Help text for app

Topic: app_help_text

riffusion/streamlit/pages/audio_to_audio.py

@@ -20,6 +20,21 @@ def render_audio_to_audio() -> None:
     """
     )
 
+    with st.expander("Help", False):
+        st.write(
+            """
+            This tool allows you to upload an audio file of arbitrary length and modify it with
+            a text prompt. It does this by sweeping over the audio in overlapping clips, doing
+            img2img style transfer with riffusion, then stitching the clips back together with
+            cross fading to eliminate seams.
+
+            Try a denoising strength of 0.4 for light modification and 0.55 for more heavy
+            modification. The best specific denoising depends on how different the prompt is
+            from the source audio. You can play with the seed to get infinite variations.
+            Currently the same seed is used for all clips along the track.
+            """
+        )
+
     device = streamlit_util.select_device(st.sidebar)
 
     audio_file = st.file_uploader(

riffusion/streamlit/pages/image_to_audio.py

@@ -18,6 +18,16 @@ def render_image_to_audio() -> None:
     """
     )
 
+    with st.expander("Help", False):
+        st.write(
+            """
+            This tool takes an existing spectrogram image and reconstructs it into an audio
+            waveform. It also displays the EXIF metadata stored inside the image, which can
+            contain the parameters used to create the spectrogram image. If no EXIF is contained,
+            assumes default parameters.
+            """
+        )
+
     device = streamlit_util.select_device(st.sidebar)
 
     image_file = st.file_uploader(

riffusion/streamlit/pages/sample_clips.py

@@ -17,6 +17,15 @@ def render_sample_clips() -> None:
     """
     )
 
+    with st.expander("Help", False):
+        st.write(
+            """
+            This tool simply allows uploading an audio file and randomly sampling short clips
+            from it. It's useful for generating a large number of short clips from a single
+            audio file. Outputs can be saved to a given directory with a given audio extension.
+            """
+        )
+
     audio_file = st.file_uploader(
         "Upload a file",
         type=["wav", "mp3", "ogg"],

riffusion/streamlit/pages/split_audio.py

@@ -15,6 +15,19 @@ def render_split_audio() -> None:
     """
     )
 
+    with st.expander("Help", False):
+        st.write(
+            """
+            This tool allows uploading an audio file of arbitrary length and splits it into
+            stems of vocals, drums, bass, and other. It does this using a deep network that
+            sweeps over the audio in clips, extracts the stems, and then cross fades the clips
+            back together to construct the full length stems. It's particularly useful in
+            combination with audio_to_audio, for example to split and preserve vocals while
+            modifying the rest of the track with a prompt. Or, to pull out drums to add later
+            in a DAW.
+            """
+        )
+
     device = streamlit_util.select_device(st.sidebar)
 
     audio_file = st.file_uploader(

riffusion/streamlit/pages/text_to_audio.py

@@ -12,11 +12,20 @@ def render_text_to_audio() -> None:
     st.subheader(":pencil2: Text to Audio")
     st.write(
         """
-    Generate audio from text prompts.  \nRuns the model directly without a seed image or
-    interpolation.
+    Generate audio from text prompts.
     """
     )
 
+    with st.expander("Help", False):
+        st.write(
+            """
+            This tool runs riffusion in the simplest text to image form to generate an audio
+            clip from a text prompt. There is no seed image or interpolation here. This mode
+            allows more diversity and creativity than when using a seed image, but it also
+            leads to having less control. Play with the seed to get infinite variations.
+            """
+        )
+
     device = streamlit_util.select_device(st.sidebar)
 
     prompt = st.text_input("Prompt")

riffusion/streamlit/pages/text_to_audio_batch.py

@@ -38,11 +38,20 @@ def render_text_to_audio_batch() -> None:
     st.subheader(":scroll: Text to Audio Batch")
     st.write(
         """
-    Generate audio in batch from a JSON file of text prompts.  \nThe input
-    file contains a global params block and a list of entries with positive and negative
-    prompts.
+    Generate audio in batch from a JSON file of text prompts.
     """
     )
+
+    with st.expander("Help", False):
+        st.write(
+            """
+            This tool is a batch form of text_to_audio, where the inputs are read in from a JSON
+            file. The input file contains a global params block and a list of entries with positive
+            and negative prompts. It's useful for automating a larger set of generations. See the
+            example inputs below for the format of the file.
+            """
+        )
+
     device = streamlit_util.select_device(st.sidebar)
 
     # Upload a JSON file