mirror of
https://github.com/modelscope/DiffSynth-Studio.git
synced 2026-03-19 06:48:12 +00:00
update docs
This commit is contained in:
Artiprocher
@@ -1,87 +0,0 @@
|
||||
|
||||
# 基于Flux的文生图示例
|
||||
|
||||
以下是如何使用FLUX.1模型进行文生图任务的示例。该脚本提供了一个简单的设置，用于从文本描述生成图像。包括下载必要的模型、配置pipeline，以及在启用和禁用 classifier-free guidance 的情况下生成图像。
|
||||
|
||||
其他 DiffSynth 支持的模型详见 [模型.md](模型.md)
|
||||
|
||||
## 准备
|
||||
|
||||
首先，确保已下载并配置了必要的模型：
|
||||
|
||||
```python
|
||||
import torch
|
||||
from diffsynth import ModelManager, FluxImagePipeline, download_models
|
||||
|
||||
# Download the FLUX.1-dev model files
|
||||
download_models(["FLUX.1-dev"])
|
||||
```
|
||||
|
||||
下载模型的用法详见 [下载模型.md](下载模型.md)
|
||||
|
||||
## 加载模型
|
||||
|
||||
使用您的设备和数据类型初始化模型管理器
|
||||
|
||||
```python
|
||||
model_manager = ModelManager(torch_dtype=torch.bfloat16, device="cuda")
|
||||
model_manager.load_models([
|
||||
"models/FLUX/FLUX.1-dev/text_encoder/model.safetensors",
|
||||
"models/FLUX/FLUX.1-dev/text_encoder_2",
|
||||
"models/FLUX/FLUX.1-dev/ae.safetensors",
|
||||
"models/FLUX/FLUX.1-dev/flux1-dev.safetensors"
|
||||
])
|
||||
```
|
||||
|
||||
模型加载的用法详见 [ModelManager.md](ModelManager.md)
|
||||
|
||||
## 创建 Pipeline
|
||||
|
||||
从加载的模型管理器中创建FluxImagePipeline实例：
|
||||
|
||||
```python
|
||||
pipe = FluxImagePipeline.from_model_manager(model_manager)
|
||||
```
|
||||
|
||||
Pipeline 的用法详见 [Pipeline.md](Pipeline.md)
|
||||
|
||||
## 文生图
|
||||
|
||||
使用简短的提示语生成图像。以下是启用和禁用 classifier-free guidance 的图像生成示例。
|
||||
|
||||
### 基础文生图
|
||||
|
||||
```python
|
||||
prompt = "A cute little turtle"
|
||||
negative_prompt = ""
|
||||
|
||||
torch.manual_seed(6)
|
||||
image = pipe(
|
||||
prompt=prompt,
|
||||
num_inference_steps=30, embedded_guidance=3.5
|
||||
)
|
||||
image.save("image_1024.jpg")
|
||||
```
|
||||
|
||||
### 使用 Classifier-Free Guidance 生成
|
||||
```python
|
||||
torch.manual_seed(6)
|
||||
image = pipe(
|
||||
prompt=prompt, negative_prompt=negative_prompt,
|
||||
num_inference_steps=30, cfg_scale=2.0, embedded_guidance=3.5
|
||||
)
|
||||
image.save("image_1024_cfg.jpg")
|
||||
```
|
||||
|
||||
### 高分辨率修复
|
||||
|
||||
```python
|
||||
torch.manual_seed(7)
|
||||
image = pipe(
|
||||
prompt=prompt,
|
||||
num_inference_steps=30, embedded_guidance=3.5,
|
||||
input_image=image.resize((2048, 2048)), height=2048, width=2048, denoising_strength=0.6, tiled=True
|
||||
)
|
||||
image.save("image_2048_highres.jpg")
|
||||
```
|
||||
|
||||
@@ -1,20 +0,0 @@
|
||||
# 下载模型
|
||||
|
||||
下载预设模型，模型ID可参考 [config file](/diffsynth/configs/model_config.py).
|
||||
|
||||
```python
|
||||
from diffsynth import download_models
|
||||
|
||||
download_models(["FLUX.1-dev", "Kolors"])
|
||||
```
|
||||
|
||||
下载非预设模型，可以选择 [ModelScope](https://modelscope.cn/models) 和 [HuggingFace](https://huggingface.co/models) 两个下载源中的模型。
|
||||
|
||||
```python
|
||||
from diffsynth.models.downloader import download_from_huggingface, download_from_modelscope
|
||||
|
||||
# From Modelscope (recommended)
|
||||
download_from_modelscope("Kwai-Kolors/Kolors", "vae/diffusion_pytorch_model.fp16.bin", "models/kolors/Kolors/vae")
|
||||
# From Huggingface
|
||||
download_from_huggingface("Kwai-Kolors/Kolors", "vae/diffusion_pytorch_model.fp16.safetensors", "models/kolors/Kolors/vae")
|
||||
```
|
||||
@@ -1,431 +0,0 @@
|
||||
# 微调
|
||||
|
||||
我们实现了一个用于文本到图像扩散模型的训练框架，使用户能够轻松地使用我们的框架训练 LoRA 模型。我们提供的脚本具有以下特点：
|
||||
|
||||
* **全面功能与用户友好性**：我们的训练框架支持多GPU和多机器配置，便于使用 DeepSpeed 加速，并包括梯度检查点优化，适用于内存需求较大的模型。
|
||||
* **代码简洁与研究者可及性**：我们避免了大块复杂的代码。通用模块实现于 `diffsynth/trainers/text_to_image.py` 中，而模型特定的训练脚本仅包含与模型架构相关的最少代码，便于研究人员使用。
|
||||
* **模块化设计与开发者灵活性**：基于通用的 Pytorch-Lightning 框架，我们的训练框架在功能上是解耦的，允许开发者通过修改我们的脚本轻松引入额外的训练技术，以满足他们的需求。
|
||||
|
||||
LoRA 微调的图像示例。提示词为 "一只小狗蹦蹦跳跳，周围是姹紫嫣红的鲜花，远处是山脉"（针对中文模型）或 "a dog is jumping, flowers around the dog, the background is mountains and clouds"（针对英文模型）。
|
||||
|
||||
||Kolors|Stable Diffusion 3|Hunyuan-DiT|
|
||||
|-|-|-|-|
|
||||
|Without LoRA||||
|
||||
|With LoRA||||
|
||||
|
||||
## 下载需要的包
|
||||
|
||||
```bash
|
||||
pip install peft lightning
|
||||
```
|
||||
|
||||
## 准备你的数据
|
||||
|
||||
我们提供了一个 [示例数据集](https://modelscope.cn/datasets/buptwq/lora-stable-diffusion-finetune/files)。你需要将训练数据集按照如下形式组织：
|
||||
|
||||
|
||||
```
|
||||
data/dog/
|
||||
└── train
|
||||
├── 00.jpg
|
||||
├── 01.jpg
|
||||
├── 02.jpg
|
||||
├── 03.jpg
|
||||
├── 04.jpg
|
||||
└── metadata.csv
|
||||
```
|
||||
|
||||
`metadata.csv`:
|
||||
|
||||
```
|
||||
file_name,text
|
||||
00.jpg,a dog
|
||||
01.jpg,a dog
|
||||
02.jpg,a dog
|
||||
03.jpg,a dog
|
||||
04.jpg,a dog
|
||||
```
|
||||
|
||||
请注意，如果模型是中文模型（例如，Hunyuan-DiT 和 Kolors），我们建议在数据集中使用中文文本。例如：
|
||||
|
||||
```
|
||||
file_name,text
|
||||
00.jpg,一只小狗
|
||||
01.jpg,一只小狗
|
||||
02.jpg,一只小狗
|
||||
03.jpg,一只小狗
|
||||
04.jpg,一只小狗
|
||||
```
|
||||
|
||||
## 训练 LoRA 模型
|
||||
|
||||
参数选项：
|
||||
|
||||
```
|
||||
--lora_target_modules LORA_TARGET_MODULES
|
||||
LoRA 模块所在的层。
|
||||
--dataset_path DATASET_PATH
|
||||
数据集的路径。
|
||||
--output_path OUTPUT_PATH
|
||||
模型保存路径。
|
||||
--steps_per_epoch STEPS_PER_EPOCH
|
||||
每个周期的步数。
|
||||
--height HEIGHT 图像高度。
|
||||
--width WIDTH 图像宽度。
|
||||
--center_crop 是否将输入图像中心裁剪到指定分辨率。如果未设置，图像将被随机裁剪。图像会在裁剪前先调整到指定分辨率。
|
||||
--random_flip 是否随机水平翻转图像。
|
||||
--batch_size BATCH_SIZE
|
||||
训练数据加载器的批量大小（每设备）。
|
||||
--dataloader_num_workers DATALOADER_NUM_WORKERS
|
||||
数据加载使用的子进程数量。0 表示数据将在主进程中加载。
|
||||
--precision {32,16,16-mixed}
|
||||
训练精度。
|
||||
--learning_rate LEARNING_RATE
|
||||
学习率。
|
||||
--lora_rank LORA_RANK
|
||||
LoRA 更新矩阵的维度。
|
||||
--lora_alpha LORA_ALPHA
|
||||
LoRA 更新矩阵的权重。
|
||||
--use_gradient_checkpointing
|
||||
是否使用梯度检查点。
|
||||
--accumulate_grad_batches ACCUMULATE_GRAD_BATCHES
