doc, in kohya lora training do you know if there is a difference between using 240mb clip L vs 1.6gb
doc, in kohya lora training do you know if there is a difference between using 240mb clip L vs 1.6gb clip L. Is this something you have tested?
Fyes i tested
Fi feel like 1.6gb clip a bit better
![Dazzastrous [4090]](https://cdn.discordapp.com/avatars/871389071686135838/c4a5f4d81d9eeb18d856295d692e2c07.webp?size=40)
Dthe 1.6gb clip L is larger
D
Dhunyuan img to video is dissapointing ive heard
VYea, im just trying to figure out if that's something I can trade off on to cut training time. On 4090 tier_4 (lowest of the 4 configs) on 200 epochs im getting a 5.5hr ETA on training for 30 images for FLUX. Intermediary checkpoints are cooking well, but the time it wants to take is a bit of a bummer.
Dcloud services ?
Vno local
V2hrs felt normal for sdxl/sd15 training. Just getting into it on the flux side. Not sure if this is what the average is across the board.
Drent a more powerful gpu to cut the time
![Dazzastrous [4090]](https://cdn.discordapp.com/avatars/871389071686135838/c4a5f4d81d9eeb18d856295d692e2c07.webp?size=16)
Fwan 2.1 king
Fi am still waiting teacache
Fif arrives it will be 100% to 200% faster
Falso there is video improvement thing
Vthank you for telling me the obvious better solution, its good to be reminded of that. Lots of this is time intensive. But i am seeking to optimize where i can which is the basis for the original query.
Fi requested its to be added too
Fyou can reply and make a comment , can make authors to add 
Dthank god you put the 2.1 in there Doc lol
SWould love to hear about it if you find a custom GPT or LLM Instructions that reliably understand how Flux likes to be prompted. Often, It feels like they just shotgun a cloud of text at the Clips.
AI mean the answer lies on the dataset of both the T5 and Clip models . If we had access to those and train an LLM to give prompts based on it it would be dope
Fjoy caption really good
For cogvlm 2
Fwe have both
HCan anyone maybe tell me why I have this weird noisey image? It's all models and loras
SAnyone around her ehave some experience with training loras? I want to learn how to make them, but I'm not sure what is the most effective method of doing it and it seems like tehre are quite a few ways to do it. Anyone have a recommended way? I've been using comfyui pretty much exclusively for image gen and if quality is equally as good, I'd prefer comfy
Honetrainer > kohya
Sonetrainer, I'll check that out. It's more the quality of lora I'm concerned about than anything else
Sany idea if onetrainer is planning to support wan?
B@Dr. Furkan Gözükara Im also training a style lora to get very realistic images. I have seen some LoRAs like these on civit ai https://civitai.com/models/652699/amateur-photography-flux-dev I'm trying to train to get this style but been struggling to get the skin texture right. I tried with 150 images, 8000 steps, batch size-2, rank- 64. I have also added 10 face closeups to get some skin texture. I mostly sourced images from a friend who is a photographer and then from stock image websites. What am I doing wrong? Is my dataset too less or im still picking clean looking images and should do more images which are actually taken from phone?
BGenerations with my LoRA
B
Bgenerations with the amateur photography lora
B
A128 Lora rank for max quality
Fprobably inaccurate parameters
FYouTubeSECourses
Stability AI published their most power newest model Stable Diffusion 3.5 Large. This model unlike FLUX is full model not distilled and has huge potential. I have done extensive research and publishing all of it in this video regarding how to use SD 3.5 Large with the best settings. Moreover, I am sharing how to use FLUX DEV with the best possib...
Fi have step by step excellent tutorial with config
FYouTubeSECourses
Ultimate Kohya GUI FLUX LoRA training tutorial. This tutorial is product of non-stop 9 days research and training. I have trained over 73 FLUX LoRA models and analyzed all to prepare this tutorial video. The research still going on and hopefully the results will be significantly improved and latest configs and findings will be shared. Please wat...
Fnope not. for Wan training I plan kohya or ostris
Fprobably kohya
SNice. I just spun up some runpod with a simple gradio ui.
Fi have runpod tutorial too
FYouTubeSECourses
Unlock the power of FLUX LoRA training, even if you're short on GPUs or looking to boost speed and scale! This comprehensive guide takes you from novice to expert, showing you how to use Kohya GUI for creating top-notch FLUX LoRAs in the cloud. We'll cover everything: maximizing quality, optimizing speed, and finding the best deals. With our exc...
Fyour dataset
Falso you shouldnt increase batch size like that because batch size impact learning rate. you can use our lora config
Mhi @Dr. Furkan Gözükara could you tell me if this parameters of the SamplingFlux exists in SwarmUI? thnks
Fnot sure you can ask here