ComfyUI Node: FL Kohya Adv Config
Category
🏵️Fill Nodes/Training
Inputs
xformers
- enable
- disable
sdpa
- enable
- disable
fp8_base
- enable
- disable
mixed_precision
- no
- fp16
- bf16
gradient_accumulation_steps INT
gradient_checkpointing
- enable
- disable
cache_latents
- enable
- disable
cache_latents_to_disk
- enable
- disable
network_dim INT
network_alpha INT
network_module
- networks.lora
- networks.dylora
- networks.oft
network_train_unet_only
- enable
- disable
lr_scheduler
- linear
- cosine
- cosine_with_restarts
- polynomial
- constant
- constant_with_warmup
- adafactor
lr_scheduler_num_cycles INT
optimizer_type
- AdamW
- AdamW8bit
- PagedAdamW
- PagedAdamW8bit
- PagedAdamW32bit
- Lion8bit
- PagedLion8bit
- Lion
- SGDNesterov
- SGDNesterov8bit
- DAdaptation
- DAdaptAdaGrad
- DAdaptAdam
- DAdaptAdan
- DAdaptAdanIP
- DAdaptLion
- DAdaptSGD
- AdaFactor
lr_warmup_steps INT
unet_lr STRING
text_encoder_lr STRING
shuffle_caption
- enable
- disable
save_precision
- float
- fp16
- bf16
persistent_data_loader_workers
- enable
- disable
no_metadata
- enable
- disable
noise_offset FLOAT
no_half_vae
- enable
- disable
lowram
- enable
- disable
Outputs
FL_TT_SS_AdvConfig
Extension: ComfyUI_FL-Trainer
Train Image Loras on both sd1.5 and SDXL. This repo git clones the pieces needed to train. It pops open a second terminal window do do the training. It will also display the inference samples in the node itself so you can track the results.
Authored by filliptm
Run ComfyUI workflows in the Cloud!
No downloads or installs are required. Pay only for active GPU usage, not idle time. No complex setups and dependency issues
Learn more