Update app.py

app.py

@@ -149,7 +149,7 @@ def generate_cmx_edl(edl_title, segments, clip_metadata, fps=25):
             src_out_frames = seconds_to_frames(src_end_val, fps) + start_offset_frames
             duration_frames = src_out_frames - src_in_frames
             
-            current_source = source_clip.replace(" ", "_")[:8] # Short EDL reel ID
+            current_source = source_clip.replace(" ", "_")[:8]
             note_text = seg.get('note', 'Clip')
             source_name = source_clip
             
@@ -473,7 +473,7 @@ def call_gemini_for_edl(transcripts_dict, story_prompt, api_key, previous_edit=N
 st.set_page_config(page_title="Junior Editor Pro", layout="wide")
 st.title("Junior Editor (Multi-Clip Edition)")
 
-with st.expander("Instructions - Please Read First", expanded=False):
+with st.expander("Instructions & Prompt Guide - Please Read First", expanded=False):
     st.markdown("""
     * This program will generate an EDL or XML containing multiple interview cut down and interwoven as directed by you.
     * Set your export timeline FPS and EDL or XML. If you are doing sequences it will have to be XML, but this will only work in Resolve (I haven’t tried FCP). If you are exporting sequences have the TC start at 00:00:00:00 for speed later. 
@@ -483,8 +483,30 @@ with st.expander("Instructions - Please Read First", expanded=False):
     * Junior Editor will transcribe and separate speakers. This takes approx 10 mins per hour of audio. Once completed.
     * You can then instruct it to find engaging bits or construct a narrative. You can ask it to add in suggested VO or Graphics cards if you are planning to use voice over. 
     * It will create an EDL or XML to import back into Resolve. 
-    * If you don’t like the first attempt, give it new instructions and ask it to generate a new sequence. 
     * Resolve Users: Uncheck "Automatically import source clips into media pool" during import.
+    
+    ---
+    ### 💡 Pro-Tips for Prompting the AI
+    **1. Think in Clips, Not Minutes:** The AI cannot accurately calculate running time. If you want a video longer than 3-5 minutes, **do not ask for a time limit**. Instead, ask for a specific number of clips (assume roughly 8-10 clips per minute).
+    
+    **2. Use Bullet Points:** The AI follows technical, structured lists much better than paragraphs.
+    
+    **3. Do the Math for It:** If you want specific ratios, give the AI hard numbers (e.g., "Use 30 clips from Cam A and 10 from Cam B") instead of percentages (e.g., "75% Cam A").
+    
+    **Example of a Great Prompt:**
+    > **TASK:** Build a comprehensive highlights reel from the provided event transcripts. 
+    > 
+    > **LENGTH CONSTRAINT:** You MUST generate a minimum of 75 to 80 individual clip segments. Do not stop early.
+    > 
+    > **SOURCE ALLOCATION:**
+    > * **~32 clips** from `BeutlerClips2` (Use heavily as the inspiring backbone).
+    > * **~8 clips** from `Bruce Interview`.
+    > * **~40 clips** mixed from the other attendee interviews. 
+    > 
+    > **NARRATIVE & PACING RULES:**
+    > * **No VO/Graphics.** > * **The Hook:** Start with an intriguing hook, followed by a clip explaining the event.
+    > * **The Chain:** Each clip must organically lead into the next. Do not repeat the same info. 
+    > * **The Ending:** Finish with a highly uplifting statement.
     """)
 
 st.divider()
@@ -507,6 +529,8 @@ if "transcripts" not in st.session_state:
     st.session_state.transcripts = {}
 if "clip_metadata" not in st.session_state:
     st.session_state.clip_metadata = {}
+if "revision_count" not in st.session_state:
+    st.session_state.revision_count = 1
 
 uploaded_files = st.file_uploader("Upload Clip(s)", type=["mp4", "m4a", "wav", "mp3", "mov"], accept_multiple_files=True)
 
@@ -572,7 +596,7 @@ if uploaded_files and len(st.session_state.transcripts) == 0:
             gc.collect()
             torch.cuda.empty_cache()
             
-            diarize_segments = diarize_model(audio) # Removed min/max speaker restrictions
+            diarize_segments = diarize_model(audio) 
             final_result = whisperx.assign_word_speakers(diarize_segments, result)
             
             processed_segments = []
@@ -608,7 +632,7 @@ if uploaded_files and len(st.session_state.transcripts) == 0:
         st.rerun()
 
