Software Engineering Courses (SECourses)•5mo ago

Training wan2.2 High-Noise in musubi-tuner in progress. Next Low-Noise and i will see result of m

Training wan2.2 High-Noise in musubi-tuner in progress.

Next Low-Noise and i will see result of my first wan2.2 lora.

Ccyberbol use mouse copy and paste, not CTRL+C etc... sometimes not work at forst times.

Ella•8/7/25, 10:13 PM

thank you

Ccyberbol Training wan2.2 High-Noise in musubi-tuner in progress. Next Low-Noise and i ...

JonkoXL•8/8/25, 2:01 PM

t2v or i2v?

JonkoXL•8/8/25, 2:01 PM

t2v works for me

JJonkoXL t2v or i2v?

cyberbolOP•8/8/25, 2:26 PM

@Ella t2v. But I used just 30 photos.

--task: t2v-A14B
- --dataset_config: dataset/dataset.toml
- --mixed_precision: fp16
- --fp8_base
- --xformers
- --optimizer_type: adamw8bit
- --gradient_checkpointing
- --learning_rate: 3e-4
- --max_train_epochs: 100
- --gradient_accumulation_steps: 1
- --seed: 5
- --optimizer_args: weight_decay=0.1
- --network_module: networks.lora_wan
- --network_dim: 16
- --network_alpha: 16
- --lr_scheduler: polynomial
- --lr_scheduler_power: 8
- --lr_scheduler_min_lr_ratio: "5e-5"
- --output_dir: output
- --save_every_n_epochs: 10
- --timestep_sampling: shift
- --discrete_flow_shift: 1.0
  For High-Noise Training:
- --min_timestep: 875
- --max_timestep: 1000
  For Low-Noise Training:
- --min_timestep: 0
- --max_timestep: 875

cyberbolOP•8/8/25, 2:27 PM

Today will try in diffusion pipe.

JonkoXL•8/8/25, 2:40 PM

probably the epochs and learning rate can be tweaked for faster results, ill try it out

JonkoXL•8/8/25, 2:40 PM

can you post some results after training?

�

🍭🎀 𝒜𝓋𝒶 𝐹𝓇𝒾𝑔𝑔 🎀🍭•8/8/25, 7:34 PM

I trained lora on a real person and still get plastic skin and plastic face in comfyui....what am I doing wrong here

Sskyrrr Hey does anyone tried a full finetune with Flux Krea ? Any samples to share/insi...

Furkan Gözükara SECourses•8/9/25, 10:55 AM

it works with our configs

Furkan Gözükara SECourses•8/9/25, 10:55 AM

people tested but i didnt have chance yet on vacation

FFurkan Gözükara SECourses people tested but i didnt have chance yet on vacation

skyrrr•8/11/25, 12:36 PM

Well deserved ! Enjoy your holidays

JonkoXL•8/11/25, 4:13 PM

@Furkan Gözükara SECourses
For Musubi Tuner Wan 2.2 training, you can add the following arguments to the accelerate command, to make it generate samples

--sample_prompts prompt.txt --sample_every_n_epochs 1 --sample_at_first

In the promp.txt file you can add the positive prompt, and also some arguments (within that textfile!)

A blonde girl dancing --w 640 --h 480 --f 1 --d 1 --s 20 --n blurry

JonkoXL•8/12/25, 8:33 PM

Anyone else trainign Wan 2.2. and somehow realizing the plugging the Lora for the low model into the high model and the Lora for the high model into the low model, giving you better results?

JonkoXL•8/12/25, 8:34 PM

I checked the MD5 hash with the Hugging Face page, it's strange..

JJonkoXL Anyone else trainign Wan 2.2. and somehow realizing the plugging the Lora for th...

