multimodalart HF staff commited on
Commit
3a1e48f
β€’
1 Parent(s): 758de8d

Create app.py

Browse files
Files changed (1) hide show
  1. app.py +52 -0
app.py ADDED
@@ -0,0 +1,52 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import cv2
2
+ from insightface.app import FaceAnalysis
3
+ import torch
4
+ from diffusers import StableDiffusionPipeline, DDIMScheduler, AutoencoderKL
5
+ from PIL import Image
6
+ from ip_adapter.ip_adapter_faceid import IPAdapterFaceID
7
+
8
+ app = FaceAnalysis(name="buffalo_l", providers=['CUDAExecutionProvider', 'CPUExecutionProvider'])
9
+ app.prepare(ctx_id=0, det_size=(640, 640))
10
+
11
+ base_model_path = "SG161222/Realistic_Vision_V4.0_noVAE"
12
+ vae_model_path = "stabilityai/sd-vae-ft-mse"
13
+ ip_ckpt = hf_hub_download(repo_id='h94/IP-Adapter-FaceID', filename="ip-adapter-faceid_sd15.bin", repo_type="model")
14
+
15
+ device = "cuda"
16
+
17
+ noise_scheduler = DDIMScheduler(
18
+ num_train_timesteps=1000,
19
+ beta_start=0.00085,
20
+ beta_end=0.012,
21
+ beta_schedule="scaled_linear",
22
+ clip_sample=False,
23
+ set_alpha_to_one=False,
24
+ steps_offset=1,
25
+ )
26
+ vae = AutoencoderKL.from_pretrained(vae_model_path).to(dtype=torch.float16)
27
+ pipe = StableDiffusionPipeline.from_pretrained(
28
+ base_model_path,
29
+ torch_dtype=torch.float16,
30
+ scheduler=noise_scheduler,
31
+ vae=vae,
32
+ #feature_extractor=None,
33
+ #safety_checker=None
34
+ )
35
+
36
+ ip_model = IPAdapterFaceID(pipe, ip_ckpt, device)
37
+
38
+ def generate_faceid_embeddings(image):
39
+ #image = cv2.imread("person.jpg")
40
+ faces = app.get(image)
41
+ faceid_embeds = torch.from_numpy(faces[0].normed_embedding).unsqueeze(0)
42
+ return faceid_embeds
43
+
44
+ def generate_image(image, prompt, negative_prompt):
45
+ faceid_embeds = generate_faceid_embeddings(image)
46
+ images = ip_model.generate(
47
+ prompt=prompt, negative_prompt=negative_prompt, faceid_embeds=faceid_embeds, width=512, height=512, num_inference_steps=30
48
+ )
49
+ return images.image[0]
50
+
51
+ demo = gr.Interface(fn=generate_image, inputs=[gr.Image(label="Your face"), gr.Textbox(label="Prompt"), gr.Textbox(label="Negative Prompt")], outputs=[gr.Image(label="Generated Image")])
52
+ demo.launch()