|
||||
梯度累积的批次数量。
|
||||
--training_strategy {auto,deepspeed_stage_1,deepspeed_stage_2,deepspeed_stage_3}
|
||||
训练策略。
|
||||
--max_epochs MAX_EPOCHS
|
||||
训练周期数。
|
||||
--modelscope_model_id MODELSCOPE_MODEL_ID
|
||||
ModelScope 上的模型 ID (https://www.modelscope.cn/)。如果提供模型 ID，模型将自动上传到 ModelScope。
|
||||
|
||||
```
|
||||
|
||||
### Kolors
|
||||
|
||||
以下文件将用于构建 Kolors。你可以从 [HuggingFace](https://huggingface.co/Kwai-Kolors/Kolors) 或 [ModelScope](https://modelscope.cn/models/Kwai-Kolors/Kolors) 下载 Kolors。由于精度溢出问题，我们需要下载额外的 VAE 模型（从 [HuggingFace](https://huggingface.co/madebyollin/sdxl-vae-fp16-fix) 或 [ModelScope](https://modelscope.cn/models/AI-ModelScope/sdxl-vae-fp16-fix)）。你可以使用以下代码下载这些文件：
|
||||
|
||||
|
||||
```python
|
||||
from diffsynth import download_models
|
||||
|
||||
download_models(["Kolors", "SDXL-vae-fp16-fix"])
|
||||
```
|
||||
|
||||
```
|
||||
models
|
||||
├── kolors
|
||||
│ └── Kolors
|
||||
│ ├── text_encoder
|
||||
│ │ ├── config.json
|
||||
│ │ ├── pytorch_model-00001-of-00007.bin
|
||||
│ │ ├── pytorch_model-00002-of-00007.bin
|
||||
│ │ ├── pytorch_model-00003-of-00007.bin
|
||||
│ │ ├── pytorch_model-00004-of-00007.bin
|
||||
│ │ ├── pytorch_model-00005-of-00007.bin
|
||||
│ │ ├── pytorch_model-00006-of-00007.bin
|
||||
│ │ ├── pytorch_model-00007-of-00007.bin
|
||||
│ │ └── pytorch_model.bin.index.json
|
||||
│ ├── unet
|
||||
│ │ └── diffusion_pytorch_model.safetensors
|
||||
│ └── vae
|
||||
│ └── diffusion_pytorch_model.safetensors
|
||||
└── sdxl-vae-fp16-fix
|
||||
└── diffusion_pytorch_model.safetensors
|
||||
```
|
||||
|
||||
使用下面的命令启动训练任务：
|
||||
|
||||
```
|
||||
CUDA_VISIBLE_DEVICES="0" python examples/train/kolors/train_kolors_lora.py \
|
||||
--pretrained_unet_path models/kolors/Kolors/unet/diffusion_pytorch_model.safetensors \
|
||||
--pretrained_text_encoder_path models/kolors/Kolors/text_encoder \
|
||||
--pretrained_fp16_vae_path models/sdxl-vae-fp16-fix/diffusion_pytorch_model.safetensors \
|
||||
--dataset_path data/dog \
|
||||
--output_path ./models \
|
||||
--max_epochs 1 \
|
||||
--steps_per_epoch 500 \
|
||||
--height 1024 \
|
||||
--width 1024 \
|
||||
--center_crop \
|
||||
--precision "16-mixed" \
|
||||
--learning_rate 1e-4 \
|
||||
--lora_rank 4 \
|
||||
--lora_alpha 4 \
|
||||
--use_gradient_checkpointing
|
||||
```
|
||||
|
||||
有关参数的更多信息，请使用 `python examples/train/kolors/train_kolors_lora.py -h` 查看详细信息。
|
||||
|
||||
训练完成后，使用 `model_manager.load_lora` 加载 LoRA 以进行推理。
|
||||
|
||||
|
||||
|
||||
```python
|
||||
from diffsynth import ModelManager, SD3ImagePipeline
|
||||
import torch
|
||||
|
||||
model_manager = ModelManager(torch_dtype=torch.float16, device="cuda",
|
||||
file_path_list=["models/stable_diffusion_3/sd3_medium_incl_clips.safetensors"])
|
||||
model_manager.load_lora("models/lightning_logs/version_0/checkpoints/epoch=0-step=500.ckpt", lora_alpha=1.0)
|
||||
pipe = SD3ImagePipeline.from_model_manager(model_manager)
|
||||
|
||||
torch.manual_seed(0)
|
||||
image = pipe(
|
||||
prompt="a dog is jumping, flowers around the dog, the background is mountains and clouds",
|
||||
negative_prompt="bad quality, poor quality, doll, disfigured, jpg, toy, bad anatomy, missing limbs, missing fingers, 3d, cgi, extra tails",
|
||||
cfg_scale=7.5,
|
||||
num_inference_steps=100, width=1024, height=1024,
|
||||
)
|
||||
image.save("image_with_lora.jpg")
|
||||
```
|
||||
|
||||
|
||||
### Stable Diffusion 3
|
||||
|
||||
训练脚本只需要一个文件。你可以使用 [`sd3_medium_incl_clips.safetensors`](https://huggingface.co/stabilityai/stable-diffusion-3-medium/resolve/main/sd3_medium_incl_clips.safetensors)（没有 T5 Encoder）或 [`sd3_medium_incl_clips_t5xxlfp16.safetensors`](https://huggingface.co/stabilityai/stable-diffusion-3-medium/resolve/main/sd3_medium_incl_clips_t5xxlfp16.safetensors)（有 T5 Encoder）。请使用以下代码下载这些文件：
|
||||
|
||||
|
||||
```python
|
||||
from diffsynth import download_models
|
||||
|
||||
download_models(["StableDiffusion3", "StableDiffusion3_without_T5"])
|
||||
```
|
||||
|
||||
```
|
||||
models/stable_diffusion_3/
|
||||
├── Put Stable Diffusion 3 checkpoints here.txt
|
||||
├── sd3_medium_incl_clips.safetensors
|
||||
└── sd3_medium_incl_clips_t5xxlfp16.safetensors
|
||||
```
|
||||
|
||||
使用下面的命令启动训练任务：
|
||||
|
||||
```
|
||||
CUDA_VISIBLE_DEVICES="0" python examples/train/stable_diffusion_3/train_sd3_lora.py \
|
||||
--pretrained_path models/stable_diffusion_3/sd3_medium_incl_clips.safetensors \
|
||||
--dataset_path data/dog \
|
||||
--output_path ./models \
|
||||
--max_epochs 1 \
|
||||
--steps_per_epoch 500 \
|
||||
--height 1024 \
|
||||
--width 1024 \
|
||||
--center_crop \
|
||||
--precision "16-mixed" \
|
||||
--learning_rate 1e-4 \
|
||||
--lora_rank 4 \
|
||||
--lora_alpha 4 \
|
||||
--use_gradient_checkpointing
|
||||
```
|
||||
|
||||
有关参数的更多信息，请使用 `python examples/train/stable_diffusion_3/train_sd3_lora.py -h` 查看详细信息。
|
||||
|
||||
训练完成后，使用 `model_manager.load_lora` 加载 LoRA 以进行推理。
|
||||
|
||||
```python
|
||||
from diffsynth import ModelManager, SD3ImagePipeline
|
||||
import torch
|
||||
|
||||
model_manager = ModelManager(torch_dtype=torch.float16, device="cuda",
|
||||
file_path_list=["models/stable_diffusion_3/sd3_medium_incl_clips.safetensors"])
|
||||
model_manager.load_lora("models/lightning_logs/version_0/checkpoints/epoch=0-step=500.ckpt", lora_alpha=1.0)
|
||||
pipe = SD3ImagePipeline.from_model_manager(model_manager)
|
||||
|
||||
torch.manual_seed(0)
|
||||
image = pipe(
|
||||
prompt="a dog is jumping, flowers around the dog, the background is mountains and clouds",
|
||||
negative_prompt="bad quality, poor quality, doll, disfigured, jpg, toy, bad anatomy, missing limbs, missing fingers, 3d, cgi, extra tails",
|
||||
