|
--- |
|
license: other |
|
license_name: flux-1-dev-non-commercial-license |
|
license_link: https://huggingface.co/black-forest-labs/FLUX.1-dev/blob/main/LICENSE.md |
|
--- |
|
# Modern Anime LoRA Adapter for FLUX.1 dev |
|
|
|
 |
|
|
|
## Usage |
|
- ComfyUI |
|
1. Download [this model](modern_anime_full.safetensors) . |
|
2. Move the file to ComfyUI/models/unet. |
|
3. Lunch ComfyUI. |
|
4. Load [the workflow](anime-full-workflow.json). |
|
5. Queue prompt. (trigger words: modern anime style,) |
|
6. Get the following image. |
|
 |
|
|
|
# How to full finetuning |
|
I used sd3 branch on sd-scripts. |
|
I ran the following command: |
|
```bash |
|
accelerate launch --num_cpu_threads_per_process 1 flux_train.py --pretrained_model_name_or_path 'flux/flux1-dev.safetensors' --clip_l 'flux/clip_l.safetensors' --t5xxl 'flux/t5xxl_fp16.safetensors' --ae 'flux/ae.safetensors' --cache_latents --save_model_as safetensors --sdpa --persistent_data_loader_workers --max_data_loader_n_workers 2 --seed 42 --gradient_checkpointing --save_precision bf16 --learning_rate 5e-6 --optimizer_type adafactor --optimizer_args "relative_step=False" "scale_parameter=False" "warmup_init=False" --cache_text_encoder_outputs --max_train_epochs 10 --save_every_n_epochs 5 --dataset_config modern_anme_full.toml --output_dir flux --output_name modern_anime_full --timestep_sampling sigmoid --model_prediction_type raw --guidance_scale 1.0 --full_bf16 --max_bucket_reso 2048 --min_bucket_reso 512 --apply_t5_attn_mask --mixed_precision bf16 --cache_text_encoder_outputs_to_disk --cache_latents_to_disk --blockwise_fused_optimizer |
|
``` |
|
|
