@Sipriyani you try sana for text?
@Sipriyani you try sana for text?
S
Ssure I would
Zlike how good is it for making text?
Zcan it do more than a sentance?
Fi think yes
Vhttps://www.youtube.com/watch?v=FvpWy1x5etM
I thought initially I might try to do the tutorial on google cloud because I still have free credits there.... but as I dont have much experience with running any models in cloud or locally (only have 2gb vram), I think I will better stick to runpod or massed compute. Which one would you recommend more, runpod or massed compute? What is the pricing difference? Or are these questions covered somewhere?
I thought initially I might try to do the tutorial on google cloud because I still have free credits there.... but as I dont have much experience with running any models in cloud or locally (only have 2gb vram), I think I will better stick to runpod or massed compute. Which one would you recommend more, runpod or massed compute? What is the pricing difference? Or are these questions covered somewhere?
YouTubeSECourses
If you want to train FLUX with maximum possible quality, this is the tutorial looking for. In this comprehensive tutorial, you will learn how to install Kohya GUI and use it to fully Fine-Tune / DreamBooth FLUX model. After that how to use SwarmUI to compare generated checkpoints / models and find the very best one to generate most amazing image...
Fmassed compute
Fbest one
Ftesting ohwx man vs ohwx, man
Fstarted 6 more training lets see i hope i can achieve preventing bleed
Vthx!!!
Fif you cant get single A6000 there due to all being used, get 2x A6000 ALT - has more ram - use our coupon SECourses
Fbut massed compute team working to add more machines
Vwhat does it mean?
NHey, everybody, how's it going?
What's the latest super interesting thing that's happened in the month I've been in the bunker?
What's the latest super interesting thing that's happened in the month I've been in the bunker?
DFlux tools 
DAnd they work!!
Sif i'm training a flux dreambooth model in kohya, and my computer crashes in the middle of it, is there a way to continue from last saved checkpoint?
Pu need to check "Save training state" in settings.
Pthis way pyu can continue from the saved state(last epich u saved)
Ninteresting to see
Swhere is this setting located? for dreambooth checkpoint training? i have saved the model every 20 epochs
CSANA is fine, nice, fast on my 4090, but most problematic are eyes
Phello! i just launched training with settings fyoum your patreon for 24gb Flux dev. I have questions regarding Repeats and epocs:
Is 2 epoch 1 reeat = 1 epoch 2 repeats or not? its just checkpoints are huge and take lots of time while saving. I was wondering if i use 20 repeats instead of 1 - will it affect training or be the same as training 20 epocs with 1 repeat? THANKS!
Is 2 epoch 1 reeat = 1 epoch 2 repeats or not? its just checkpoints are huge and take lots of time while saving. I was wondering if i use 20 repeats instead of 1 - will it affect training or be the same as training 20 epocs with 1 repeat? THANKS!
TGuys can we do a grid generation in swarmui with comfyui workflow? I would like to test with detail daemon
TAny update?
Aanyone know what is the best video faceswap / video deepfake software ?
A@Furkan Gözükara SECourses Rank 16 = 16, node neural network and Rank 128 = 128 node neural network?
AMeaning more nodes = more data flow = better understanding of the subject?
ASo Rank would be equal to the size of the neural network itself?
AHey, I have question regarding this: https://www.youtube.com/watch?v=FvpWy1x5etM I would like to master topic as you (or at least be able to reproduce results with my face or the set of cartoon characters I'm working on). I have question, what tier do I need in Patreon if any and am I be able to do all of that on my local computer which has RTX 3090 24GB and 128GB of RAM? Or this is this tutorial too much bound to RunPod and other such paid cloud solutions.
YouTubeSECourses
If you want to train FLUX with maximum possible quality, this is the tutorial looking for. In this comprehensive tutorial, you will learn how to install Kohya GUI and use it to fully Fine-Tune / DreamBooth FLUX model. After that how to use SwarmUI to compare generated checkpoints / models and find the very best one to generate most amazing image...
DYou can do it at home. Your machine is very good.
AYou can start with a Tier 1 , doctor has an amazing library of scripts and configs for litteraly everything
AYou're saving time and effort basically and yes you can do it locally
DIt's really good. I tried today, one shot inpainting & outpainting both in comfy and swarm.
DIt also takes care of problem areas like hands, etc. real quick.
DRope Live comes highly recommended, though I haven't been able to produce as good faceswaps with it as other people. There are multiple settings, and limited 'in-depth' tutorials out there. Most will just cover the basics.
I think there are better paid apps for this kinda thing, since people produce really good faceswaps, but I haven't been able to pin point one particular leader amongst the paid ones.
Perhaps some other people on the group can recommend.
I think there are better paid apps for this kinda thing, since people produce really good faceswaps, but I haven't been able to pin point one particular leader amongst the paid ones.
Perhaps some other people on the group can recommend.
LYou can't fine tune dreambooth flux with rtx 4090?
DWhy would you think so? Ofc, you can
BMy training set for Dreambooth had no captions and Im finding all loras destroying my likeness, would captions have helped this?
Padvanced settings
Pthe best would be Deepface Lab. if you got tons of photos of a person and willing to spend few days in training. Secon best is FaceFusin/roop/rope/reactor takes 1 photo and lightning fast but low quality 128x128 only. DeepfaceLab works up to 512x512 for the face
DIs there a bug or something with Kohya? I start the training with 24GB Tier3, everything is loaded, memory looks fine, gpu utilisation is fine, yet I'm getting horrible speeds
D
AHoly 
FOne day later...
F:bceTired3Dead4:
Fsomeone proposed that this way captioning may work better
Fcurrently training
F