cfg_scale=7.5,
|
||||
num_inference_steps=100, width=1024, height=1024,
|
||||
)
|
||||
image.save("image_with_lora.jpg")
|
||||
```
|
||||
|
||||
### Hunyuan-DiT
|
||||
|
||||
构建 Hunyuan DiT 需要四个文件。你可以从 [HuggingFace](https://huggingface.co/Tencent-Hunyuan/HunyuanDiT) 或 [ModelScope](https://www.modelscope.cn/models/modelscope/HunyuanDiT/summary) 下载这些文件。你可以使用以下代码下载这些文件：
|
||||
|
||||
|
||||
```python
|
||||
from diffsynth import download_models
|
||||
|
||||
download_models(["HunyuanDiT"])
|
||||
```
|
||||
|
||||
```
|
||||
models/HunyuanDiT/
|
||||
├── Put Hunyuan DiT checkpoints here.txt
|
||||
└── t2i
|
||||
├── clip_text_encoder
|
||||
│ └── pytorch_model.bin
|
||||
├── model
|
||||
│ └── pytorch_model_ema.pt
|
||||
├── mt5
|
||||
│ └── pytorch_model.bin
|
||||
└── sdxl-vae-fp16-fix
|
||||
└── diffusion_pytorch_model.bin
|
||||
```
|
||||
|
||||
Launch the training task using the following command:
|
||||
|
||||
```
|
||||
CUDA_VISIBLE_DEVICES="0" python examples/train/hunyuan_dit/train_hunyuan_dit_lora.py \
|
||||
--pretrained_path models/HunyuanDiT/t2i \
|
||||
--dataset_path data/dog \
|
||||
--output_path ./models \
|
||||
--max_epochs 1 \
|
||||
--steps_per_epoch 500 \
|
||||
--height 1024 \
|
||||
--width 1024 \
|
||||
--center_crop \
|
||||
--precision "16-mixed" \
|
||||
--learning_rate 1e-4 \
|
||||
--lora_rank 4 \
|
||||
--lora_alpha 4 \
|
||||
--use_gradient_checkpointing
|
||||
```
|
||||
|
||||
有关参数的更多信息，请使用 `python examples/train/hunyuan_dit/train_hunyuan_dit_lora.py -h` 查看详细信息。
|
||||
|
||||
训练完成后，使用 `model_manager.load_lora` 加载 LoRA 以进行推理。
|
||||
|
||||
|
||||
```python
|
||||
from diffsynth import ModelManager, HunyuanDiTImagePipeline
|
||||
import torch
|
||||
|
||||
model_manager = ModelManager(torch_dtype=torch.float16, device="cuda",
|
||||
file_path_list=[
|
||||
"models/HunyuanDiT/t2i/clip_text_encoder/pytorch_model.bin",
|
||||
"models/HunyuanDiT/t2i/model/pytorch_model_ema.pt",
|
||||
"models/HunyuanDiT/t2i/mt5/pytorch_model.bin",
|
||||
"models/HunyuanDiT/t2i/sdxl-vae-fp16-fix/diffusion_pytorch_model.bin"
|
||||
])
|
||||
model_manager.load_lora("models/lightning_logs/version_0/checkpoints/epoch=0-step=500.ckpt", lora_alpha=1.0)
|
||||
pipe = HunyuanDiTImagePipeline.from_model_manager(model_manager)
|
||||
|
||||
torch.manual_seed(0)
|
||||
image = pipe(
|
||||
prompt="一只小狗蹦蹦跳跳，周围是姹紫嫣红的鲜花，远处是山脉",
|
||||
negative_prompt="",
|
||||
cfg_scale=7.5,
|
||||
num_inference_steps=100, width=1024, height=1024,
|
||||
)
|
||||
image.save("image_with_lora.jpg")
|
||||
```
|
||||
|
||||
### Stable Diffusion
|
||||
|
||||
训练脚本只需要一个文件。我们支持 [CivitAI](https://civitai.com/) 中的主流检查点。默认情况下，我们使用基础的 Stable Diffusion v1.5。你可以从 [HuggingFace](https://huggingface.co/runwayml/stable-diffusion-v1-5/resolve/main/v1-5-pruned-emaonly.safetensors) 或 [ModelScope](https://www.modelscope.cn/models/AI-ModelScope/stable-diffusion-v1-5/resolve/master/v1-5-pruned-emaonly.safetensors) 下载。你可以使用以下代码下载这个文件：
|
||||
|
||||
```python
|
||||
from diffsynth import download_models
|
||||
|
||||
download_models(["StableDiffusion_v15"])
|
||||
```
|
||||
|
||||
```
|
||||
models/stable_diffusion
|
||||
├── Put Stable Diffusion checkpoints here.txt
|
||||
└── v1-5-pruned-emaonly.safetensors
|
||||
```
|
||||
|
||||
Launch the training task using the following command:
|
||||
|
||||
```
|
||||
CUDA_VISIBLE_DEVICES="0" python examples/train/stable_diffusion/train_sd_lora.py \
|
||||
--pretrained_path models/stable_diffusion/v1-5-pruned-emaonly.safetensors \
|
||||
--dataset_path data/dog \
|
||||
--output_path ./models \
|
||||
--max_epochs 1 \
|
||||
--steps_per_epoch 500 \
|
||||
--height 512 \
|
||||
--width 512 \
|
||||
--center_crop \
|
||||
--precision "16-mixed" \
|
||||
--learning_rate 1e-4 \
|
||||
--lora_rank 4 \
|
||||
--lora_alpha 4 \
|
||||
--use_gradient_checkpointing
|
||||
```
|
||||
|
||||
有关参数的更多信息，请使用 `python examples/train/stable_diffusion/train_sd_lora.py -h` 查看详细信息。
|
||||
|
||||
训练完成后，使用 `model_manager.load_lora` 加载 LoRA 以进行推理。
|
||||
|
||||
|
||||
|
||||
```python
|
||||
from diffsynth import ModelManager, SDImagePipeline
|
||||
import torch
|
||||
|
||||
model_manager = ModelManager(torch_dtype=torch.float16, device="cuda",
|
||||
file_path_list=["models/stable_diffusion/v1-5-pruned-emaonly.safetensors"])
|
||||
model_manager.load_lora("models/lightning_logs/version_0/checkpoints/epoch=0-step=500.ckpt", lora_alpha=1.0)
|
||||
pipe = SDImagePipeline.from_model_manager(model_manager)
|
||||
|
||||
torch.manual_seed(0)
|
||||
image = pipe(
|
||||
prompt="a dog is jumping, flowers around the dog, the background is mountains and clouds",
|
||||
negative_prompt="bad quality, poor quality, doll, disfigured, jpg, toy, bad anatomy, missing limbs, missing fingers, 3d, cgi, extra tails",
|
||||
cfg_scale=7.5,
|
||||
num_inference_steps=100, width=512, height=512,
|
||||
)
|
||||
image.save("image_with_lora.jpg")
|
||||
```
|
||||
|
||||
### Stable Diffusion XL
|
||||
|
||||
训练脚本只需要一个文件。我们支持 [CivitAI](https://civitai.com/) 中的主流检查点。默认情况下，我们使用基础的 Stable Diffusion XL。你可以从 [HuggingFace](https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0/resolve/main/sd_xl_base_1.0.safetensors) 或 [ModelScope](https://www.modelscope.cn/models/AI-ModelScope/stable-diffusion-xl-base-1.0/resolve/master/sd_xl_base_1.0.safetensors) 下载。也可以使用以下代码下载这个文件：
|
||||
|
||||
```python
|
||||
from diffsynth import download_models
|
||||
|
||||
download_models(["StableDiffusionXL_v1"])
|
||||
```
|
||||
|
||||
```
|
||||
models/stable_diffusion_xl
|
||||
├── Put Stable Diffusion XL checkpoints here.txt
|
||||
└── sd_xl_base_1.0.safetensors
|
||||
```
|
||||
|
||||
We observed that Stable Diffusion XL is not float16-safe, thus we recommand users to use float32.
