chunzhi/stable-diffusion-cli-on-modal
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Update some dependencies. Repair some codes.

Browse Source
This commit is contained in:
hodanov 2024-04-21 12:40:44 +09:00
parent 4fe518038d
commit e1639039d7
8 changed files with 63 additions and 65 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
app/Dockerfile
View File

@ -5,7 +5,7 @@ RUN apt-get update \
&& apt-get autoremove -y \ && apt-get autoremove -y \
&& apt-get clean -y \ && apt-get clean -y \
&& rm -rf /var/lib/apt/lists/* \ && rm -rf /var/lib/apt/lists/* \
&& pip install -r requirements.txt --extra-index-url https://download.pytorch.org/whl/cu117 --no-cache-dir \ && pip install -r requirements.txt --extra-index-url https://download.pytorch.org/whl/cu121 --no-cache-dir \
&& mkdir -p /vol/cache/esrgan \ && mkdir -p /vol/cache/esrgan \
&& wget --progress=dot:giga https://github.com/xinntao/Real-ESRGAN/releases/download/v0.1.0/RealESRGAN_x4plus.pth -P /vol/cache/esrgan \ && wget --progress=dot:giga https://github.com/xinntao/Real-ESRGAN/releases/download/v0.1.0/RealESRGAN_x4plus.pth -P /vol/cache/esrgan \
&& wget --progress=dot:giga https://github.com/xinntao/Real-ESRGAN/releases/download/v0.1.1/RealESRNet_x4plus.pth -P /vol/cache/esrgan \ && wget --progress=dot:giga https://github.com/xinntao/Real-ESRGAN/releases/download/v0.1.1/RealESRNet_x4plus.pth -P /vol/cache/esrgan \

4
app/__main__.py
View File

@ -2,10 +2,10 @@ from __future__ import annotations
import stable_diffusion_1_5 import stable_diffusion_1_5
import stable_diffusion_xl import stable_diffusion_xl
from setup import stub from setup import app
@stub.function(gpu="A10G") @app.function(gpu="A10G")
def main(): def main():
stable_diffusion_1_5.SD15 stable_diffusion_1_5.SD15
stable_diffusion_xl.SDXLTxt2Img stable_diffusion_xl.SDXLTxt2Img

16
app/requirements.txt
View File

@ -1,21 +1,21 @@
invisible_watermark invisible_watermark
accelerate accelerate
diffusers[torch]==0.24.0 diffusers[torch]==0.27.2
onnxruntime==1.16.3 onnxruntime==1.17.3
safetensors==0.4.1 safetensors==0.4.3
torch==2.1.0 torch==2.2.2
transformers==4.39.1 transformers==4.39.3
xformers==0.0.22.post7 xformers==0.0.25.post1
realesrgan==0.3.0 realesrgan==0.3.0
basicsr>=1.4.2 basicsr>=1.4.2
facexlib>=0.3.0 facexlib>=0.3.0
gfpgan>=1.3.8 gfpgan>=1.3.8
scipy==1.12.0 scipy==1.13.0
opencv-python opencv-python
Pillow Pillow
pillow-avif-plugin pillow-avif-plugin
torchvision torchvision==0.17.2
tqdm tqdm
controlnet_aux controlnet_aux

8
app/setup.py
View File

@ -3,7 +3,7 @@ from __future__ import annotations
import os import os
import diffusers import diffusers
from modal import Image, Mount, Secret, Stub from modal import App, Image, Mount, Secret
BASE_CACHE_PATH = "/vol/cache" BASE_CACHE_PATH = "/vol/cache"
BASE_CACHE_PATH_LORA = "/vol/cache/lora" BASE_CACHE_PATH_LORA = "/vol/cache/lora"
@ -58,7 +58,7 @@ def download_model(name: str, model_url: str, token: str):
cache_path = os.path.join(BASE_CACHE_PATH, name) cache_path = os.path.join(BASE_CACHE_PATH, name)
pipe = diffusers.StableDiffusionPipeline.from_single_file( pipe = diffusers.StableDiffusionPipeline.from_single_file(
pretrained_model_link_or_path=model_url, pretrained_model_link_or_path=model_url,
use_auth_token=token, token=token,
cache_dir=cache_path, cache_dir=cache_path,
) )
pipe.save_pretrained(cache_path, safe_serialization=True) pipe.save_pretrained(cache_path, safe_serialization=True)
@ -131,12 +131,12 @@ def build_image():
) )
stub = Stub("stable-diffusion-cli") app = App("stable-diffusion-cli")
base_stub = Image.from_dockerfile( base_stub = Image.from_dockerfile(
path="Dockerfile", path="Dockerfile",
context_mount=Mount.from_local_file("requirements.txt"), context_mount=Mount.from_local_file("requirements.txt"),
) )
stub.image = base_stub.dockerfile_commands( app.image = base_stub.dockerfile_commands(
"FROM base", "FROM base",
"COPY config.yml /", "COPY config.yml /",
context_mount=Mount.from_local_file("config.yml"), context_mount=Mount.from_local_file("config.yml"),

