Spaces:
Running
on
Zero
Running
on
Zero
Update app.py
Browse files
app.py
CHANGED
@@ -9,44 +9,15 @@ from safetensors.torch import load_file
|
|
9 |
from omegaconf import OmegaConf
|
10 |
|
11 |
from image_datasets.dataset import image_resize
|
12 |
-
|
13 |
-
tensor = in_image.squeeze(0)
|
14 |
-
tensor = (tensor + 1) / 2
|
15 |
-
tensor = tensor * 255
|
16 |
-
numpy_array = tensor.permute(1, 2, 0).byte().numpy()
|
17 |
-
pil_image = Image.fromarray(numpy_array)
|
18 |
-
return pil_image
|
19 |
-
# from src.flux.xflux_pipeline import XFluxSampler
|
20 |
args = OmegaConf.load("inference_configs/inference.yaml")
|
21 |
-
# is_schnell = args.model_name == "flux-schnell"
|
22 |
-
# sampler = None
|
23 |
device = torch.device("cuda")
|
24 |
dtype = torch.bfloat16
|
25 |
-
# dit = load_flow_model2(args.model_name, device="cpu").to(device, dtype=dtype)
|
26 |
-
# vae = load_ae(args.model_name, device="cpu").to(device, dtype=dtype)
|
27 |
-
# t5 = load_t5(device="cpu", max_length=256 if is_schnell else 512).to(device, dtype=dtype)
|
28 |
-
# clip = load_clip("cpu").to(device, dtype=dtype)
|
29 |
-
#test push
|
30 |
@spaces.GPU
|
31 |
def generate(image: Image.Image, edit_prompt: str):
|
32 |
from src.flux.xflux_pipeline import XFluxSampler
|
33 |
|
34 |
|
35 |
-
|
36 |
-
|
37 |
-
# vae.requires_grad_(False)
|
38 |
-
# t5.requires_grad_(False)
|
39 |
-
# clip.requires_grad_(False)
|
40 |
-
|
41 |
-
# model_path = hf_hub_download(
|
42 |
-
# repo_id="Boese0601/ByteMorpher",
|
43 |
-
# filename="dit.safetensors",
|
44 |
-
# use_auth_token=os.getenv("HF_TOKEN")
|
45 |
-
# )
|
46 |
-
# state_dict = load_file(model_path)
|
47 |
-
# dit.load_state_dict(state_dict)
|
48 |
-
# dit.eval()
|
49 |
-
# dit.to(device, dtype=dtype)
|
50 |
|
51 |
sampler = XFluxSampler(
|
52 |
device = device,
|
@@ -56,42 +27,7 @@ def generate(image: Image.Image, edit_prompt: str):
|
|
56 |
image_encoder=None,
|
57 |
improj=None
|
58 |
)
|
59 |
-
|
60 |
-
# device = torch.device("cuda")
|
61 |
-
# dtype = torch.bfloat16
|
62 |
-
|
63 |
-
# if sampler is None:
|
64 |
-
# dit = load_flow_model2(args.model_name, device="cpu").to(device, dtype=dtype)
|
65 |
-
# vae = load_ae(args.model_name, device="cpu").to(device, dtype=dtype)
|
66 |
-
# t5 = load_t5(device="cpu", max_length=256 if is_schnell else 512).to(device, dtype=dtype)
|
67 |
-
# clip = load_clip("cpu").to(device, dtype=dtype)
|
68 |
-
|
69 |
-
# vae.requires_grad_(False)
|
70 |
-
# t5.requires_grad_(False)
|
71 |
-
# clip.requires_grad_(False)
|
72 |
-
|
73 |
-
# model_path = hf_hub_download(
|
74 |
-
# repo_id="Boese0601/ByteMorpher",
|
75 |
-
# filename="dit.safetensors",
|
76 |
-
# use_auth_token=os.getenv("HF_TOKEN")
|
77 |
-
# )
|
78 |
-
# state_dict = load_file(model_path)
|
79 |
-
# dit.load_state_dict(state_dict)
|
80 |
-
# dit.eval()
|
81 |
-
|
82 |
-
# sampler = XFluxSampler(
|
83 |
-
# clip=clip,
|
84 |
-
# t5=t5,
|
85 |
-
# ae=vae,
|
86 |
-
# model=dit,
|
87 |
-
# device=device,
|
88 |
-
# ip_loaded=False,
|
89 |
-
# spatial_condition=False,
|
90 |
-
# clip_image_processor=None,
|
91 |
-
# image_encoder=None,
|
92 |
-
# improj=None
|
93 |
-
# )
|
94 |
-
|
95 |
img = image_resize(image, 512)
|
96 |
w, h = img.size
|
97 |
img = img.resize(((w // 32) * 32, (h // 32) * 32))
|
|
|
9 |
from omegaconf import OmegaConf
|
10 |
|
11 |
from image_datasets.dataset import image_resize
|
12 |
+
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
13 |
args = OmegaConf.load("inference_configs/inference.yaml")
|
|
|
|
|
14 |
device = torch.device("cuda")
|
15 |
dtype = torch.bfloat16
|
|
|
|
|
|
|
|
|
|
|
16 |
@spaces.GPU
|
17 |
def generate(image: Image.Image, edit_prompt: str):
|
18 |
from src.flux.xflux_pipeline import XFluxSampler
|
19 |
|
20 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
21 |
|
22 |
sampler = XFluxSampler(
|
23 |
device = device,
|
|
|
27 |
image_encoder=None,
|
28 |
improj=None
|
29 |
)
|
30 |
+
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
31 |
img = image_resize(image, 512)
|
32 |
w, h = img.size
|
33 |
img = img.resize(((w // 32) * 32, (h // 32) * 32))
|