 # -- Chat / Generation Area --
-if len(st.session_state.transcripts) > 0:
+if len(st.session_state.transcripts) > 0 and not st.session_state.get("edit_generated"):
     st.divider()
     
     with st.expander("View Source Transcripts", expanded=False):
@@ -620,7 +644,7 @@ if len(st.session_state.transcripts) > 0:
             st.divider()
 
     st.subheader("Your Instruction")
-    brief = st.text_area("What should the Junior Editor do?", placeholder="e.g. Build a 1-minute intro cutting between Cam_A and Cam_B.")
+    brief = st.text_area("What should the Junior Editor do?", placeholder="e.g. Build a sequence using 40 clips, starting with an intriguing hook...")
     
     if st.button("Generate Multi-Clip Edit", type="primary"):
         with st.spinner("Analyzing all transcripts and building sequence..."):
@@ -630,13 +654,13 @@ if len(st.session_state.transcripts) > 0:
                 safe_title = "".join([c for c in sequence_title if c.isalnum() or c in (' ', '_', '-')]).strip().replace(' ', '_')
                 
                 if "XML" in export_format:
-                    final_output = generate_xml(safe_title, edl_segments, st.session_state.clip_metadata, fps)
+                    final_output = generate_xml(f"{safe_title}_V1", edl_segments, st.session_state.clip_metadata, fps)
                     ext = "xml"
                 else:
-                    final_output = generate_cmx_edl(safe_title, edl_segments, st.session_state.clip_metadata, fps)
+                    final_output = generate_cmx_edl(f"{safe_title}_V1", edl_segments, st.session_state.clip_metadata, fps)
                     ext = "edl"
                     
-                transcript_txt = generate_transcript_txt(safe_title, edl_segments, st.session_state.clip_metadata, fps, st.session_state.transcripts)
+                transcript_txt = generate_transcript_txt(f"{safe_title}_V1", edl_segments, st.session_state.clip_metadata, fps, st.session_state.transcripts)
                 
                 st.session_state.edit_generated = True
                 st.session_state.revision_count = 1
@@ -646,7 +670,9 @@ if len(st.session_state.transcripts) > 0:
                 st.session_state.transcript_txt = transcript_txt
                 st.session_state.used_model = used_model
                 st.session_state.edl_segments = edl_segments
+                st.rerun()
 
+# -- Post-Generation / Revision Area --
 if st.session_state.get("edit_generated"):
     st.success(f"✅ Multi-clip sequence generated successfully using **{st.session_state.used_model}**.")
     
@@ -656,7 +682,40 @@ if st.session_state.get("edit_generated"):
     with col2:
         st.download_button(f"Download Transcript (.TXT)", data=st.session_state.transcript_txt, file_name=f"{st.session_state.safe_title}_V{st.session_state.revision_count}_Transcript.txt")
 
-    if st.button("Start Over / Clear Bin"):
-        for key in ["transcripts", "clip_metadata", "edit_generated"]:
-            if key in st.session_state: del st.session_state[key]
-        st.rerun()
+    st.divider()
+    
+    st.subheader("Revise This Edit")
+    revision_brief = st.text_area("Want changes? Give Junior Editor new instructions based on the current cut:", placeholder="e.g. Make it longer (add 20 more clips), swap the opening clip, or focus more on AstraZeneca leaders.")
+    
+    col3, col4 = st.columns([1, 4])
+    with col3:
+        if st.button("Apply Revisions", type="primary"):
+            with st.spinner("Junior Editor is revising the sequence..."):
+                new_edl_segments, used_model, used_endpoint = call_gemini_for_edl(
+                    st.session_state.transcripts, 
+                    revision_brief, 
+                    ACTIVE_GEMINI_KEY, 
+                    previous_edit=st.session_state.edl_segments
+                )
+                
+                if new_edl_segments:
+                    st.session_state.revision_count += 1
+                    safe_title = st.session_state.safe_title
+                    
+                    if st.session_state.ext == "xml":
+                        final_output = generate_xml(f"{safe_title}_V{st.session_state.revision_count}", new_edl_segments, st.session_state.clip_metadata, fps)
+                    else:
+                        final_output = generate_cmx_edl(f"{safe_title}_V{st.session_state.revision_count}", new_edl_segments, st.session_state.clip_metadata, fps)
+                        
+                    transcript_txt = generate_transcript_txt(f"{safe_title}_V{st.session_state.revision_count}", new_edl_segments, st.session_state.clip_metadata, fps, st.session_state.transcripts)
+                    
+                    st.session_state.final_output = final_output
+                    st.session_state.transcript_txt = transcript_txt
+                    st.session_state.used_model = used_model
+                    st.session_state.edl_segments = new_edl_segments
+                    st.rerun()
+    with col4:
+        if st.button("Start Over / Clear Bin"):
+            for key in ["transcripts", "clip_metadata", "edit_generated", "revision_count"]:
+                if key in st.session_state: del st.session_state[key]
+            st.rerun()