Spaces:
Running
Running
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,84 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
import torch
|
3 |
+
import os
|
4 |
+
import gc
|
5 |
+
import random
|
6 |
+
from huggingface_hub import snapshot_download
|
7 |
+
from diffusers import StableDiffusionXLPipeline, LCMScheduler
|
8 |
+
from PIL import Image
|
9 |
+
|
10 |
+
os.environ["XDG_CACHE_HOME"] = "/home/user/.cache"
|
11 |
+
os.environ["TRANSFORMERS_CACHE"] = "/home/user/.cache/huggingface/transformers"
|
12 |
+
os.environ["HF_HOME"] = "/home/user/.cache/huggingface"
|
13 |
+
models = [
|
14 |
+
"Niggendar/fastPhotoPony_v80MixB",
|
15 |
+
"Niggendar/realisticPonyPhoto_v10",
|
16 |
+
"Niggendar/realmix_v10",
|
17 |
+
"Niggendar/realmixpony_v01",
|
18 |
+
"Niggendar/realmixpony_v02",
|
19 |
+
"Niggendar/recondiff_v10",
|
20 |
+
"Niggendar/Regro",
|
21 |
+
"Niggendar/relhCheckpoint_v20",
|
22 |
+
]
|
23 |
+
loras = ["openskyml/lcm-lora-sdxl-turbo"]
|
24 |
+
pipe = None
|
25 |
+
cached = {}
|
26 |
+
cached_loras = {}
|
27 |
+
def get_lora(lora_id):
|
28 |
+
if lora_id in cached_loras:
|
29 |
+
return cached_loras[lora_id]
|
30 |
+
lora_dir = snapshot_download(repo_id=lora_id, allow_patterns=["*.safetensors", "*.bin"])
|
31 |
+
lora_files = [f for f in os.listdir(lora_dir) if f.endswith((".safetensors", ".bin"))]
|
32 |
+
lora_path = os.path.join(lora_dir, lora_files[0])
|
33 |
+
cached_loras[lora_id] = lora_path
|
34 |
+
return lora_path
|
35 |
+
def load_pipe(model_id, lora_id):
|
36 |
+
global pipe
|
37 |
+
if (model_id, lora_id) in cached:
|
38 |
+
pipe = cached[(model_id, lora_id)]
|
39 |
+
return
|
40 |
+
if pipe is not None:
|
41 |
+
pipe.to("meta")
|
42 |
+
pipe.unet = None
|
43 |
+
pipe.vae = None
|
44 |
+
pipe.text_encoder = None
|
45 |
+
del pipe
|
46 |
+
gc.collect()
|
47 |
+
cached.clear()
|
48 |
+
pipe = StableDiffusionXLPipeline.from_pretrained(model_id,torch_dtype=torch.float32,low_cpu_mem_usage=True )
|
49 |
+
pipe.scheduler = LCMScheduler.from_config(pipe.scheduler.config)
|
50 |
+
pipe.load_lora_weights(get_lora(lora_id))
|
51 |
+
pipe.to("cpu", dtype=torch.float32)
|
52 |
+
pipe.enable_attention_slicing()
|
53 |
+
cached[(model_id, lora_id)] = pipe
|
54 |
+
return gr.update(value='-')
|
55 |
+
def infer(model_id, lora_id, prompt, seed=None, steps=4, guid=0.1):
|
56 |
+
if seed is None or seed == "":
|
57 |
+
seed = random.randint(0, 2**32 - 1)
|
58 |
+
yield Image.new("RGB", (512, 512), color="gray"), gr.update(value='-')
|
59 |
+
image = pipe( prompt, generator=torch.manual_seed(int(seed)), num_inference_steps=steps,
|
60 |
+
guidance_scale=guid,width=128+256, height=128+256, added_cond_kwargs={} ).images[0]
|
61 |
+
yield image, gr.update(value='-')
|
62 |
+
|
63 |
+
|
64 |
+
with gr.Blocks() as demo:
|
65 |
+
with gr.Row():
|
66 |
+
with gr.Column(scale=2):
|
67 |
+
text2=gr.Textbox(label="Time",placeholder="timer",container=False,value='-')
|
68 |
+
mbtn=gr.Button(value="Load Pair")
|
69 |
+
modeldrop=gr.Dropdown(models, label="Model")
|
70 |
+
loradrop=gr.Dropdown(loras, label="LCM LoRA")
|
71 |
+
with gr.Accordion(label="Settings", open=False):
|
72 |
+
seed=gr.Textbox(label="Seed",visible=False)
|
73 |
+
steps=gr.Slider(1, 15, value=4, step=1, label="Steps")
|
74 |
+
guidance=gr.Slider(0.0, 2.0, value=0.1, step=0.1, label="Guidance Scale")
|
75 |
+
with gr.Column(scale=3):
|
76 |
+
text= gr.Textbox(label="Prompt",container=False,placeholder="Prompt",value='')
|
77 |
+
gbtn=gr.Button(value="Generate")
|
78 |
+
imageout=gr.Image()
|
79 |
+
mbtn.click(fn=load_pipe, inputs=[ modeldrop, loradrop ], outputs=[text2])
|
80 |
+
gbtn.click(fn=infer, inputs=[ modeldrop, loradrop, text, seed,steps, guidance ], outputs=[imageout,text2])
|
81 |
+
|
82 |
+
demo.queue()
|
83 |
+
demo.launch(server_name="0.0.0.0", server_port=7860)
|
84 |
+
|