Furkan Gözükara SECourses•8/12/25, 9:57 PM

wow weird

FFurkan Gözükara SECourses wow weird

JonkoXL•8/12/25, 9:59 PM

Yes, but managed to get my Cartoon style Lora finished today

JonkoXL•8/12/25, 9:59 PM

https://civitai.com/models/1862320/wan22-t2v-lora-cartoon-style

(Wan2.2 T2V Lora) - Cartoon Style - Wan 2.2 T2V LORA - Low | WanVid...

(Wan2.2 T2V Lora) - Cartoon Style Create Cartoon Style images and videos. (see examples) Use "r3tr0c0m1c" and "cartoon" as a prompt keyword. Resolu...

JonkoXL•8/12/25, 10:06 PM

JonkoXL•8/12/25, 10:21 PM

JonkoXL•8/12/25, 10:23 PM

JJonkoXL https://civitai.com/models/1862320/wan22-t2v-lora-cartoon-style

Furkan Gözükara SECourses•8/12/25, 10:28 PM

nice

FFurkan Gözükara SECourses nice

JonkoXL•8/12/25, 10:35 PM

Thanks!
Wan 2.2 is very cool, my favorite model to play with.
I just wish they would launch a controlnet version for 2.2 already.

JonkoXL•8/12/25, 10:39 PM

My results so far is 3000 steps for 25 images at 2e4 learning rate, on dim 32 alpha 24, trained on 720x720 resolution.
Just a bit less than 2 hours training per model on my RTX 5090 without any offloading.
Lora strength of 1 works for 960x720 and 720x480. Needs a higher strength for higher res or if prompt gets too long.

JJonkoXL My results so far is 3000 steps for 25 images at 2e4 learning rate, on dim 32 al...

Furkan Gözükara SECourses•8/12/25, 10:42 PM

nice info thanks

Newtus•8/13/25, 5:39 AM

I was interested in creating a Chorma lora using Kohya SD scripts. I'm thinking the issues I'm having are just my ignorance of command line. Does anyone know of a youtube or idiots guid for Kohya SD-SCripts? It's probably too much to ask for Chroma training, but if there was a Flux I think I can learn on that and eventually wing it. Thanks.

NNewtus I was interested in creating a Chorma lora using Kohya SD scripts. I'm thinking ...

JonkoXL•8/13/25, 6:51 AM

what does your command look like?

NNewtus I was interested in creating a Chorma lora using Kohya SD scripts. I'm thinking ...

Furkan Gözükara SECourses•8/13/25, 7:17 AM

this is literally gui for kohya sd scripts

Furkan Gözükara SECourses•8/13/25, 7:17 AM

https://youtu.be/FvpWy1x5etM

YouTubeSECourses

FLUX Full Fine-Tuning / DreamBooth Training Master Tutorial for Win...

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...

FFurkan Gözükara SECourses this is literally gui for kohya sd scripts

AiInfluence•8/13/25, 7:17 AM

im getting very bad results

AiInfluence•8/13/25, 7:18 AM

on wan 2.2 preset not working

JJonkoXL what does your command look like?

Newtus•8/13/25, 7:21 AM

Thank you in advanced. I'm almost ashamed to post it for fear I forgot something simple. I put the read me into chatgpt to try and help me create the command line and the dataset.toml file: accelerate launch --mixed_precision bf16 --num_cpu_threads_per_process 1 flux_train_network.py ^
--pretrained_model_name_or_path "D:/APPS/KohyaSD/TestModels/Chroma.safetensors" ^
--clip_l "D:/APPS/KohyaSD/TestsModels/clip_l.safetensors" ^
--t5xxl "D:/APPS/KohyaSD/TestModels/model-00001-of-00002.safetensors" ^
--ae "D:/APPS/KohyaSD/TestModels/diffusion_pytorch_model.safetensors" ^
--model_type chroma ^
--apply_t5_attn_mask ^
--cache_latents_to_disk ^
--save_model_as safetensors ^
--sdpa ^
--persistent_data_loader_workers ^
--max_data_loader_n_workers 2 ^
--seed 42 ^
--gradient_checkpointing ^
--mixed_precision bf16 ^
--save_precision bf16 ^
--network_module networks.lora_flux ^
--network_dim 4 ^
--network_train_unet_only ^
--optimizer_type adamw8bit ^
--learning_rate 1e-4 ^
--cache_text_encoder_outputs ^
--cache_text_encoder_outputs_to_disk ^
--highvram ^
--max_train_epochs 4 ^
--save_every_n_epochs 1 ^
--dataset_config "D:/APPS/KohyaSD/TestCFG/C4TS_dataset_multi_bs.toml" ^
--output_dir "D:/APPS/KohyaSD/TestOutput" ^
--output_name chroma-lora-name ^
--timestep_sampling shift ^
--discrete_flow_shift 3.1582 ^
--model_prediction_type raw ^
--guidance_scale 1.0