|
||||
|
||||
```
|
||||
CUDA_VISIBLE_DEVICES="0" python examples/train/stable_diffusion_xl/train_sdxl_lora.py \
|
||||
--pretrained_path models/stable_diffusion_xl/sd_xl_base_1.0.safetensors \
|
||||
--dataset_path data/dog \
|
||||
--output_path ./models \
|
||||
--max_epochs 1 \
|
||||
--steps_per_epoch 500 \
|
||||
--height 1024 \
|
||||
--width 1024 \
|
||||
--center_crop \
|
||||
--precision "32" \
|
||||
--learning_rate 1e-4 \
|
||||
--lora_rank 4 \
|
||||
--lora_alpha 4 \
|
||||
--use_gradient_checkpointing
|
||||
```
|
||||
|
||||
有关参数的更多信息，请使用 `python examples/train/stable_diffusion_xl/train_sdxl_lora.py -h` 查看详细信息。
|
||||
|
||||
训练完成后，使用 `model_manager.load_lora` 加载 LoRA 以进行推理。
|
||||
|
||||
```python
|
||||
from diffsynth import ModelManager, SDXLImagePipeline
|
||||
import torch
|
||||
|
||||
model_manager = ModelManager(torch_dtype=torch.float16, device="cuda",
|
||||
file_path_list=["models/stable_diffusion_xl/sd_xl_base_1.0.safetensors"])
|
||||
model_manager.load_lora("models/lightning_logs/version_0/checkpoints/epoch=0-step=500.ckpt", lora_alpha=1.0)
|
||||
pipe = SDXLImagePipeline.from_model_manager(model_manager)
|
||||
|
||||
torch.manual_seed(0)
|
||||
image = pipe(
|
||||
prompt="a dog is jumping, flowers around the dog, the background is mountains and clouds",
|
||||
negative_prompt="bad quality, poor quality, doll, disfigured, jpg, toy, bad anatomy, missing limbs, missing fingers, 3d, cgi, extra tails",
|
||||
cfg_scale=7.5,
|
||||
num_inference_steps=100, width=1024, height=1024,
|
||||
)
|
||||
image.save("image_with_lora.jpg")
|
||||
```
|
||||
98
docs/source/finetune/overview.md
Normal file
98
docs/source/finetune/overview.md
Normal file
@@ -0,0 +1,98 @@
|
||||
# 训练框架
|
||||
|
||||
我们实现了一个用于文本到图像扩散模型的训练框架，使用户能够轻松地使用我们的框架训练 LoRA 模型。我们提供的脚本具有以下特点：
|
||||
|
||||
* **功能全面**：我们的训练框架支持多GPU和多机器配置，便于使用 DeepSpeed 加速，并包括梯度检查点优化，适用于内存需求较大的模型。
|
||||
* **代码简洁**：我们避免了大块复杂的代码。通用模块实现于 `diffsynth/trainers/text_to_image.py` 中，而模型特定的训练脚本仅包含与模型架构相关的最少代码，便于学术研究人员使用。
|
||||
* **模块化设计**：基于通用的 Pytorch-Lightning 框架，我们的训练框架在功能上是解耦的，允许开发者通过修改我们的脚本轻松引入额外的训练技术，以满足他们的需求。
|
||||
|
||||
LoRA 微调的图像示例。提示词为 "一只小狗蹦蹦跳跳，周围是姹紫嫣红的鲜花，远处是山脉"（针对中文模型）或 "a dog is jumping, flowers around the dog, the background is mountains and clouds"（针对英文模型）。
|
||||
|
||||
||FLUX.1-dev|Kolors|Stable Diffusion 3|Hunyuan-DiT|
|
||||
|-|-|-|-|-|
|
||||
|Without LoRA|||||
|
||||
|With LoRA|||||
|
||||
|
||||
## 安装额外包
|
||||
|
||||
```
|
||||
pip install peft lightning
|
||||
```
|
||||
|
||||
## 准备数据集
|
||||
|
||||
我们提供了一个[示例数据集](https://modelscope.cn/datasets/buptwq/lora-stable-diffusion-finetune/files)。你需要将训练数据集按照如下形式组织：
|
||||
|
||||
```
|
||||
data/dog/
|
||||
└── train
|
||||
├── 00.jpg
|
||||
├── 01.jpg
|
||||
├── 02.jpg
|
||||
├── 03.jpg
|
||||
├── 04.jpg
|
||||
└── metadata.csv
|
||||
```
|
||||
|
||||
`metadata.csv`:
|
||||
|
||||
```
|
||||
file_name,text
|
||||
00.jpg,a dog
|
||||
01.jpg,a dog
|
||||
02.jpg,a dog
|
||||
03.jpg,a dog
|
||||
04.jpg,a dog
|
||||
```
|
||||
|
||||
请注意，如果模型是中文模型（例如，Hunyuan-DiT 和 Kolors），我们建议在数据集中使用中文文本。例如：
|
||||
|
||||
```
|
||||
file_name,text
|
||||
00.jpg,一只小狗
|
||||
01.jpg,一只小狗
|
||||
02.jpg,一只小狗
|
||||
03.jpg,一只小狗
|
||||
04.jpg,一只小狗
|
||||
```
|
||||
|
||||
## 训练 LoRA 模型
|
||||
|
||||
通用参数选项：
|
||||
|
||||
```
|
||||
--lora_target_modules LORA_TARGET_MODULES
|
||||
LoRA 模块所在的层。
|
||||
--dataset_path DATASET_PATH
|
||||
数据集的路径。
|
||||
--output_path OUTPUT_PATH
|
||||
模型保存路径。
|
||||
--steps_per_epoch STEPS_PER_EPOCH
|
||||
每个周期的步数。
|
||||
--height HEIGHT 图像高度。
|
||||
--width WIDTH 图像宽度。
|
||||
--center_crop 是否将输入图像中心裁剪到指定分辨率。如果未设置，图像将被随机裁剪。图像会在裁剪前先调整到指定分辨率。
|
||||
--random_flip 是否随机水平翻转图像。
|
||||
--batch_size BATCH_SIZE
|
||||
训练数据加载器的批量大小（每设备）。
|
||||
--dataloader_num_workers DATALOADER_NUM_WORKERS
|
||||
数据加载使用的子进程数量。0 表示数据将在主进程中加载。
|
||||
--precision {32,16,16-mixed}
|
||||
训练精度。
|
||||
--learning_rate LEARNING_RATE
|
||||
学习率。
|
||||
--lora_rank LORA_RANK
|
||||
LoRA 更新矩阵的维度。
|
||||
--lora_alpha LORA_ALPHA
|
||||
LoRA 更新矩阵的权重。
|
||||
--use_gradient_checkpointing
|
||||
是否使用梯度检查点。
|
||||
--accumulate_grad_batches ACCUMULATE_GRAD_BATCHES
|
||||
梯度累积的批次数量。
|
||||
--training_strategy {auto,deepspeed_stage_1,deepspeed_stage_2,deepspeed_stage_3}
|
||||
训练策略。
|
||||
--max_epochs MAX_EPOCHS
|
||||
训练轮数。
|
||||
--modelscope_model_id MODELSCOPE_MODEL_ID
|
||||
ModelScope 上的模型 ID (https://www.modelscope.cn/)。如果提供模型 ID，模型将自动上传到 ModelScope。
|
||||
```
|
||||
71
docs/source/finetune/train_flux_lora.md
Normal file
71
docs/source/finetune/train_flux_lora.md
Normal file
@@ -0,0 +1,71 @@
|
||||
# 训练 FLUX LoRA
|
||||
|
||||
以下文件将会被用于构建 FLUX 模型。 你可以从[huggingface](https://huggingface.co/black-forest-labs/FLUX.1-dev)或[modelscope](https://www.modelscope.cn/models/ai-modelscope/flux.1-dev)下载，也可以使用以下代码下载这些文件:
|
||||
|
||||
```python
|
||||
from diffsynth import download_models
|
||||
|
||||
download_models(["FLUX.1-dev"])
|
||||
```
|
||||
|
||||
```
|
||||
models/FLUX/
|
||||
└── FLUX.1-dev
|
||||
├── ae.safetensors
|
||||
├── flux1-dev.safetensors
|
||||
├── text_encoder
|
||||
│ └── model.safetensors
|
||||
└── text_encoder_2
|
||||
├── config.json
|
||||
├── model-00001-of-00002.safetensors
|
||||
├── model-00002-of-00002.safetensors
|
||||
└── model.safetensors.index.json
|
||||
```
|
||||
|
||||
使用以下命令启动训练任务：
|
||||
|
||||
```
|
||||
CUDA_VISIBLE_DEVICES="0" python examples/train/flux/train_flux_lora.py \
|
||||
--pretrained_text_encoder_path models/FLUX/FLUX.1-dev/text_encoder/model.safetensors \
|
||||
--pretrained_text_encoder_2_path models/FLUX/FLUX.1-dev/text_encoder_2 \
|
||||
--pretrained_dit_path models/FLUX/FLUX.1-dev/flux1-dev.safetensors \
|
||||
--pretrained_vae_path models/FLUX/FLUX.1-dev/ae.safetensors \
|
||||
--dataset_path data/dog \
|
||||
--output_path ./models \
|
||||
--max_epochs 1 \
|
||||
--steps_per_epoch 500 \
|
||||
--height 1024 \
|
||||
--width 1024 \
|
||||
--center_crop \
|
||||
--precision "bf16" \
|
||||
--learning_rate 1e-4 \
|
||||
