intro text

app.py

@@ -9,7 +9,29 @@ import operator
 pd.options.plotting.backend = "plotly"
 
 
-TITLE = "Diffusion Professions Cluster Explorer"
+TITLE = "Identity Biases in Diffusion Models: Professions"
+
+_INTRO = """
+# Identity Biases in Diffusion Models: Professions
+        
+Explore profession-level social biases in the data from [DiffusionBiasExplorer](https://hf.co/spaces/society-ethics/DiffusionBiasExplorer)!
+This demo leverages the gender and ethnicity representation clusters described in the [companion app](https://hf.co/spaces/society-ethics/DiffusionFaceClustering)
+to analyze social trends in machine-generated visual representations of professions.
+The **Professions Overview** tab lets you compare the distribution over
+[identity clusters](https://hf.co/spaces/society-ethics/DiffusionFaceClustering "Identity clusters identify visual features in the systems' output space correlated with variation of gender and ethnicity in input prompts.")
+across professions for Stable Diffusion and Dalle-2 systems (or aggregated for `All Models`).
+The **Professions Focus** tab provides more details for each of the individual professions, including direct system comparisons and examples of profession images for each cluster.
+This work was done in the scope of the [Stable Bias Project](https://hf.co/spaces/society-ethics/StableBias).
+As you use this demo, please share findings and comments [in the discussions tab](https://hf.co/spaces/society-ethics/DiffusionClustering/discussions)!
+"""
+
+_ = """
+ For example, you can use this tool to investigate:
+ - How do each model's representation of professions correlate with the gender ratios reported by the [U.S. Bureau of Labor
+Statistics](https://www.bls.gov/cps/cpsaat11.htm "The reported percentage of women in each profession in the US is indicated in the `Labor Women` column in the Professions Overview tab.")?
+Are social trends reflected, are they exaggerated?
+- Which professions have the starkest differences in how different models represent them?
+"""
 
 professions_dset = load_from_disk("professions")
 professions_df = professions_dset.to_pandas()
@@ -92,7 +114,7 @@ def make_profession_plot(num_clusters, prof_name):
     )
     df = pd.DataFrame.from_dict(pre_pandas)
     prof_plot = df.plot(kind="bar", barmode="group")
-    cl_summary_text = f"Profession ``{prof_name}'':\n"
+    cl_summary_text = f"Profession '{prof_name}':\n"
     for cl_id, _ in sorted_cl_scores:
         cl_summary_text += f"- {cluster_summaries_by_size[str(num_clusters)][int(cl_id)].replace(' gender terms', '').replace('; ethnicity terms:', ',')} \n"
     return (
@@ -206,20 +228,16 @@ def show_examplars(num_clusters, prof_name, cl_id, confidence_threshold=0.6):
 
 
 with gr.Blocks(title=TITLE) as demo:
-    gr.Markdown(
-        """
-        # Identity Biases in Diffusion Models: Professions
-        
-        This tool helps you explore the different clusters that we discovered in the images generated by 3 text-to-image models: Dall-E 2, Stable Diffusion v.1.4 and v.2.
-        This work was done in the scope of the [Stable Bias Project](https://huggingface.co/spaces/society-ethics/StableBias).
-        """
-    )
+    gr.Markdown(_INTRO)
     gr.HTML(
         """<span style="color:red" font-size:smaller>⚠️ DISCLAIMER: the images displayed by this tool were generated by text-to-image systems and may depict offensive stereotypes or contain explicit content.</span>"""
     )
     with gr.Tab("Professions Overview"):
         gr.Markdown(
-            "Select one or more professions and models from the dropdowns on the left to see which clusters are most representative for this combination. Try choosing different numbers of clusters to see if the results change, and then go to the 'Profession Focus' tab to go more in-depth into these results."
+            """
+            Select one or more professions and models from the dropdowns on the left to see which clusters are most representative for this combination.
+            Try choosing different numbers of clusters to see if the results change, and then go to the 'Profession Focus' tab to go more in-depth into these results.
+            """
         )
         with gr.Row():
             with gr.Column(scale=1):
@@ -277,7 +295,7 @@ with gr.Blocks(title=TITLE) as demo:
         with gr.Row():
             with gr.Column():
                 gr.Markdown(
-                    "Select profession to visualize and see which clusters and identity groups are most represented in the profession, as well as some examples of generated images below."
+                    "Select a profession to visualize and see which clusters and identity groups are most represented in the profession, as well as some examples of generated images below."
                 )
                 profession_choice_focus = gr.Dropdown(
                     choices=professions,