Upload handler and requirements
Browse files- handler.py +87 -0
- requirements.txt +1 -0
handler.py
ADDED
@@ -0,0 +1,87 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from typing import Dict, List, Any
|
2 |
+
import torch
|
3 |
+
import base64
|
4 |
+
from PIL import Image
|
5 |
+
from io import BytesIO
|
6 |
+
from diffusers import T2IAdapter, StableDiffusionXLAdapterPipeline, AutoencoderKL
|
7 |
+
from controlnet_aux.pidi import PidiNetDetector
|
8 |
+
|
9 |
+
# set device
|
10 |
+
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
|
11 |
+
|
12 |
+
if device.type != 'cuda':
|
13 |
+
raise ValueError("need to run on GPU")
|
14 |
+
|
15 |
+
class EndpointHandler():
|
16 |
+
def __init__(self, path=""):
|
17 |
+
# Preload all the elements you are going to need at inference.
|
18 |
+
# pseudo:
|
19 |
+
# self.model= load_model(path)
|
20 |
+
|
21 |
+
adapter = T2IAdapter.from_pretrained(
|
22 |
+
"Adapter/t2iadapter",
|
23 |
+
subfolder="sketch_sdxl_1.0",
|
24 |
+
torch_dtype=torch.float16,
|
25 |
+
adapter_type="full_adapter_xl"
|
26 |
+
)
|
27 |
+
|
28 |
+
vae = AutoencoderKL.from_pretrained(
|
29 |
+
"madebyollin/sdxl-vae-fp16-fix",
|
30 |
+
torch_dtype=torch.float16,
|
31 |
+
use_safetensors=True
|
32 |
+
)
|
33 |
+
|
34 |
+
self.pipeline = StableDiffusionXLAdapterPipeline.from_pretrained(
|
35 |
+
"stabilityai/stable-diffusion-xl-base-1.0",
|
36 |
+
adapter=adapter,
|
37 |
+
vae=vae,
|
38 |
+
torch_dtype=torch.float16,
|
39 |
+
variant="fp16"
|
40 |
+
).to("cuda")
|
41 |
+
self.pipeline.enable_sequential_cpu_offload()
|
42 |
+
|
43 |
+
self.pidinet = PidiNetDetector.from_pretrained("lllyasviel/Annotators").to("cuda")
|
44 |
+
|
45 |
+
def __call__(self, data: Dict[str, Any]) -> List[Dict[str, Any]]:
|
46 |
+
"""
|
47 |
+
data args:
|
48 |
+
inputs (:obj: `str` | `PIL.Image` | `np.array`)
|
49 |
+
kwargs
|
50 |
+
Return:
|
51 |
+
A :obj:`list` | `dict`: will be serialized and returned
|
52 |
+
"""
|
53 |
+
|
54 |
+
# pseudo
|
55 |
+
# self.model(input)
|
56 |
+
|
57 |
+
# get inputs
|
58 |
+
inputs = data.pop("inputs", "")
|
59 |
+
encoded_image = data.pop("image", None)
|
60 |
+
|
61 |
+
# Decode image and convert to black and white sketch
|
62 |
+
decoded_image = self.decode_base64_image(encoded_image).convert('RGB')
|
63 |
+
sketch_image = self.pidinet(
|
64 |
+
decoded_image,
|
65 |
+
detect_resolution=1024,
|
66 |
+
image_resolution=1024,
|
67 |
+
apply_filter=True
|
68 |
+
).convert('L')
|
69 |
+
|
70 |
+
# sketch_image.save("./output1.png")
|
71 |
+
|
72 |
+
output_image = self.pipeline(
|
73 |
+
prompt=inputs,
|
74 |
+
negative_prompt="extra digit, fewer digits, cropped, worst quality, low quality",
|
75 |
+
image=sketch_image,
|
76 |
+
guidance_scale=7.5,
|
77 |
+
).images[0]
|
78 |
+
|
79 |
+
# output_image.save("./output2.png")
|
80 |
+
return output_image
|
81 |
+
|
82 |
+
# helper to decode input image
|
83 |
+
def decode_base64_image(self, image_string):
|
84 |
+
base64_image = base64.b64decode(image_string)
|
85 |
+
buffer = BytesIO(base64_image)
|
86 |
+
image = Image.open(buffer)
|
87 |
+
return image
|
requirements.txt
ADDED
@@ -0,0 +1 @@
|
|
|
|
|
1 |
+
controlnet-aux
|