--lora_rank 4 \
|
||||
--lora_alpha 4 \
|
||||
--use_gradient_checkpointing
|
||||
```
|
||||
|
||||
有关参数的更多信息，请使用 `python examples/train/flux/train_flux_lora.py -h` 查看详细信息。
|
||||
|
||||
训练完成后，使用 `model_manager.load_lora` 加载 LoRA 以进行推理。
|
||||
|
||||
```python
|
||||
from diffsynth import ModelManager, FluxImagePipeline
|
||||
import torch
|
||||
|
||||
model_manager = ModelManager(torch_dtype=torch.float16, device="cuda",
|
||||
file_path_list=[
|
||||
"models/FLUX/FLUX.1-dev/text_encoder/model.safetensors",
|
||||
"models/FLUX/FLUX.1-dev/text_encoder_2",
|
||||
"models/FLUX/FLUX.1-dev/ae.safetensors",
|
||||
"models/FLUX/FLUX.1-dev/flux1-dev.safetensors"
|
||||
])
|
||||
model_manager.load_lora("models/lightning_logs/version_0/checkpoints/epoch=0-step=500.ckpt", lora_alpha=1.0)
|
||||
pipe = SDXLImagePipeline.from_model_manager(model_manager)
|
||||
|
||||
torch.manual_seed(0)
|
||||
image = pipe(
|
||||
prompt=prompt,
|
||||
num_inference_steps=30, embedded_guidance=3.5
|
||||
)
|
||||
image.save("image_with_lora.jpg")
|
||||
```
|
||||
72
docs/source/finetune/train_hunyuan_dit_lora.md
Normal file
72
docs/source/finetune/train_hunyuan_dit_lora.md
Normal file
@@ -0,0 +1,72 @@
|
||||
# 训练 Hunyuan-DiT LoRA
|
||||
|
||||
构建 Hunyuan DiT 需要四个文件。你可以从 [HuggingFace](https://huggingface.co/Tencent-Hunyuan/HunyuanDiT) 或 [ModelScope](https://www.modelscope.cn/models/modelscope/HunyuanDiT/summary) 下载这些文件。你可以使用以下代码下载这些文件：
|
||||
|
||||
|
||||
```python
|
||||
from diffsynth import download_models
|
||||
|
||||
download_models(["HunyuanDiT"])
|
||||
```
|
||||
|
||||
```
|
||||
models/HunyuanDiT/
|
||||
├── Put Hunyuan DiT checkpoints here.txt
|
||||
└── t2i
|
||||
├── clip_text_encoder
|
||||
│ └── pytorch_model.bin
|
||||
├── model
|
||||
│ └── pytorch_model_ema.pt
|
||||
├── mt5
|
||||
│ └── pytorch_model.bin
|
||||
└── sdxl-vae-fp16-fix
|
||||
└── diffusion_pytorch_model.bin
|
||||
```
|
||||
|
||||
使用以下命令启动训练任务：
|
||||
|
||||
```
|
||||
CUDA_VISIBLE_DEVICES="0" python examples/train/hunyuan_dit/train_hunyuan_dit_lora.py \
|
||||
--pretrained_path models/HunyuanDiT/t2i \
|
||||
--dataset_path data/dog \
|
||||
--output_path ./models \
|
||||
--max_epochs 1 \
|
||||
--steps_per_epoch 500 \
|
||||
--height 1024 \
|
||||
--width 1024 \
|
||||
--center_crop \
|
||||
--precision "16-mixed" \
|
||||
--learning_rate 1e-4 \
|
||||
--lora_rank 4 \
|
||||
--lora_alpha 4 \
|
||||
--use_gradient_checkpointing
|
||||
```
|
||||
|
||||
有关参数的更多信息，请使用 `python examples/train/hunyuan_dit/train_hunyuan_dit_lora.py -h` 查看详细信息。
|
||||
|
||||
训练完成后，使用 `model_manager.load_lora` 加载 LoRA 以进行推理。
|
||||
|
||||
|
||||
```python
|
||||
from diffsynth import ModelManager, HunyuanDiTImagePipeline
|
||||
import torch
|
||||
|
||||
model_manager = ModelManager(torch_dtype=torch.float16, device="cuda",
|
||||
file_path_list=[
|
||||
"models/HunyuanDiT/t2i/clip_text_encoder/pytorch_model.bin",
|
||||
"models/HunyuanDiT/t2i/model/pytorch_model_ema.pt",
|
||||
"models/HunyuanDiT/t2i/mt5/pytorch_model.bin",
|
||||
"models/HunyuanDiT/t2i/sdxl-vae-fp16-fix/diffusion_pytorch_model.bin"
|
||||
])
|
||||
model_manager.load_lora("models/lightning_logs/version_0/checkpoints/epoch=0-step=500.ckpt", lora_alpha=1.0)
|
||||
pipe = HunyuanDiTImagePipeline.from_model_manager(model_manager)
|
||||
|
||||
torch.manual_seed(0)
|
||||
image = pipe(
|
||||
prompt="一只小狗蹦蹦跳跳，周围是姹紫嫣红的鲜花，远处是山脉",
|
||||
negative_prompt="",
|
||||
cfg_scale=7.5,
|
||||
num_inference_steps=100, width=1024, height=1024,
|
||||
)
|
||||
image.save("image_with_lora.jpg")
|
||||
```
|
||||
78
docs/source/finetune/train_kolors_lora.md
Normal file
78
docs/source/finetune/train_kolors_lora.md
Normal file
@@ -0,0 +1,78 @@
|
||||
# 训练 Kolors LoRA
|
||||
|
||||
以下文件将用于构建 Kolors。你可以从 [HuggingFace](https://huggingface.co/Kwai-Kolors/Kolors) 或 [ModelScope](https://modelscope.cn/models/Kwai-Kolors/Kolors) 下载 Kolors。由于精度溢出问题，我们需要下载额外的 VAE 模型（从 [HuggingFace](https://huggingface.co/madebyollin/sdxl-vae-fp16-fix) 或 [ModelScope](https://modelscope.cn/models/AI-ModelScope/sdxl-vae-fp16-fix)）。你可以使用以下代码下载这些文件：
|
||||
|
||||
|
||||
```python
|
||||
from diffsynth import download_models
|
||||
|
||||
download_models(["Kolors", "SDXL-vae-fp16-fix"])
|
||||
```
|
||||
|
||||
```
|
||||
models
|
||||
├── kolors
|
||||
│ └── Kolors
|
||||
│ ├── text_encoder
|
||||
│ │ ├── config.json
|
||||
│ │ ├── pytorch_model-00001-of-00007.bin
|
||||
│ │ ├── pytorch_model-00002-of-00007.bin
|
||||
│ │ ├── pytorch_model-00003-of-00007.bin
|
||||
│ │ ├── pytorch_model-00004-of-00007.bin
|
||||
│ │ ├── pytorch_model-00005-of-00007.bin
|
||||
│ │ ├── pytorch_model-00006-of-00007.bin
|
||||
│ │ ├── pytorch_model-00007-of-00007.bin
|
||||
│ │ └── pytorch_model.bin.index.json
|
||||
│ ├── unet
|
||||
│ │ └── diffusion_pytorch_model.safetensors
|
||||
│ └── vae
|
||||
│ └── diffusion_pytorch_model.safetensors
|
||||
└── sdxl-vae-fp16-fix
|
||||
└── diffusion_pytorch_model.safetensors
|
||||
```
|
||||
|
||||
使用下面的命令启动训练任务：
|
||||
|
||||
```
|
||||
CUDA_VISIBLE_DEVICES="0" python examples/train/kolors/train_kolors_lora.py \
|
||||
--pretrained_unet_path models/kolors/Kolors/unet/diffusion_pytorch_model.safetensors \
|
||||
--pretrained_text_encoder_path models/kolors/Kolors/text_encoder \
|
||||
--pretrained_fp16_vae_path models/sdxl-vae-fp16-fix/diffusion_pytorch_model.safetensors \
|
||||
--dataset_path data/dog \
|
||||
--output_path ./models \
|
||||
--max_epochs 1 \
|
||||
--steps_per_epoch 500 \
|
||||
--height 1024 \
|
||||
--width 1024 \
|
||||
--center_crop \
|
||||
--precision "16-mixed" \
|
||||
--learning_rate 1e-4 \
|
||||
--lora_rank 4 \
|
||||
--lora_alpha 4 \
|
||||
--use_gradient_checkpointing
|
||||
```
|
||||
|
||||
有关参数的更多信息，请使用 `python examples/train/kolors/train_kolors_lora.py -h` 查看详细信息。
|
||||
|
||||
训练完成后，使用 `model_manager.load_lora` 加载 LoRA 以进行推理。
|
||||
|
||||
|
||||
|
||||
```python
|
||||
from diffsynth import ModelManager, SD3ImagePipeline
|
||||
import torch
|
