fffiloni commited on
Commit
396214f
1 Parent(s): 6a3a19b

Update app.py

Browse files
Files changed (1) hide show
  1. app.py +28 -9
app.py CHANGED
@@ -65,7 +65,28 @@ def get_caption_from_MD(image_in):
65
  print(result)
66
  return result
67
 
68
-
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
69
  import re
70
  import torch
71
  from transformers import pipeline
@@ -102,8 +123,10 @@ def infer(image_in):
102
  cleaned_text = re.sub(pattern, '', outputs[0]["generated_text"], flags=re.DOTALL)
103
 
104
  print(f"SUGGESTED Musical prompt: {cleaned_text}")
 
 
105
 
106
- return user_prompt, cleaned_text.lstrip("\n")
107
 
108
  title = "Image to Music V2",
109
  description = "Get music from a picture"
@@ -133,14 +156,10 @@ with gr.Blocks(css=css) as demo:
133
  submit_btn = gr.Button("Make LLM system from my pic !")
134
  with gr.Column():
135
  caption = gr.Textbox(
136
- label = "Image caption (Kosmos2)",
137
- elem_id = "image-caption"
138
  )
139
- result = gr.Textbox(
140
- label = "Suggested System",
141
- lines = 6,
142
- max_lines = 30,
143
- elem_id = "suggested-system-prompt"
144
  )
145
  with gr.Row():
146
  gr.Examples(
 
65
  print(result)
66
  return result
67
 
68
+ def get_magnet(prompt):
69
+ amended_prompt = f"{prompt}"
70
+ print(amended_prompt)
71
+ client = Client("https://fffiloni-magnet.hf.space/")
72
+ result = client.predict(
73
+ "facebook/magnet-small-10secs", # Literal['facebook/magnet-small-10secs', 'facebook/magnet-medium-10secs', 'facebook/magnet-small-30secs', 'facebook/magnet-medium-30secs', 'facebook/audio-magnet-small', 'facebook/audio-magnet-medium'] in 'Model' Radio component
74
+ "", # str in 'Model Path (custom models)' Textbox component
75
+ amended_prompt, # str in 'Input Text' Textbox component
76
+ 3, # float in 'Temperature' Number component
77
+ 0.9, # float in 'Top-p' Number component
78
+ 10, # float in 'Max CFG coefficient' Number component
79
+ 1, # float in 'Min CFG coefficient' Number component
80
+ 20, # float in 'Decoding Steps (stage 1)' Number component
81
+ 10, # float in 'Decoding Steps (stage 2)' Number component
82
+ 10, # float in 'Decoding Steps (stage 3)' Number component
83
+ 10, # float in 'Decoding Steps (stage 4)' Number component
84
+ "prod-stride1 (new!)", # Literal['max-nonoverlap', 'prod-stride1 (new!)'] in 'Span Scoring' Radio component
85
+ api_name="/predict_full"
86
+ )
87
+ print(result)
88
+ return result[1]
89
+
90
  import re
91
  import torch
92
  from transformers import pipeline
 
123
  cleaned_text = re.sub(pattern, '', outputs[0]["generated_text"], flags=re.DOTALL)
124
 
125
  print(f"SUGGESTED Musical prompt: {cleaned_text}")
126
+
127
+ music_o = get_magnet(cleaned_text)
128
 
129
+ return cleaned_text, music_o
130
 
131
  title = "Image to Music V2",
132
  description = "Get music from a picture"
 
156
  submit_btn = gr.Button("Make LLM system from my pic !")
157
  with gr.Column():
158
  caption = gr.Textbox(
159
+ label = "Musical prompt"
 
160
  )
161
+ result = gr.Audio(
162
+ label = "Music"
 
 
 
163
  )
164
  with gr.Row():
165
  gr.Examples(