Caching will be automatically before trening process or we will need do do separate?
Caching will be automatically before trening process or we will need do do separate?
JAwesome
JI am impressed 
JWill you make it generic or easily adaptable so you can reuse it for Wan 2.2 training?
Fgood question i need to ensure it happens auto
Fwhen selected
Fyes i plan to add wan 2.2 as well
Fafter qwen done
CHow is going ?? WHen you plan publish ?? I asking as i am about start new big dataset trainig and not sure if wait for GUI or ... ???
S@cyberbol what GPU did you use for Wan2.2 training?
C4090 (24 GB VRAM) – it was fine, but I’m now collecting a better dataset for my final LoRA for WAN 2.2.
For example, right now when I train WAN 2.2 with high noise and low noise… the low noise results are actually better when I use my old LoRA from WAN 2.1.
So currently I’m using WAN 2.2 high noise for the high noise pass, and my old WAN 2.1 LoRA for the low noise pass.
I’m not sure why… Most of the issues seem to be with human eyes.
I think what I need to do is:
for low noise → collect a dataset with high resolution, close-ups on skin details, eyes, etc.
for high noise → use more general shots of the body, face, etc.
For example, right now when I train WAN 2.2 with high noise and low noise… the low noise results are actually better when I use my old LoRA from WAN 2.1.
So currently I’m using WAN 2.2 high noise for the high noise pass, and my old WAN 2.1 LoRA for the low noise pass.
I’m not sure why… Most of the issues seem to be with human eyes.
I think what I need to do is:
for low noise → collect a dataset with high resolution, close-ups on skin details, eyes, etc.
for high noise → use more general shots of the body, face, etc.
CKey Training Parameters for WAN2.2 (for RTX 4090)
For the High-Noise model:
Optimizer: adamw8bit (a key optimization for 24GB VRAM)
Precision: fp16 and fp8_base
Learning Rate: 3e-4
LR Scheduler: polynomial with power=8
Network Dim / Alpha (LoRA Size): 16 / 16
Gradient Checkpointing: Enabled
Timestep Range:
min_timestep: 875
max_timestep: 1000
For the Low-Noise model:
All parameters are identical to the High-Noise model, except for the timestep range.
Optimizer: adamw8bit
Precision: fp16 and fp8_base
Learning Rate: 3e-4 or other 2e-4
LR Scheduler: polynomial with power=8
Network Dim / Alpha (LoRA Size): 16 / 16
Gradient Checkpointing: Enabled
Timestep Range:
min_timestep: 0
max_timestep: 875
For the High-Noise model:
Optimizer: adamw8bit (a key optimization for 24GB VRAM)
Precision: fp16 and fp8_base
Learning Rate: 3e-4
LR Scheduler: polynomial with power=8
Network Dim / Alpha (LoRA Size): 16 / 16
Gradient Checkpointing: Enabled
Timestep Range:
min_timestep: 875
max_timestep: 1000
For the Low-Noise model:
All parameters are identical to the High-Noise model, except for the timestep range.
Optimizer: adamw8bit
Precision: fp16 and fp8_base
Learning Rate: 3e-4 or other 2e-4
LR Scheduler: polynomial with power=8
Network Dim / Alpha (LoRA Size): 16 / 16
Gradient Checkpointing: Enabled
Timestep Range:
min_timestep: 0
max_timestep: 875
Fmy research would take at best several days
Cuntil publish gui ?? ohhhh..
Fyep
Fi need to train and test
Cok, so i will use just command by now
CI'm looking for the latest info and best practices for training WAN 2.2 LoRAs, especially for character likeness on a 24GB card.
From what I've gathered so far, it seems the best dataset approach is to train the high-noise model on low-resolution videos (to learn motion) and the low-noise model on high-resolution images (to learn the fine details).
I also saw in the new Musubi-Tuner docs that you can apparently train both high and low noise LoRAs at the same time now, using the --dit (for low) and --dit_high_noise (for high) parameters in a single command. For 24GB VRAM, the key seems to be adding the --lazy_loading option to only keep the active model on the GPU.
Has anyone tried this simultaneous training method yet?
From what I've gathered so far, it seems the best dataset approach is to train the high-noise model on low-resolution videos (to learn motion) and the low-noise model on high-resolution images (to learn the fine details).
I also saw in the new Musubi-Tuner docs that you can apparently train both high and low noise LoRAs at the same time now, using the --dit (for low) and --dit_high_noise (for high) parameters in a single command. For 24GB VRAM, the key seems to be adding the --lazy_loading option to only keep the active model on the GPU.
Has anyone tried this simultaneous training method yet?
JNo need for the lazy-loading as you can just train the two lora's after each other (in command line it auto continues to the 2nd command)
CI know. But y can also Train both same time
JIt probably wont be faster, and you could decide to just turn off your pc after the 1st lora training if you decide that
CProbably you right. I just experiment..we will know much better in few weeks probably..
J+ I may want to use less epochs or learning rate for the other model
Ji noticed the high noise model gets overtrained quicker
C@JonkoXL Do you successfully Train any high and low noise LORA and it's working together with OTHER lora keeps human eyes good quality?
JI think eyes in Wan 2.2. will still be a problem unless we get an even higher base resolution
JJThis is my only Lora I succesfully trained so far
Jhaven't tried it with other loras
CI Train three and when using with others LORA human eyes always problem.
Wan2 1 was always fine
Wan2 1 was always fine
Jhave you tried switching your high and low lora?
Jfor some reason i sometimes do that and apparently it can work out
COo. No..but will try when back home
JI played a bit with Wan 2.1 Lora training too, it seemed easier or more forgiving. But only tested it
C
JOOh nice :p
Jwhat's your learning_rate and how many steps did you train for?
CWan2.1? I don't remember..but I Keep on drive all dataset and configs.
Sthanks - right now im doing Wan2.2 training on my dataset ll see how it turns out
CLet me know please. Share also what datasets you used if you can. Just photo's or mixed or video clips?
Sdue to nda i cant share dataset but these are studio photos and few natural shots of a person. Unfortunetly only photos for now
Mhello, i got out of memory error using massed-compute 2x RTX 6000 ADA
Jtraining what?
M40 images of a human for Flux
SNo LoRA vs LoRA Wan2.2
Sshe is little bit damaged tho its 40 epoch [65img's dataset] 32DIM
Jlearning rate?
Fdid you load lora config into dreambooth or dreambooth into lora?
Fconfig name?
MNo, I loaded dreambooth config into Dreambooth. I tried both 48GB_GPU_28200MB_6.3_second_it_Tier_1 and Batch_Size_7_48GB_GPU_46250MB_29.1_second_it_Tier_1. out of memory errors for both
Fwhat is your images resolution?