||||
|
||||
model_manager = ModelManager(torch_dtype=torch.float16, device="cuda",
|
||||
file_path_list=["models/stable_diffusion_3/sd3_medium_incl_clips.safetensors"])
|
||||
model_manager.load_lora("models/lightning_logs/version_0/checkpoints/epoch=0-step=500.ckpt", lora_alpha=1.0)
|
||||
pipe = SD3ImagePipeline.from_model_manager(model_manager)
|
||||
|
||||
torch.manual_seed(0)
|
||||
image = pipe(
|
||||
prompt="a dog is jumping, flowers around the dog, the background is mountains and clouds",
|
||||
negative_prompt="bad quality, poor quality, doll, disfigured, jpg, toy, bad anatomy, missing limbs, missing fingers, 3d, cgi, extra tails",
|
||||
cfg_scale=7.5,
|
||||
num_inference_steps=100, width=1024, height=1024,
|
||||
)
|
||||
image.save("image_with_lora.jpg")
|
||||
```
|
||||
59
docs/source/finetune/train_sd3_lora.md
Normal file
59
docs/source/finetune/train_sd3_lora.md
Normal file
@@ -0,0 +1,59 @@
|
||||
# 训练 Stable Diffusion 3 LoRA
|
||||
|
||||
训练脚本只需要一个文件。你可以使用 [`sd3_medium_incl_clips.safetensors`](https://huggingface.co/stabilityai/stable-diffusion-3-medium/resolve/main/sd3_medium_incl_clips.safetensors)（没有 T5 Encoder）或 [`sd3_medium_incl_clips_t5xxlfp16.safetensors`](https://huggingface.co/stabilityai/stable-diffusion-3-medium/resolve/main/sd3_medium_incl_clips_t5xxlfp16.safetensors)（有 T5 Encoder）。请使用以下代码下载这些文件：
|
||||
|
||||
|
||||
```python
|
||||
from diffsynth import download_models
|
||||
|
||||
download_models(["StableDiffusion3", "StableDiffusion3_without_T5"])
|
||||
```
|
||||
|
||||
```
|
||||
models/stable_diffusion_3/
|
||||
├── Put Stable Diffusion 3 checkpoints here.txt
|
||||
├── sd3_medium_incl_clips.safetensors
|
||||
└── sd3_medium_incl_clips_t5xxlfp16.safetensors
|
||||
```
|
||||
|
||||
使用下面的命令启动训练任务：
|
||||
|
||||
```
|
||||
CUDA_VISIBLE_DEVICES="0" python examples/train/stable_diffusion_3/train_sd3_lora.py \
|
||||
--pretrained_path models/stable_diffusion_3/sd3_medium_incl_clips.safetensors \
|
||||
--dataset_path data/dog \
|
||||
--output_path ./models \
|
||||
--max_epochs 1 \
|
||||
--steps_per_epoch 500 \
|
||||
--height 1024 \
|
||||
--width 1024 \
|
||||
--center_crop \
|
||||
--precision "16-mixed" \
|
||||
--learning_rate 1e-4 \
|
||||
--lora_rank 4 \
|
||||
--lora_alpha 4 \
|
||||
--use_gradient_checkpointing
|
||||
```
|
||||
|
||||
有关参数的更多信息，请使用 `python examples/train/stable_diffusion_3/train_sd3_lora.py -h` 查看详细信息。
|
||||
|
||||
训练完成后，使用 `model_manager.load_lora` 加载 LoRA 以进行推理。
|
||||
|
||||
```python
|
||||
from diffsynth import ModelManager, SD3ImagePipeline
|
||||
import torch
|
||||
|
||||
model_manager = ModelManager(torch_dtype=torch.float16, device="cuda",
|
||||
file_path_list=["models/stable_diffusion_3/sd3_medium_incl_clips.safetensors"])
|
||||
model_manager.load_lora("models/lightning_logs/version_0/checkpoints/epoch=0-step=500.ckpt", lora_alpha=1.0)
|
||||
pipe = SD3ImagePipeline.from_model_manager(model_manager)
|
||||
|
||||
torch.manual_seed(0)
|
||||
image = pipe(
|
||||
prompt="a dog is jumping, flowers around the dog, the background is mountains and clouds",
|
||||
negative_prompt="bad quality, poor quality, doll, disfigured, jpg, toy, bad anatomy, missing limbs, missing fingers, 3d, cgi, extra tails",
|
||||
cfg_scale=7.5,
|
||||
num_inference_steps=100, width=1024, height=1024,
|
||||
)
|
||||
image.save("image_with_lora.jpg")
|
||||
```
|
||||
59
docs/source/finetune/train_sd_lora.md
Normal file
59
docs/source/finetune/train_sd_lora.md
Normal file
@@ -0,0 +1,59 @@
|
||||
# 训练 Stable Diffusion LoRA
|
||||
|
||||
训练脚本只需要一个文件。我们支持 [CivitAI](https://civitai.com/) 中的主流检查点。默认情况下，我们使用基础的 Stable Diffusion v1.5。你可以从 [HuggingFace](https://huggingface.co/runwayml/stable-diffusion-v1-5/resolve/main/v1-5-pruned-emaonly.safetensors) 或 [ModelScope](https://www.modelscope.cn/models/AI-ModelScope/stable-diffusion-v1-5/resolve/master/v1-5-pruned-emaonly.safetensors) 下载。你可以使用以下代码下载这个文件：
|
||||
|
||||
```python
|
||||
from diffsynth import download_models
|
||||
|
||||
download_models(["StableDiffusion_v15"])
|
||||
```
|
||||
|
||||
```
|
||||
models/stable_diffusion
|
||||
├── Put Stable Diffusion checkpoints here.txt
|
||||
└── v1-5-pruned-emaonly.safetensors
|
||||
```
|
||||
|
||||
使用以下命令启动训练任务：
|
||||
|
||||
```
|
||||
CUDA_VISIBLE_DEVICES="0" python examples/train/stable_diffusion/train_sd_lora.py \
|
||||
--pretrained_path models/stable_diffusion/v1-5-pruned-emaonly.safetensors \
|
||||
--dataset_path data/dog \
|
||||
--output_path ./models \
|
||||
--max_epochs 1 \
|
||||
--steps_per_epoch 500 \
|
||||
--height 512 \
|
||||
--width 512 \
|
||||
--center_crop \
|
||||
--precision "16-mixed" \
|
||||
--learning_rate 1e-4 \
|
||||
--lora_rank 4 \
|
||||
--lora_alpha 4 \
|
||||
--use_gradient_checkpointing
|
||||
```
|
||||
|
||||
有关参数的更多信息，请使用 `python examples/train/stable_diffusion/train_sd_lora.py -h` 查看详细信息。
|
||||
|
||||
训练完成后，使用 `model_manager.load_lora` 加载 LoRA 以进行推理。
|
||||
|
||||
|
||||
|
||||
```python
|
||||
from diffsynth import ModelManager, SDImagePipeline
|
||||
import torch
|
||||
|
||||
model_manager = ModelManager(torch_dtype=torch.float16, device="cuda",
|
||||
file_path_list=["models/stable_diffusion/v1-5-pruned-emaonly.safetensors"])
|
||||
model_manager.load_lora("models/lightning_logs/version_0/checkpoints/epoch=0-step=500.ckpt", lora_alpha=1.0)
|
||||
pipe = SDImagePipeline.from_model_manager(model_manager)
|
||||
|
||||
torch.manual_seed(0)
|
||||
image = pipe(
|
||||
prompt="a dog is jumping, flowers around the dog, the background is mountains and clouds",
|
||||
negative_prompt="bad quality, poor quality, doll, disfigured, jpg, toy, bad anatomy, missing limbs, missing fingers, 3d, cgi, extra tails",
|
||||
cfg_scale=7.5,
|
||||
num_inference_steps=100, width=512, height=512,
|
||||
)
|
||||
image.save("image_with_lora.jpg")
|
||||
```
|
||||
57
docs/source/finetune/train_sdxl_lora.md
Normal file
57
docs/source/finetune/train_sdxl_lora.md
Normal file
@@ -0,0 +1,57 @@
|
||||