23
app/stable_diffusion_1_5.py
View File

@ -10,11 +10,11 @@ from setup import (
BASE_CACHE_PATH_CONTROLNET, BASE_CACHE_PATH_CONTROLNET,
BASE_CACHE_PATH_LORA, BASE_CACHE_PATH_LORA,
BASE_CACHE_PATH_TEXTUAL_INVERSION, BASE_CACHE_PATH_TEXTUAL_INVERSION,
stub, app,
) )
@stub.cls( @app.cls(
gpu="A10G", gpu="A10G",
secrets=[Secret.from_dotenv(__file__)], secrets=[Secret.from_dotenv(__file__)],
) )
@ -187,15 +187,16 @@ class SD15:
generated_images.extend(fixed_by_controlnet) generated_images.extend(fixed_by_controlnet)
base_images = fixed_by_controlnet base_images = fixed_by_controlnet
if upscaler != "": # TODO: Upscaler stopped working due to update of dependent packages. Replace with diffusers upscaler.
upscaled = self._upscale( # if upscaler != "":
base_images=base_images, # upscaled = self._upscale(
half_precision=False, # base_images=base_images,
tile=700, # half_precision=False,
upscaler=upscaler, # tile=700,
use_face_enhancer=use_face_enhancer, # upscaler=upscaler,
) # use_face_enhancer=use_face_enhancer,
generated_images.extend(upscaled) # )
# generated_images.extend(upscaled)
image_output = [] image_output = []
for image in generated_images: for image in generated_images:

55
app/stable_diffusion_xl.py
View File

@ -5,10 +5,10 @@ import os
import PIL.Image import PIL.Image
from modal import Secret, enter, method from modal import Secret, enter, method
from setup import BASE_CACHE_PATH, BASE_CACHE_PATH_CONTROLNET, stub from setup import BASE_CACHE_PATH, app
@stub.cls( @app.cls(
gpu="A10G", gpu="A10G",
secrets=[Secret.from_dotenv(__file__)], secrets=[Secret.from_dotenv(__file__)],
) )
@ -39,13 +39,13 @@ class SDXLTxt2Img:
variant="fp16", variant="fp16",
) )
# self.refiner_cache_path = self.cache_path + "-refiner" self.refiner_cache_path = self.cache_path + "-refiner"
# self.refiner = diffusers.StableDiffusionXLImg2ImgPipeline.from_pretrained( self.refiner = diffusers.StableDiffusionXLImg2ImgPipeline.from_pretrained(
# self.refiner_cache_path, self.refiner_cache_path,
# torch_dtype=torch.float16, torch_dtype=torch.float16,
# use_safetensors=True, use_safetensors=True,
# variant="fp16", variant="fp16",
# ) )
# controlnets = config.get("controlnets") # controlnets = config.get("controlnets")
# if controlnets is not None: # if controlnets is not None:
@ -94,12 +94,10 @@ class SDXLTxt2Img:
n_prompt: str, n_prompt: str,
height: int = 1024, height: int = 1024,
width: int = 1024, width: int = 1024,
batch_size: int = 1,
steps: int = 30, steps: int = 30,
seed: int = 1, seed: int = 1,
upscaler: str = "", upscaler: str = "",
use_face_enhancer: bool = False, use_face_enhancer: bool = False,
fix_by_controlnet_tile: bool = False,
output_format: str = "png", output_format: str = "png",
) -> list[bytes]: ) -> list[bytes]:
""" """
@ -119,37 +117,33 @@ class SDXLTxt2Img:
).images ).images
base_images = generated_images base_images = generated_images
# for image in base_images: for image in base_images:
# image = self._resize_image(image=image, scale_factor=2) image = self._resize_image(image=image, scale_factor=2)
# self.refiner.to("cuda") self.refiner.to("cuda")
# refined_images = self.refiner( refined_images = self.refiner(
# prompt=prompt, prompt=prompt,
# negative_prompt=n_prompt, negative_prompt=n_prompt,
# num_inference_steps=steps, num_inference_steps=steps,
# strength=0.1, strength=0.1,
# # guidance_scale=7.5, # guidance_scale=7.5,
# generator=generator, generator=generator,
# image=image, image=image,
# ).images ).images
# generated_images.extend(refined_images) generated_images.extend(refined_images)
# base_images = refined_images base_images = refined_images
""" """
Fix the generated images by the control_v11f1e_sd15_tile when `fix_by_controlnet_tile` is `True`. Fix the generated images by the control_v11f1e_sd15_tile when `fix_by_controlnet_tile` is `True`.
https://huggingface.co/lllyasviel/control_v11f1e_sd15_tile https://huggingface.co/lllyasviel/control_v11f1e_sd15_tile
""" """
# if fix_by_controlnet_tile: # if fix_by_controlnet_tile:
# max_embeddings_multiples = self._count_token(p=prompt, n=n_prompt) # max_embeddings_multiples = self._count_token(p=prompt, n=n_prompt)
# print("========================確認用========================")
# print("Step1")
# self.controlnet_pipe.to("cuda") # self.controlnet_pipe.to("cuda")
# self.controlnet_pipe.enable_vae_tiling() # self.controlnet_pipe.enable_vae_tiling()
# self.controlnet_pipe.enable_xformers_memory_efficient_attention() # self.controlnet_pipe.enable_xformers_memory_efficient_attention()
# print("Step2")
# for image in base_images: # for image in base_images:
# image = self._resize_image(image=image, scale_factor=2) # image = self._resize_image(image=image, scale_factor=2)
# print("Step3")
# with torch.autocast("cuda"): # with torch.autocast("cuda"):
# print("Step4")
# fixed_by_controlnet = self.controlnet_pipe( # fixed_by_controlnet = self.controlnet_pipe(
# prompt=prompt * batch_size, # prompt=prompt * batch_size,
# negative_prompt=n_prompt * batch_size, # negative_prompt=n_prompt * batch_size,
@ -160,7 +154,6 @@ class SDXLTxt2Img:
# generator=generator, # generator=generator,
# image=image, # image=image,
# ).images # ).images
# print("Step5")
# generated_images.extend(fixed_by_controlnet) # generated_images.extend(fixed_by_controlnet)
# base_images = fixed_by_controlnet # base_images = fixed_by_controlnet

