Stable Diffusion Training¶

You can also check configs/stable_diffusion/README.md file.

Configs¶

All configuration files are placed under the configs/stable_diffusion folder.

Following is the example config fixed from the stable_diffusion_v15_pokemon_blip config file in configs/stable_diffusion/stable_diffusion_v15_pokemon_blip.py:

from mmengine.config import read_base

with read_base():
    from .._base_.datasets.pokemon_blip import *
    from .._base_.default_runtime import *
    from .._base_.models.stable_diffusion_v15 import *
    from .._base_.schedules.stable_diffusion_50e import *

Finetuning the text encoder and UNet¶

The script also allows you to finetune the text_encoder along with the unet.

from mmengine.config import read_base

with read_base():
    from .._base_.datasets.pokemon_blip import *
    from .._base_.default_runtime import *
    from .._base_.models.stable_diffusion_v15 import *
    from .._base_.schedules.stable_diffusion_50e import *

model.update(finetune_text_encoder=True)  # fine tune text encoder

We also provide configs/stable_diffusion/stable_diffusion_v15_textencoder_pokemon_blip.py as a whole config.

Finetuning with Unet EMA¶

The script also allows you to finetune with Unet EMA.

from mmengine.config import read_base

with read_base():
    from .._base_.datasets.pokemon_blip import *
    from .._base_.default_runtime import *
    from .._base_.models.stable_diffusion_v15 import *
    from .._base_.schedules.stable_diffusion_50e import *

custom_hooks = [  # Hook is list, we should write all custom_hooks again.
    dict(type=VisualizationHook, prompt=['yoda pokemon'] * 4),
    dict(type=SDCheckpointHook),
    dict(type=UnetEMAHook, momentum=1e-4, priority='ABOVE_NORMAL')  # setup EMA Hook
]

We also provide configs/stable_diffusion/stable_diffusion_v15_ema_pokemon_blip.py as a whole config.

Finetuning with Min-SNR Weighting Strategy¶

The script also allows you to finetune with Min-SNR Weighting Strategy.

from mmengine.config import read_base

with read_base():
    from .._base_.datasets.pokemon_blip import *
    from .._base_.default_runtime import *
    from .._base_.models.stable_diffusion_v15 import *
    from .._base_.schedules.stable_diffusion_50e import *

model.update(loss=dict(type='SNRL2Loss', snr_gamma=5.0, loss_weight=1.0))  # setup Min-SNR Weighting Strategy

We also provide configs/min_snr_loss/stable_diffusion_v15_snr_pokemon_blip.py as a whole config.

Run training¶

Run train

# single gpu
$ diffengine train ${CONFIG_FILE}
# Example
$ diffengine train stable_diffusion_v15_pokemon_blip

# multi gpus
$ NPROC_PER_NODE=${GPU_NUM} diffengine train ${CONFIG_FILE}

Inference with diffusers¶

Once you have trained a model, specify the path to the saved model and utilize it for inference using the diffusers.pipeline module.

Before inferencing, we should convert weights for diffusers format,

$ diffengine convert ${CONFIG_FILE} ${INPUT_FILENAME} ${OUTPUT_DIR} --save-keys ${SAVE_KEYS}
# Example
$ diffengine convert stable_diffusion_v15_pokemon_blip work_dirs/stable_diffusion_v15_pokemon_blip/epoch_50.pth work_dirs/stable_diffusion_v15_pokemon_blip --save-keys unet

Then we can run inference.

import torch
from diffusers import DiffusionPipeline, UNet2DConditionModel

prompt = 'yoda pokemon'
checkpoint = 'work_dirs/stable_diffusion_v15_pokemon_blip'

unet = UNet2DConditionModel.from_pretrained(
    checkpoint, subfolder='unet', torch_dtype=torch.float16)
pipe = DiffusionPipeline.from_pretrained(
    'runwayml/stable-diffusion-v1-5', unet=unet, torch_dtype=torch.float16)
pipe.to('cuda')

image = pipe(
    prompt,
    num_inference_steps=50,
).images[0]
image.save('demo.png')

Inference Text Encoder and Unet finetuned weight with diffusers¶

Once you have trained a model, specify the path to the saved model and utilize it for inference using the diffusers.pipeline module.

Before inferencing, we should convert weights for diffusers format,

$ diffengine convert ${CONFIG_FILE} ${INPUT_FILENAME} ${OUTPUT_DIR} --save-keys ${SAVE_KEYS}
# Example
$ diffengine convert stable_diffusion_v15_textencoder_pokemon_blip work_dirs/stable_diffusion_v15_textencoder_pokemon_blip/epoch_50.pth work_dirs/stable_diffusion_v15_textencoder_pokemon_blip --save-keys unet text_encoder

Then we can run inference.

import torch
from transformers import CLIPTextModel
from diffusers import DiffusionPipeline, UNet2DConditionModel

prompt = 'yoda pokemon'
checkpoint = 'work_dirs/stable_diffusion_v15_pokemon_blip'

text_encoder = CLIPTextModel.from_pretrained(
            checkpoint,
            subfolder='text_encoder',
            torch_dtype=torch.float16)
unet = UNet2DConditionModel.from_pretrained(
    checkpoint, subfolder='unet', torch_dtype=torch.float16)
pipe = DiffusionPipeline.from_pretrained(
    'runwayml/stable-diffusion-v1-5', unet=unet, text_encoder=text_encoder, torch_dtype=torch.float16)
pipe.to('cuda')

image = pipe(
    prompt,
    num_inference_steps=50,
).images[0]
image.save('demo.png')

Results Example¶

stable_diffusion_v15_pokemon_blip¶

example1

You can check configs/stable_diffusion/README.md for more details.