# 训练 Stable Diffusion XL LoRA
|
||||
|
||||
训练脚本只需要一个文件。我们支持 [CivitAI](https://civitai.com/) 中的主流检查点。默认情况下，我们使用基础的 Stable Diffusion XL。你可以从 [HuggingFace](https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0/resolve/main/sd_xl_base_1.0.safetensors) 或 [ModelScope](https://www.modelscope.cn/models/AI-ModelScope/stable-diffusion-xl-base-1.0/resolve/master/sd_xl_base_1.0.safetensors) 下载。也可以使用以下代码下载这个文件：
|
||||
|
||||
```python
|
||||
from diffsynth import download_models
|
||||
|
||||
download_models(["StableDiffusionXL_v1"])
|
||||
```
|
||||
|
||||
```
|
||||
models/stable_diffusion_xl
|
||||
├── Put Stable Diffusion XL checkpoints here.txt
|
||||
└── sd_xl_base_1.0.safetensors
|
||||
```
|
||||
|
||||
我们观察到 Stable Diffusion XL 在 float16 精度下会出现数值精度溢出，因此我们建议用户使用 float32 精度训练，使用以下命令启动训练任务：
|
||||
|
||||
```
|
||||
CUDA_VISIBLE_DEVICES="0" python examples/train/stable_diffusion_xl/train_sdxl_lora.py \
|
||||
--pretrained_path models/stable_diffusion_xl/sd_xl_base_1.0.safetensors \
|
||||
--dataset_path data/dog \
|
||||
--output_path ./models \
|
||||
--max_epochs 1 \
|
||||
--steps_per_epoch 500 \
|
||||
--height 1024 \
|
||||
--width 1024 \
|
||||
--center_crop \
|
||||
--precision "32" \
|
||||
--learning_rate 1e-4 \
|
||||
--lora_rank 4 \
|
||||
--lora_alpha 4 \
|
||||
--use_gradient_checkpointing
|
||||
```
|
||||
|
||||
有关参数的更多信息，请使用 `python examples/train/stable_diffusion_xl/train_sdxl_lora.py -h` 查看详细信息。
|
||||
|
||||
训练完成后，使用 `model_manager.load_lora` 加载 LoRA 以进行推理。
|
||||
|
||||
```python
|
||||
from diffsynth import ModelManager, SDXLImagePipeline
|
||||
import torch
|
||||
|
||||
model_manager = ModelManager(torch_dtype=torch.float16, device="cuda",
|
||||
file_path_list=["models/stable_diffusion_xl/sd_xl_base_1.0.safetensors"])
|
||||
model_manager.load_lora("models/lightning_logs/version_0/checkpoints/epoch=0-step=500.ckpt", lora_alpha=1.0)
|
||||
pipe = SDXLImagePipeline.from_model_manager(model_manager)
|
||||
|
||||
torch.manual_seed(0)
|
||||
image = pipe(
|
||||
prompt="a dog is jumping, flowers around the dog, the background is mountains and clouds",
|
||||
negative_prompt="bad quality, poor quality, doll, disfigured, jpg, toy, bad anatomy, missing limbs, missing fingers, 3d, cgi, extra tails",
|
||||
cfg_scale=7.5,
|
||||
num_inference_steps=100, width=1024, height=1024,
|
||||
)
|
||||
image.save("image_with_lora.jpg")
|
||||
```
|
||||
@@ -6,28 +6,30 @@
|
||||
DiffSynth-Studio 文档
|
||||
==============================
|
||||
|
||||
Add your content using ``reStructuredText`` syntax. See the
|
||||
`reStructuredText <https://www.sphinx-doc.org/en/master/usage/restructuredtext/index.html>`_
|
||||
documentation for details.
|
||||
|
||||
欢迎来到 DiffSynth-Studio，我们旨在构建 Diffusion 模型的开源互联生态，在这里，你可以体验到 AIGC（AI Generated Content）技术魔法般的魅力！
|
||||
|
||||
.. toctree::
|
||||
:maxdepth: 1
|
||||
:caption: Contents:
|
||||
|
||||
GetStarted/A_simple_example.md
|
||||
GetStarted/Download_models.md
|
||||
GetStarted/ModelManager.md
|
||||
GetStarted/Models.md
|
||||
GetStarted/Pipelines.md
|
||||
GetStarted/PromptProcessing.md
|
||||
GetStarted/Schedulers.md
|
||||
GetStarted/Fine-tuning.md
|
||||
GetStarted/Extensions.md
|
||||
GetStarted/WebUI.md
|
||||
|
||||
:caption: 快速开始
|
||||
|
||||
tutorial/ASimpleExample.md
|
||||
tutorial/Installation.md
|
||||
tutorial/DownloadModels.md
|
||||
tutorial/Models.md
|
||||
tutorial/Pipelines.md
|
||||
tutorial/PromptProcessing.md
|
||||
tutorial/Extensions.md
|
||||
tutorial/Schedulers.md
|
||||
|
||||
.. toctree::
|
||||
:maxdepth: 1
|
||||
:caption: API Docs
|
||||
:caption: 微调
|
||||
|
||||
finetune/overview.md
|
||||
finetune/train_flux_lora.md
|
||||
finetune/train_kolors_lora.md
|
||||
finetune/train_sd3_lora.md
|
||||
finetune/train_hunyuan_dit_lora.md
|
||||
finetune/train_sdxl_lora.md
|
||||
finetune/train_sd_lora.md
|
||||
|
||||
|
||||
81
docs/source/tutorial/ASimpleExample.md
Normal file
81
docs/source/tutorial/ASimpleExample.md
Normal file
@@ -0,0 +1,81 @@
|
||||
# 快速开始
|
||||
|
||||
在这篇文档中，我们通过一段代码为你介绍如何快速上手使用 DiffSynth-Studio 进行创作。
|
||||
|
||||
## 安装
|
||||
|
||||
使用以下命令从 GitHub 克隆并安装 DiffSynth-Studio。更多信息请参考[安装](./Installation.md)。
|
||||
|
||||
```shell
|
||||
git clone https://github.com/modelscope/DiffSynth-Studio.git
|
||||
cd DiffSynth-Studio
|
||||
pip install -e .
|
||||
```
|
||||
|
||||
## 下载模型
|
||||
|
||||
我们在 DiffSynth-Studio 中预置了一些主流 Diffusion 模型的下载链接，你可以直接使用 `download_models` 函数下载预置的模型文件。
|
||||
|
||||
```python
|
||||
from diffsynth import download_models
|
||||
|
||||
download_models(["FLUX.1-dev"])
|
||||
```
|
||||
|
||||
我们支持从 [ModelScope](https://www.modelscope.cn/) 和 [HuggingFace](https://huggingface.co/) 下载模型，也支持下载非预置的模型，请参考[模型下载](./DownloadModels.md)。
|
||||
|
||||
## 加载模型
|
||||
|
||||
在 DiffSynth-Studio 中，模型由统一的 `ModelManager` 维护。以 FLUX.1-dev 模型为例，模型包括两个文本编码器、一个 DiT、一个 VAE，使用方式如下所示：
|
||||
|
||||
```python
|
||||
import torch
|
||||
from diffsynth import ModelManager
|
||||
|
||||
model_manager = ModelManager(torch_dtype=torch.bfloat16, device="cuda")
|
||||
model_manager.load_models([
|
||||
"models/FLUX/FLUX.1-dev/text_encoder/model.safetensors",
|
||||
"models/FLUX/FLUX.1-dev/text_encoder_2",
|
||||
"models/FLUX/FLUX.1-dev/ae.safetensors",
|
||||
"models/FLUX/FLUX.1-dev/flux1-dev.safetensors"
|
||||
])
|
||||
```
|
||||
|
||||
你可以把所有想要加载的模型路径放入其中。对于 `.safetensors` 等格式的模型权重文件，`ModelManager` 在加载后会自动判断模型类型；对于文件夹格式的模型，`ModelManager` 会尝试解析其中的 `config.json` 文件并尝试调用 `transformers` 等第三方库中的对应模块。关于 DiffSynth-Studio 支持的模型，请参考[支持的模型](./Models.md)。
|
||||
|
||||
## 构建 Pipeline
|
||||
|
||||
DiffSynth-Studio 提供了多个推理 `Pipeline`，这些 `Pipeline` 可以直接通过 `ModelManager` 获取所需的模型并初始化。例如，FLUX.1-dev 模型的文生图 `Pipeline` 可以这样构建：
|
||||
|
||||
```python
|
||||
pipe = FluxImagePipeline.from_model_manager(model_manager)
|
||||
```
|
||||
|
||||
更多用于图像生成和视频生成的 `Pipeline` 详见[推理流水线](./Pipelines.md)。
|
||||
|
||||
## 生成！
|
||||
|
||||
写好你的提示词，交给 DiffSynth-Studio，启动生成任务吧！
|
||||
|
||||
```python
|
||||
