Update app.py
Browse files
app.py
CHANGED
@@ -1,102 +1,61 @@
|
|
|
|
1 |
import gradio as gr
|
2 |
-
import torch
|
3 |
-
import base64
|
4 |
-
import io
|
5 |
from PIL import Image
|
6 |
-
from diffusers import
|
7 |
-
from
|
8 |
-
|
9 |
-
|
10 |
-
|
11 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
12 |
|
13 |
-
# Load Base Model and LoRA
|
14 |
-
base_model = "black-forest-labs/FLUX.1-dev"
|
15 |
-
lora_path = "checkpoints/models/Ghibli.safetensors"
|
16 |
-
|
17 |
-
# Load the main pipeline
|
18 |
-
pipe = FluxPipeline.from_pretrained(base_model, torch_dtype=torch.float16)
|
19 |
-
transformer = FluxTransformer2DModel.from_pretrained(base_model, subfolder="transformer", torch_dtype=torch.float16)
|
20 |
-
pipe.transformer = transformer
|
21 |
-
pipe.to("cuda")
|
22 |
-
|
23 |
-
# Load LoRA
|
24 |
-
set_single_lora(pipe.transformer, lora_path, lora_weights=[1], cond_size=512)
|
25 |
-
|
26 |
-
# Base64 to Image
|
27 |
-
def base64_to_image(base64_str):
|
28 |
-
image_data = base64.b64decode(base64_str)
|
29 |
-
return Image.open(io.BytesIO(image_data)).convert("RGB")
|
30 |
-
|
31 |
-
# Image to Base64
|
32 |
-
def image_to_base64(image):
|
33 |
-
buffered = io.BytesIO()
|
34 |
-
image.save(buffered, format="PNG")
|
35 |
-
return base64.b64encode(buffered.getvalue()).decode()
|
36 |
-
|
37 |
-
# Cartoonizer function
|
38 |
-
def cartoonize_base64(b64_image, prompt="Ghibli Studio style, hand-drawn anime illustration", height=768, width=768, seed=42):
|
39 |
-
input_image = base64_to_image(b64_image)
|
40 |
-
|
41 |
-
generator = torch.Generator(device="cuda").manual_seed(int(seed))
|
42 |
-
|
43 |
-
result = pipe(
|
44 |
-
prompt=prompt,
|
45 |
-
height=int(height),
|
46 |
-
width=int(width),
|
47 |
-
guidance_scale=3.5,
|
48 |
-
num_inference_steps=25,
|
49 |
-
generator=generator,
|
50 |
-
spatial_images=[input_image],
|
51 |
-
cond_size=512
|
52 |
-
).images[0]
|
53 |
-
|
54 |
-
clear_cache(pipe.transformer)
|
55 |
-
|
56 |
-
return image_to_base64(result)
|
57 |
-
|
58 |
-
# Gradio UI function
|
59 |
-
def ui_cartoonize(image, prompt, height, width, seed):
|
60 |
-
buffered = io.BytesIO()
|
61 |
-
image.save(buffered, format="PNG")
|
62 |
-
b64_image = base64.b64encode(buffered.getvalue()).decode()
|
63 |
-
cartoon_b64 = cartoonize_base64(b64_image, prompt, height, width, seed)
|
64 |
-
cartoon_image = base64_to_image(cartoon_b64)
|
65 |
-
return cartoon_image
|
66 |
-
|
67 |
-
# Gradio App
|
68 |
with gr.Blocks() as demo:
|
69 |
-
gr.Markdown("#
|
70 |
-
|
71 |
-
with gr.