8
cmd/sd15_txt2img.py
View File

@ -3,11 +3,11 @@ import time
import modal import modal
import util import util
stub = modal.Stub("run-stable-diffusion-cli") app = modal.App("run-stable-diffusion-cli")
stub.run_inference = modal.Function.from_name("stable-diffusion-cli", "SD15.run_txt2img_inference") app.run_inference = modal.Function.from_name("stable-diffusion-cli", "SD15.run_txt2img_inference")
@stub.local_entrypoint() @app.local_entrypoint()
def main( def main(
prompt: str, prompt: str,
n_prompt: str, n_prompt: str,
@ -33,7 +33,7 @@ def main(
if seed == -1: if seed == -1:
seed_generated = util.generate_seed() seed_generated = util.generate_seed()
start_time = time.time() start_time = time.time()
images = stub.run_inference.remote( images = app.run_inference.remote(
prompt=prompt, prompt=prompt,
n_prompt=n_prompt, n_prompt=n_prompt,
height=height, height=height,

12
cmd/sdxl_txt2img.py
View File

@ -3,16 +3,18 @@ import time
import modal import modal
import util import util
stub = modal.Stub("run-stable-diffusion-cli") app = modal.Stub("run-stable-diffusion-cli")
stub.run_inference = modal.Function.from_name("stable-diffusion-cli", "SDXLTxt2Img.run_inference") app.run_inference = modal.Function.from_name("stable-diffusion-cli", "SDXLTxt2Img.run_inference")
@stub.local_entrypoint() @app.local_entrypoint()
def main( def main(
prompt: str, prompt: str,
n_prompt: str,
height: int = 1024, height: int = 1024,
width: int = 1024, width: int = 1024,
samples: int = 5, samples: int = 5,
steps: int = 20,
seed: int = -1, seed: int = -1,
upscaler: str = "", upscaler: str = "",
use_face_enhancer: str = "False", use_face_enhancer: str = "False",
@ -29,10 +31,12 @@ def main(
if seed == -1: if seed == -1:
seed_generated = util.generate_seed() seed_generated = util.generate_seed()
start_time = time.time() start_time = time.time()
images = stub.run_inference.remote( images = app.run_inference.remote(
prompt=prompt, prompt=prompt,
n_prompt=n_prompt,
height=height, height=height,
width=width, width=width,
steps=steps,
seed=seed_generated, seed=seed_generated,
upscaler=upscaler, upscaler=upscaler,
use_face_enhancer=use_face_enhancer == "True", use_face_enhancer=use_face_enhancer == "True",