JonkoXL•8/13/25, 7:23 AM

Whats the error you get?

JonkoXL•8/13/25, 7:25 AM

I think the --t5xxl needs to be a single safetensor file or pth file, and not a partial file (00001 of 00002)

JonkoXL•8/13/25, 7:25 AM

Also discrete_flow_shift maybe rounded number?

JonkoXL•8/13/25, 7:26 AM

did you manually edit the datset.toml? how does it look?

AAiInfluence on wan 2.2 preset not working

Furkan Gözükara SECourses•8/13/25, 7:29 AM

which preset exactly you using?

Newtus•8/13/25, 7:34 AM

Yes, I edited the toml file. Also I am working off a 4090. It looks like it attempts to run. I get the error: File "D:\APPS\KohyaSD\sd-scripts-sd3\venv\lib\site-packages\torch\nn\modules\module.py", line 1362, in convert
raise NotImplementedError(
NotImplementedError: Cannot copy out of meta tensor; no data! Please use torch.nn.Module.to_empty() instead of torch.nn.Module.to() when moving module from meta to a different device.
Traceback (most recent call last):
File "D:\PROG\Python310\lib\runpy.py", line 196, in _run_module_as_main
return _run_code(code, main_globals, None, and Toml: [general]

Common settings for the dataset

flip_aug = true
color_aug = false
keep_tokens_separator = "|||"
caption_extension = ".txt"

[[datasets]]

This is the dataset definition for training

batch_size = 2
enable_bucket = true
resolution = [1024, 1024]

[[datasets.subsets]]
image_dir = "D:/APPS/KohyaSD/TestIMG/10_C4TS cats"
num_repeats = 1 A suggestion from chatgpt was to move to bf16. (on the two lines in the bat file) I changes that is and looks like it attempts to load the models. But then another set of errors. It's like moving from one set of issues to another and I'm afraid I'm just working in the wrong direction.

NNewtus Yes, I edited the toml file. Also I am working off a 4090. It looks like it atte...

Furkan Gözükara SECourses•8/13/25, 7:43 AM

you are trying to train flux?

Newtus•8/13/25, 7:48 AM

No, I would ultimately like to train Chroma.

NNewtus No, I would ultimately like to train Chroma.

Furkan Gözükara SECourses•8/13/25, 7:50 AM

you can use gui

Furkan Gözükara SECourses•8/13/25, 7:50 AM

and add some extra params

Furkan Gözükara SECourses•8/13/25, 7:50 AM

https://github.com/kohya-ss/sd-scripts/pull/2157

GitHub

Feature Chroma support by kohya-ss · Pull Request #2157 · kohya-s...

Thank you lodestones for the high quality model.
--model_type option is added to flux_train_network.py. For Chroma, --model_type chroma is required.
According to lodestones, --apply_t5_attn_mask --...

FFurkan Gözükara SECourses you can use gui

Newtus•8/13/25, 7:51 AM

Thank you I will review that.

Sakami•8/13/25, 7:51 AM

Hey Doc @Furkan Gözükara SECourses
Are there any plans for Fine-Tuning or LoRA for Wan2.2? If not in the near future, where would you recommend looking to learn more about that field?