amos1088 commited on
Commit
9f78050
·
1 Parent(s): a0c491d
Files changed (1) hide show
  1. app.py +12 -7
app.py CHANGED
@@ -12,6 +12,7 @@ import torch
12
  from diffusers import StableDiffusion3ControlNetPipeline, SD3ControlNetModel
13
  from diffusers.utils import load_image
14
  from image_gen_aux import DepthPreprocessor
 
15
 
16
  # ----------------------------
17
  # Step 1: Download IP Adapter if not exists
@@ -51,18 +52,21 @@ image_encoder_path = "google/siglip-so400m-patch14-384"
51
 
52
 
53
  controlnet = SD3ControlNetModel.from_pretrained("stabilityai/stable-diffusion-3.5-large-controlnet-depth", torch_dtype=torch.float16)
54
- pipe = StableDiffusion3ControlNetPipeline.from_pretrained(
55
- "stabilityai/stable-diffusion-3.5-large",
56
- controlnet=controlnet,
57
- torch_dtype=torch.float16,
58
- ).to("cuda")
59
 
60
- pipe.load_ip_adapter(
61
  ip_adapter_path=ip_adapter_path,
62
  image_encoder_path=image_encoder_path,
63
  nb_token=64,
 
64
  )
65
 
 
 
 
 
 
 
 
66
 
67
 
68
  # ----------------------------
@@ -77,6 +81,8 @@ def gui_generation(prompt,negative_prompt, ref_img, guidance_scale, ipadapter_sc
77
  control_image = depth_preprocessor(image, invert=True)[0].convert("RGB")
78
 
79
  generator = torch.Generator(device="cpu").manual_seed(0)
 
 
80
  image = pipe(
81
  width=1024,
82
  height=1024,
@@ -88,7 +94,6 @@ def gui_generation(prompt,negative_prompt, ref_img, guidance_scale, ipadapter_sc
88
  num_inference_steps=40,
89
  generator=generator,
90
  max_sequence_length=77,
91
- ipadapter_scale=ipadapter_scale,
92
  ).images[0]
93
 
94
  return image
 
12
  from diffusers import StableDiffusion3ControlNetPipeline, SD3ControlNetModel
13
  from diffusers.utils import load_image
14
  from image_gen_aux import DepthPreprocessor
15
+ from diffusers.models import SD3ControlNetModel, T2IAdapter
16
 
17
  # ----------------------------
18
  # Step 1: Download IP Adapter if not exists
 
52
 
53
 
54
  controlnet = SD3ControlNetModel.from_pretrained("stabilityai/stable-diffusion-3.5-large-controlnet-depth", torch_dtype=torch.float16)
 
 
 
 
 
55
 
56
+ adapter = T2IAdapter.from_pretrained(
57
  ip_adapter_path=ip_adapter_path,
58
  image_encoder_path=image_encoder_path,
59
  nb_token=64,
60
+ torch_dtype=torch.float16
61
  )
62
 
63
+ pipe = StableDiffusion3ControlNetPipeline.from_pretrained(
64
+ "stabilityai/stable-diffusion-3.5-large",
65
+ controlnet=controlnet,adapter=adapter,
66
+ torch_dtype=torch.float16,
67
+ ).to("cuda")
68
+
69
+
70
 
71
 
72
  # ----------------------------
 
81
  control_image = depth_preprocessor(image, invert=True)[0].convert("RGB")
82
 
83
  generator = torch.Generator(device="cpu").manual_seed(0)
84
+ pipe.set_ip_adapter_scale(ipadapter_scale) # Adjust the scale as needed
85
+
86
  image = pipe(
87
  width=1024,
88
  height=1024,
 
94
  num_inference_steps=40,
95
  generator=generator,
96
  max_sequence_length=77,
 
97
  ).images[0]
98
 
99
  return image