Create handler.py
Browse files- handler.py +48 -0
handler.py
ADDED
@@ -0,0 +1,48 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from typing import Dict
|
2 |
+
import torch
|
3 |
+
from diffusers import StableDiffusionXLPipeline
|
4 |
+
from io import BytesIO
|
5 |
+
import base64
|
6 |
+
|
7 |
+
class EndpointHandler:
|
8 |
+
def __init__(self, path: str = ""):
|
9 |
+
print(f"Initializing SDXL model from: {path}")
|
10 |
+
|
11 |
+
# Base SDXL Model
|
12 |
+
self.pipe = StableDiffusionXLPipeline.from_pretrained(
|
13 |
+
"stabilityai/stable-diffusion-xl-base-1.0",
|
14 |
+
torch_dtype=torch.float16,
|
15 |
+
variant="fp16"
|
16 |
+
)
|
17 |
+
|
18 |
+
print("Loading LoRA weights from: Texttra/Bh0r")
|
19 |
+
self.pipe.load_lora_weights("Texttra/Bh0r", weight_name="Bh0r-10.safetensors")
|
20 |
+
self.pipe.fuse_lora()
|
21 |
+
|
22 |
+
self.pipe.to("cuda" if torch.cuda.is_available() else "cpu")
|
23 |
+
print("Model ready.")
|
24 |
+
|
25 |
+
def __call__(self, data: Dict) -> Dict:
|
26 |
+
print("Received data:", data)
|
27 |
+
|
28 |
+
inputs = data.get("inputs", {})
|
29 |
+
prompt = inputs.get("prompt", "")
|
30 |
+
print("Extracted prompt:", prompt)
|
31 |
+
|
32 |
+
if not prompt:
|
33 |
+
return {"error": "No prompt provided."}
|
34 |
+
|
35 |
+
image = self.pipe(
|
36 |
+
prompt,
|
37 |
+
num_inference_steps=35,
|
38 |
+
guidance_scale=7.0,
|
39 |
+
).images[0]
|
40 |
+
|
41 |
+
print("Image generated.")
|
42 |
+
|
43 |
+
buffer = BytesIO()
|
44 |
+
image.save(buffer, format="PNG")
|
45 |
+
base64_image = base64.b64encode(buffer.getvalue()).decode("utf-8")
|
46 |
+
print("Returning image.")
|
47 |
+
|
48 |
+
return {"image": base64_image}
|