Nick088 commited on
Commit
dae60c0
·
verified ·
1 Parent(s): 54f830c

Update app.py

Browse files
Files changed (1) hide show
  1. app.py +34 -3
app.py CHANGED
@@ -2,8 +2,9 @@ import torch
2
  from diffusers import StableDiffusion3Pipeline
3
  import gradio as gr
4
  import os
5
- import spaces
6
  from huggingface_hub import snapshot_download
 
7
 
8
  HF_TOKEN = os.getenv("HF_TOKEN")
9
 
@@ -27,9 +28,34 @@ else:
27
  pipe = StableDiffusion3Pipeline.from_pretrained(model_path, torch_dtype=torch.float16)
28
  pipe.to(device)
29
 
 
 
 
 
30
  # Define the image generation function
31
  @spaces.GPU(duration=60)
32
- def generate_image(prompt, negative_prompt, num_inference_steps, height, width, guidance_scale, num_images_per_prompt):
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
33
  output = pipe(
34
  prompt=prompt,
35
  negative_prompt=negative_prompt,
@@ -37,6 +63,7 @@ def generate_image(prompt, negative_prompt, num_inference_steps, height, width,
37
  height=height,
38
  width=width,
39
  guidance_scale=guidance_scale,
 
40
  num_images_per_prompt=num_images_per_prompt
41
  ).images
42
  return output
@@ -45,6 +72,8 @@ def generate_image(prompt, negative_prompt, num_inference_steps, height, width,
45
 
46
  prompt = gr.Textbox(label="Prompt", info="Describe the image you want", placeholder="A cat...")
47
 
 
 
48
  negative_prompt = gr.Textbox(label="Negative Prompt", info="Describe what you don't want in the image", placeholder="Ugly, bad anatomy...")
49
 
50
  num_inference_steps = gr.Number(label="Number of Inference Steps", precision=0, value=25)
@@ -55,11 +84,13 @@ width = gr.Slider(label="Width", info="Width of the Image", minimum=256, maximum
55
 
56
  guidance_scale = gr.Number(minimum=0.1, value=7.5, label="Guidance Scale", info="The number of denoising steps of the image. More denoising steps usually lead to a higher quality image at the cost of slower inference")
57
 
 
 
58
  num_images_per_prompt = gr.Slider(label="Number of Images to generate with the settings",minimum=1, maximum=4, step=1, value=1)
59
 
60
  interface = gr.Interface(
61
  fn=generate_image,
62
- inputs=[prompt, negative_prompt, num_inference_steps, height, width, guidance_scale, num_images_per_prompt],
63
  outputs=gr.Gallery(label="Generated AI Images", elem_id="gallery", show_label=False),
64
  title="Stable Diffusion 3 Medium",
65
  description="Made by <a href='https://linktr.ee/Nick088' target='_blank'>Nick088</a> \n Join https://discord.gg/osai to talk about Open Source AI"
 
2
  from diffusers import StableDiffusion3Pipeline
3
  import gradio as gr
4
  import os
5
+ from transformers import T5Tokenizer, T5ForConditionalGeneration
6
  from huggingface_hub import snapshot_download
7
+ import spaces
8
 
9
  HF_TOKEN = os.getenv("HF_TOKEN")
10
 
 
28
  pipe = StableDiffusion3Pipeline.from_pretrained(model_path, torch_dtype=torch.float16)
29
  pipe.to(device)
30
 
31
+ tokenizer = T5Tokenizer.from_pretrained("roborovski/superprompt-v1")
32
+ model = T5ForConditionalGeneration.from_pretrained("roborovski/superprompt-v1", device_map="auto", torch_dtype="auto")
33
+ model.to(device)
34
+
35
  # Define the image generation function
36
  @spaces.GPU(duration=60)
37
+ def generate_image(prompt, negative_prompt, num_inference_steps, height, width, guidance_scale, seed, num_images_per_prompt):
38
+ if seed == 0:
39
+ seed = random.randint(1, 2**32-1)
40
+
41
+ if enhance_prompt:
42
+ transformers.set_seed(seed)
43
+
44
+ input_text = f"Expand the following prompt to add more detail: {prompt}"
45
+ input_ids = tokenizer(input_text, return_tensors="pt").input_ids.to(device)
46
+
47
+ outputs = model.generate(
48
+ input_ids,
49
+ max_new_tokens=512,
50
+ repetition_penalty=1.2,
51
+ do_sample=True,
52
+ temperature=0.7,
53
+ top_p=1,
54
+ top_k=50,
55
+ )
56
+
57
+ generator = torch.Generator().manual_seed(seed)
58
+
59
  output = pipe(
60
  prompt=prompt,
61
  negative_prompt=negative_prompt,
 
63
  height=height,
64
  width=width,
65
  guidance_scale=guidance_scale,
66
+ generator=generator,
67
  num_images_per_prompt=num_images_per_prompt
68
  ).images
69
  return output
 
72
 
73
  prompt = gr.Textbox(label="Prompt", info="Describe the image you want", placeholder="A cat...")
74
 
75
+ enhance_prompt = gr.Checkbox(label="Prompt Enhancement", info="Enhance your prompt with SuperPrompt-v1", value=True)
76
+
77
  negative_prompt = gr.Textbox(label="Negative Prompt", info="Describe what you don't want in the image", placeholder="Ugly, bad anatomy...")
78
 
79
  num_inference_steps = gr.Number(label="Number of Inference Steps", precision=0, value=25)
 
84
 
85
  guidance_scale = gr.Number(minimum=0.1, value=7.5, label="Guidance Scale", info="The number of denoising steps of the image. More denoising steps usually lead to a higher quality image at the cost of slower inference")
86
 
87
+ seed = gr.Slider(value=42, minimum=0, maximum=2**32-1, step=1, label="Seed", info="A starting point to initiate the generation process, put 0 for a random one")
88
+
89
  num_images_per_prompt = gr.Slider(label="Number of Images to generate with the settings",minimum=1, maximum=4, step=1, value=1)
90
 
91
  interface = gr.Interface(
92
  fn=generate_image,
93
+ inputs=[prompt, enhance_prompt, negative_prompt, num_inference_steps, height, width, guidance_scale, seed, num_images_per_prompt],
94
  outputs=gr.Gallery(label="Generated AI Images", elem_id="gallery", show_label=False),
95
  title="Stable Diffusion 3 Medium",
96
  description="Made by <a href='https://linktr.ee/Nick088' target='_blank'>Nick088</a> \n Join https://discord.gg/osai to talk about Open Source AI"