import torch
|
||||
from diffsynth import ModelManager, FluxImagePipeline
|
||||
|
||||
model_manager = ModelManager(torch_dtype=torch.bfloat16, device="cuda")
|
||||
model_manager.load_models([
|
||||
"models/FLUX/FLUX.1-dev/text_encoder/model.safetensors",
|
||||
"models/FLUX/FLUX.1-dev/text_encoder_2",
|
||||
"models/FLUX/FLUX.1-dev/ae.safetensors",
|
||||
"models/FLUX/FLUX.1-dev/flux1-dev.safetensors"
|
||||
])
|
||||
pipe = FluxImagePipeline.from_model_manager(model_manager)
|
||||
|
||||
torch.manual_seed(0)
|
||||
image = pipe(
|
||||
prompt="In a forest, a wooden plank sign reading DiffSynth",
|
||||
height=576, width=1024
|
||||
)
|
||||
image.save("image.jpg")
|
||||
```
|
||||
|
||||
|
||||
30
docs/source/tutorial/DownloadModels.md
Normal file
30
docs/source/tutorial/DownloadModels.md
Normal file
@@ -0,0 +1,30 @@
|
||||
# 下载模型
|
||||
|
||||
我们在 DiffSynth-Studio 中预置了一些主流 Diffusion 模型的下载链接，你可以轻松地下载并使用这些模型。
|
||||
|
||||
## 下载预置模型
|
||||
|
||||
你可以直接使用 `download_models` 函数下载预置的模型文件，其中模型 ID 可参考 [config file](/diffsynth/configs/model_config.py)。
|
||||
|
||||
```python
|
||||
from diffsynth import download_models
|
||||
|
||||
download_models(["FLUX.1-dev"])
|
||||
```
|
||||
|
||||
对于 VSCode 用户，激活 Pylance 或其他 Python 语言服务后，在代码中输入 `""` 即可显示支持的所有模型 ID。
|
||||
|
||||
|
||||
|
||||
## 下载非预置模型
|
||||
|
||||
你可以选择 [ModelScope](https://modelscope.cn/models) 和 [HuggingFace](https://huggingface.co/models) 两个下载源中的模型。当然，你也可以通过浏览器等工具选择手动下载自己所需的模型。
|
||||
|
||||
```python
|
||||
from diffsynth.models.downloader import download_from_huggingface, download_from_modelscope
|
||||
|
||||
# From Modelscope (recommended)
|
||||
download_from_modelscope("Kwai-Kolors/Kolors", "vae/diffusion_pytorch_model.fp16.bin", "models/kolors/Kolors/vae")
|
||||
# From Huggingface
|
||||
download_from_huggingface("Kwai-Kolors/Kolors", "vae/diffusion_pytorch_model.fp16.safetensors", "models/kolors/Kolors/vae")
|
||||
```
|
||||
@@ -1,5 +1,7 @@
|
||||
# 安装
|
||||
|
||||
目前，DiffSynth-Studio 支持从 GitHub 克隆安装或使用 pip 安装，我们建议用户从 GitHub 克隆安装，从而体验最新的功能。
|
||||
|
||||
## 从源码下载
|
||||
|
||||
1. 克隆源码仓库：
|
||||
@@ -2,6 +2,7 @@
|
||||
|
||||
目前为止，DiffSynth Studio 支持的模型如下所示：
|
||||
|
||||
* [CogVideo](https://huggingface.co/THUDM/CogVideoX-5b)
|
||||
* [FLUX](https://huggingface.co/black-forest-labs/FLUX.1-dev)
|
||||
* [ExVideo](https://huggingface.co/ECNU-CILab/ExVideo-SVD-128f-v1)
|
||||
* [Kolors](https://huggingface.co/Kwai-Kolors/Kolors)
|
||||
@@ -1,27 +1,22 @@
|
||||
# Pipelines
|
||||
# 流水线
|
||||
|
||||
So far, the following table lists our pipelines and the models supported by each pipeline.
|
||||
DiffSynth-Studio 中包括多个流水线，分为图像生成和视频生成两类。
|
||||
|
||||
## Image Pipelines
|
||||
|
||||
Pipelines for generating images from text descriptions. Each pipeline relies on specific encoder and decoder models.
|
||||
## 图像生成流水线
|
||||
|
||||
| Pipeline | Models |
|
||||
|----------------------------|----------------------------------------------------------------|
|
||||
| HunyuanDiTImagePipeline | text_encoder: HunyuanDiTCLIPTextEncoder<br>text_encoder_t5: HunyuanDiTT5TextEncoder<br>dit: HunyuanDiT<br>vae_decoder: SDVAEDecoder<br>vae_encoder: SDVAEEncoder |
|
||||
| SDImagePipeline | text_encoder: SDTextEncoder<br>unet: SDUNet<br>vae_decoder: SDVAEDecoder<br>vae_encoder: SDVAEEncoder<br>controlnet: MultiControlNetManager<br>ipadapter_image_encoder: IpAdapterCLIPImageEmbedder<br>ipadapter: SDIpAdapter |
|
||||
| SD3ImagePipeline | text_encoder_1: SD3TextEncoder1<br>text_encoder_2: SD3TextEncoder2<br>text_encoder_3: SD3TextEncoder3<br>dit: SD3DiT<br>vae_decoder: SD3VAEDecoder<br>vae_encoder: SD3VAEEncoder |
|
||||
| SDXLImagePipeline | text_encoder: SDXLTextEncoder<br>text_encoder_2: SDXLTextEncoder2<br>text_encoder_kolors: ChatGLMModel<br>unet: SDXLUNet<br>vae_decoder: SDXLVAEDecoder<br>vae_encoder: SDXLVAEEncoder<br>controlnet: MultiControlNetManager<br>ipadapter_image_encoder: IpAdapterXLCLIPImageEmbedder<br>ipadapter: SDXLIpAdapter |
|
||||
| SD3ImagePipeline | text_encoder_1: SD3TextEncoder1<br>text_encoder_2: SD3TextEncoder2<br>text_encoder_3: SD3TextEncoder3<br>dit: SD3DiT<br>vae_decoder: SD3VAEDecoder<br>vae_encoder: SD3VAEEncoder |
|
||||
| HunyuanDiTImagePipeline | text_encoder: HunyuanDiTCLIPTextEncoder<br>text_encoder_t5: HunyuanDiTT5TextEncoder<br>dit: HunyuanDiT<br>vae_decoder: SDVAEDecoder<br>vae_encoder: SDVAEEncoder |
|
||||
| FluxImagePipeline | text_encoder_1: FluxTextEncoder1<br>text_encoder_2: FluxTextEncoder2<br>dit: FluxDiT<br>vae_decoder: FluxVAEDecoder<br>vae_encoder: FluxVAEEncoder |
|
||||
|
||||
## Video Pipelines
|
||||
|
||||
Pipelines for generating videos from text descriptions. In addition to the models required for image generation, they include models for handling motion modules.
|
||||
## 视频生成流水线
|
||||
|
||||
| Pipeline | Models |
|
||||
|----------------------------|----------------------------------------------------------------|
|
||||
| SDVideoPipeline | text_encoder: SDTextEncoder<br>unet: SDUNet<br>vae_decoder: SDVAEDecoder<br>vae_encoder: SDVAEEncoder<br>controlnet: MultiControlNetManager<br>ipadapter_image_encoder: IpAdapterCLIPImageEmbedder<br>ipadapter: SDIpAdapter<br>motion_modules: SDMotionModel |
|
||||
| SDXLVideoPipeline | text_encoder: SDXLTextEncoder<br>text_encoder_2: SDXLTextEncoder2<br>text_encoder_kolors: ChatGLMModel<br>unet: SDXLUNet<br>vae_decoder: SDXLVAEDecoder<br>vae_encoder: SDXLVAEEncoder<br>ipadapter_image_encoder: IpAdapterXLCLIPImageEmbedder<br>ipadapter: SDXLIpAdapter<br>motion_modules: SDXLMotionModel |
|
||||
| SVDVideoPipeline | image_encoder: SVDImageEncoder<br>unet: SVDUNet<br>vae_encoder: SVDVAEEncoder<br>vae_decoder: SVDVAEDecoder |
|
||||
|
||||
|
||||
|
||||
| CogVideoPipeline | text_encoder: FluxTextEncoder2<br>dit: CogDiT<br>vae_encoder: CogVAEEncoder<br>vae_decoder: CogVAEDecoder |
|
||||
@@ -1,4 +1,4 @@
|
||||
# 提示词（Prompt）处理
|
||||
# 提示词处理
|
||||
|
||||
DiffSynth 内置了提示词处理功能，分为：
|
||||
|
||||
Reference in New Issue
Block a user