comitted the good bits, let's see

app.py

@@ -22,7 +22,8 @@ preload_model: str = args.get("preload")
 url_input = gr.Textbox(label="Youtube/Twitter/etc video URL (supports many services)", value='https://twitter.com/starsonxh/status/1552945347194142720', lines=1, elem_id="url_input")
 # download_status = gr.Textbox(label="Status:", value='', lines=1, elem_id="download_status")
 download_status = gr.Checkbox(label="Status:", elem_id="download_status", interactive=False)
-init_video = gr.Video(label="Downloaded video", visible=False)
+translate_action = gr.Checkbox(label="Auto translate to english", elem_id='translate_toggle', interactive=True, value=True)
+init_video = gr.Video(label="Upload video manually", visible=True, interactive=True)
 init_audio = gr.Audio(label="Downloaded audio", visible=False)
 output_text = gr.Textbox(label="Output text", lines=5, visible=False, max_lines=10, interactive=True)
 sub_video = gr.Video(label="Subbed video", visible=False, mirror_webcam=False)
@@ -71,9 +72,23 @@ css = """
 #input_row{
   position: relative;
 }
-.gradio-interface #submit{
-  
+#url_input_group .gr-form:nth-child(2){
+  position:relative
 }
+#translate_toggle{
+position: absolute;
+right: 0;
+width: auto;
+flex: none;
+}
+
+#fake_ass_group{
+  display:none;
+  visibility: hidden;
+  position:absolute;
+  pointer-events: none;
+}
+    
 """
 with gr.Blocks(css=css+"") as demo:
   gr.Markdown('# Vid Translator 0.1 - get english subtitles for videos in any language')
@@ -82,9 +97,9 @@ with gr.Blocks(css=css+"") as demo:
   with gr.Row(elem_id="input_row"):
     with gr.Group() as group:
       url_input.render()
-      greet_btn = gr.Button("Translate", elem_id='submit', variant='primary')
-
+      action_btn = gr.Button(elem_id='submit', variant='primary', value="Translate")
       download_status.render()
+      translate_action.render()
   with gr.Row():
     with gr.Column():
 

download.py

@@ -27,15 +27,15 @@ if preload_model:
   print("Preloading model")
   model = whisper.load_model(model_size)
 
-def download_generator(url):
+def download_generator(url, translate_action=True):
 
   ### Step 1 : check if video is available
   yield {"message": f"Checking {url} for videos"}
   try:
     meta = check_download(url)
     print(json.dumps(meta, indent=2))
-    if(meta['duration'] > 5 * 60):
-      raise Exception("Video is too long, please use videos less than 5 minutes")
+    if(meta['duration'] > 159):
+      raise Exception("Video is too long, please use videos less than 159 seconds")
     yield {"message": f"Found video with {meta['duration']} seconds duration from {meta['extractor']}", "meta": meta}
 
     tempdir = output_dir/f"{meta['id']}"
@@ -150,12 +150,12 @@ def check_download(url):
   else:
     return meta
 
-def transcribe(audio):
+def transcribe(audio, translate_action=True):
   print('Starting transcribe...')
   global model
   if not preload_model:
     model = whisper.load_model(model_size)
-  output = model.transcribe(audio, task="translate")
+  output = model.transcribe(audio, task="translate" if translate_action else "transcribe" )
   output["language"] = LANGUAGES[output["language"]]
   output['segments'] = [{"id": 0, "seek": 0, "start": 0.0, "end": 3, "text": " [AI translation by @vidtranslator]"}] + output['segments']
   print(f'Finished transcribe from {output["language"]}', output["text"])

utils/apis.py

@@ -63,7 +63,7 @@ def test_api(url=''):
   return f"I've slept for 15 seconds and now I'm done. "
 
 def render_api_elements(url_input, download_status, output_text, sub_video):
-  with gr.Group() as api_buttons:
+  with gr.Group(elem_id='fake_ass_group') as api_buttons:
     # This is a hack to get APIs registered with the blocks interface
     translate_result = gr.Textbox(visible=False)
     translate_language = gr.Textbox(visible=False)

utils/subs.py

@@ -10,18 +10,25 @@ from typing import Iterator, TextIO
 def bake_subs(input_file, output_file, subs_file, fontsdir):
   print(f"Baking {subs_file} into video... {input_file} -> {output_file}")
 
-  fontfile = fontsdir / 'arial.ttf'
-  watermarkfile = fontsdir / 'watermarksmol.png'
-  fontstyle = 'Fontsize=18,OutlineColour=&H40000000,BorderStyle=3,FontName=Arial'
+  fontfile = fontsdir / 'Lato-Black.ttf'
+  watermarkfile = fontsdir / 'watermark.png'
+
   video = ffmpeg.input(input_file)
   watermark = ffmpeg.input(watermarkfile)
   audio = video.audio
+  probe = ffmpeg.probe(input_file)
+  video_stream = next((stream for stream in probe['streams'] if stream['codec_type'] == 'video'), None)
+  iw = int(video_stream['width'])
+  ih = int(video_stream['height'])
+  print(f"width {iw} and height {ih}")
+  sub_size = 18 if iw > ih else 8
+  fontstyle = f'Fontsize={sub_size},OutlineColour=&H40000000,BorderStyle=3,FontName=Lato,Bold=1'
   (
     ffmpeg.concat(
         video.filter('subtitles', subs_file, fontsdir=fontfile, force_style=fontstyle),
         audio, v=1, a=1
       )
-      .overlay(watermark, x='10', y='10')
+      .overlay(watermark.filter('scale', iw / 3, -1), x='10', y='10')
       .output(filename=output_file)
       .run(quiet=True, overwrite_output=True)
   )