !! U are a star ! @Dr. Furkan Gözükara
Jhey brothers, is there a way to make social banners without letters/fonts errors? with stable diffusion?
JI think the latest DiffSynth-Studio fix for FP8 on WAN 2.1 14B Image-to-Video 480P has broken the FP16 quality too. I'm testing with previous commit before the FP8 fix
JConfirmed
J@Dr. Furkan Gözükara Please revert (or fix) the latest DiffSynth update. It breaks colours and quality in FP16, and in FP8 the video turns completely pink. I think it only affects WAN 2.1 14B Image-to-Video 480P model
P@JaviEM i do noit have that iissue .. strange..
I do have a issue so running video to video
RuntimeError: The size of tensor a (36) must match the size of tensor b (13) at non-singleton dimension 2
I do have a issue so running video to video
RuntimeError: The size of tensor a (36) must match the size of tensor b (13) at non-singleton dimension 2
JI reverted to previous DiffSynth commit and it's working fine again, so it's a problem from the latest commit
POk that is weird .. for me it seems to work fine on rtx 3090
Fthe offical repo authors made some fixes
Fi will move to them back
Fi hope it fixes
F2 awesome
Fwill fix this too
Fthis looks like frame count of video to video
Fi am testing newest DiffSynth i hope they added lora properly as we did
B@Dr. Furkan Gözükara can you tell me the commands need to start the swarmui in the runpod bcz i installed and runned it last time with one command and now i have installed i want to run that only so i can use gui to do some generations
Fi have updated commands here https://www.patreon.com/posts/tutorial-instructions-links-public-post-106135985
Patreon
Get more from SECourses: FLUX, Tutorials, Guides, Resources, Training, Scripts on Patreon
Fand tutorial here
FYouTubeSECourses
In this video, I demonstrate how to install and use #SwarmUI on cloud services. If you lack a powerful GPU or wish to harness more GPU power, this video is essential. You'll learn how to install and utilize SwarmUI, one of the most powerful Generative AI interfaces, on Massed Compute, RunPod, and Kaggle (which offers free dual T4 GPU access for ...
Bthanks a lot
Fi have to be afk few hours after that hopefully will fix wan 2.1 errors
S@Dr. Furkan Gözükara Would you recommend to always use adetailer for faces (even if it’s a closer face)? Using forge.
And do you know if adetailer would possibly lower likeness of a character lora? Or if it's supposed to increase it
And do you know if adetailer would possibly lower likeness of a character lora? Or if it's supposed to increase it
Phey.. i used forge before as well and switched to SwarmUi and can highly recommend it. I use the facedetaile yolo there and it works amazing. There is no way to achieve that detail out of the box in my opinion, especially witzh multiple character images 
SWill have a look into that! Thanks a lot