|
72 |
-
|
73 |
-
|
74 |
-
|
75 |
-
|
76 |
-
|
77 |
-
|
78 |
-
|
79 |
-
|
80 |
-
|
81 |
-
|
82 |
-
|
83 |
-
|
84 |
-
inputs=[input_image, prompt, height, width, seed],
|
85 |
-
outputs=output_image
|
86 |
-
)
|
87 |
-
|
88 |
-
# Gradio API: Accept base64, return base64
|
89 |
-
gr.Interface(
|
90 |
-
fn=cartoonize_base64,
|
91 |
-
inputs=[
|
92 |
-
gr.Text(label="Base64 Image Input"),
|
93 |
-
gr.Text(label="Prompt"),
|
94 |
-
gr.Number(label="Height", value=768),
|
95 |
-
gr.Number(label="Width", value=768),
|
96 |
-
gr.Number(label="Seed", value=42)
|
97 |
-
],
|
98 |
-
outputs=gr.Text(label="Base64 Cartoon Output"),
|
99 |
-
api_name="predict"
|
100 |
-
)
|
101 |
|
102 |
demo.launch()
|
|
|
1 |
+
# app.py
|
2 |
import gradio as gr
|
3 |
+
import torch, io, base64
|
|
|
|
|
4 |
from PIL import Image
|
5 |
+
from diffusers import StableDiffusionImg2ImgPipeline
|
6 |
+
from vtoonify_model import load_vtoonify # see below
|
7 |
+
|
8 |
+
# Load models
|
9 |
+
pipe_ghibli = StableDiffusionImg2ImgPipeline.from_pretrained(
|
10 |
+
"nitrosocke/Ghibli-Diffusion", torch_dtype=torch.float16
|
11 |
+
).to("cuda") # Ghibli-style fine-tuned SD :contentReference[oaicite:1]{index=1}
|
12 |
+
|
13 |
+
pipe_vtoonify = load_vtoonify().to("cuda") # cartoonization model loader
|
14 |
+
|
15 |
+
# Helpers for base64 conversion
|
16 |
+
def pil_to_b64(img: Image.Image) -> str:
|
17 |
+
buf = io.BytesIO()
|
18 |
+
img.save(buf, format="PNG")
|
19 |
+
return base64.b64encode(buf.getvalue()).decode()
|
20 |
+
|
21 |
+
def b64_to_pil(b64: str) -> Image.Image:
|
22 |
+
data = base64.b64decode(b64)
|
23 |
+
return Image.open(io.BytesIO(data)).convert("RGB")
|
24 |
+
|
25 |
+
# Core processor
|
26 |
+
def run_effect(input_b64: str, effect: str) -> dict:
|
27 |
+
img = b64_to_pil(input_b64)
|
28 |
+
if effect == "ghibli":
|
29 |
+
out = pipe_ghibli(prompt="ghibli style", image=img, strength=0.5, guidance_scale=7.5).images[0]
|
30 |
+
else:
|
31 |
+
out = pipe_vtoonify(img)
|
32 |
+
return {"output_b64": pil_to_b64(out)}
|
33 |
+
|
34 |
+
@gr.utils.decorators.thread_safe()
|
35 |
+
@spaces.GPU # enables GPU on ZeroGPU Infra
|
36 |
+
def api_process(input_b64, effect):
|
37 |
+
return run_effect(input_b64, effect)
|
38 |
+
|
39 |
+
def gradio_process(img: Image.Image, effect: str) -> Image.Image:
|
40 |
+
# Reuse logic, bypass base64
|
41 |
+
in_b64 = pil_to_b64(img)
|
42 |
+
return b64_to_pil(run_effect(in_b64, effect)["output_b64"])
|
43 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
44 |
with gr.Blocks() as demo:
|
45 |
+
gr.Markdown("# Ghibli & VToonify Effects 🎨")
|
46 |
+
|
47 |
+
with gr.Tab("Web UI"):
|
48 |
+
inp = gr.Image(type="pil", label="Upload Image")
|
49 |
+
eff = gr.Radio(["ghibli", "vtoonify"], label="Effect")
|
50 |
+
btn = gr.Button("Apply Effect")
|
51 |
+
out = gr.Image(label="Result")
|
52 |
+
btn.click(gradio_process, [inp, eff], out)
|
53 |
+
|
54 |
+
with gr.Tab("API (base64)"):
|
55 |
+
inp_b64 = gr.Textbox(lines=4, label="Input Image (base64)")
|
56 |
+
eff2 = gr.Radio(["ghibli", "vtoonify"], label="Effect")
|
57 |
+
btn2 = gr.Button("Run API")
|
58 |
+
out_b64 = gr.Textbox(lines=4, label="Output Image (base64)")
|
59 |
+
btn2.click(api_process, [inp_b64, eff2], out_b64)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
60 |
|
61 |
demo.launch()
|