File size: 2,022 Bytes
0f9a690 7b29017 20dd25c 0f9a690 d677e94 0f9a690 fbb8b7d 0f9a690 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 |
---
license: apache-2.0
datasets:
- nbeerbower/FIXBODY
base_model:
- Qwen/Qwen-Image-Edit
tags:
- lora
---
# FIXBODY-r64
An imperfect LoRA for Qwen-Image-Edit focused on correcting hands and anatomy issues in anime-style illustrations while preserving the original art style.
I am working on an improved version of this LoRA!
## Usage
```
pipeline.load_lora_weights(
"nbeerbower/FIXBODY-r64-Qwen-Image-Edit",
weight_name="pytorch_lora_weights.safetensors"
)
```
Triggered by `fix [her/his/their] hands/legs` (the pronoun can be omitted, and it works best focusing on one body part at a time)
## Model Configuration
[Training code available on Github!](https://github.com/nbeerbower/Qwen-Image-Edit-LoRA-Trainer)
```
# Configuration for FIXBODY - 86 high-quality pairs
# Model settings
pretrained_model_name_or_path: "Qwen/Qwen-Image-Edit"
# Dataset settings
dataset_name: "nbeerbower/FIXBODY"
dataset_split: "train"
max_samples: null # Use all 86
# Prompt settings
use_prompt_directly: true
prompt_template: "{prompt}"
# LoRA configuration
lora_rank: 64
lora_alpha: 128
lora_dropout: 0.01
target_modules: ["to_k", "to_q", "to_v", "to_out.0", "ff.net.0.proj", "ff.net.2"]
# Training settings
train_batch_size: 1
gradient_accumulation_steps: 2
num_train_epochs: 50
max_train_steps: 3000
learning_rate: 1e-4
lr_scheduler: "cosine"
lr_warmup_steps: 50
# Optimizer settings
use_8bit_adam: true
adam_beta1: 0.9
adam_beta2: 0.999
adam_weight_decay: 0.001
adam_epsilon: 1e-8
max_grad_norm: 1.0
# Mixed precision
mixed_precision: "bf16"
# Checkpointing and logging
output_dir: "./output_fixbody_full2"
logging_dir: "logs"
checkpointing_steps: 100
checkpoints_total_limit: 15
report_to: "wandb"
tracker_project_name: "qwen-fixbody"
run_name: "r64-1e4-50ep-0.01ld"
# Data processing
save_embeddings: true
dataloader_num_workers: 1
# Memory optimization
gradient_checkpointing: true
``` |