Restore Space to revision 33c3dda

ai_patch.py

@@ -518,119 +518,8 @@ def _make_short_frame_full(post, img_path, out_path):
     bg.save(out_path, quality=92)
 
 
-
-
-def _summary_segments_from_post(post, max_segments=7):
-    """Extract clean summary segments. No old context; each bullet/paragraph becomes a scene."""
-    raw = _clean(post.get('text') or post.get('title') or '')
-    raw = re.sub(r'^Bản tin AI viết lại:\s*', '', raw, flags=re.I)
-    raw = re.sub(r'Nguồn tham khảo:.*$', '', raw, flags=re.I|re.S).strip()
-    # Prefer bullet points if available
-    lines = []
-    for ln in raw.splitlines():
-        ln = _clean(re.sub(r'^[•\-\*\d\.\)\s]+', '', ln))
-        if not ln: continue
-        if len(ln) >= 18 and not ln.lower().startswith(('điểm chính', 'tiêu đề', 'sapo')):
-            lines.append(ln)
-    if len(lines) < 2:
-        # split into sentences/clauses
-        lines = []
-        for s in re.split(r'(?<=[\.\!\?])\s+', raw):
-            s = _clean(s)
-            if len(s) >= 25:
-                lines.append(s)
-    segs = _dedupe_units(lines, max_units=max_segments)
-    if not segs:
-        segs = [post.get('title','Bản tin VNEWS')]
-    return segs[:max_segments]
-
-
-def _make_scene_frame(post, segment, idx, total, img_path, out_path, emotion='neutral'):
-    """Create one 9:16 frame for a single summary segment."""
-    if Image is None:
-        return _make_short_frame_full(post, img_path, out_path)
-    W, H = 1080, 1920
-    bg = Image.new('RGB', (W, H), (10, 10, 10))
-    try:
-        im = Image.open(img_path).convert('RGB')
-        # background blur/cover approximation: cover whole canvas
-        ratio = im.width / max(1, im.height)
-        target_ratio = W / H
-        if ratio > target_ratio:
-            nh = H; nw = int(nh * ratio)
-        else:
-            nw = W; nh = int(nw / ratio)
-        im2 = im.resize((nw, nh))
-        left=(nw-W)//2; top=(nh-H)//2
-        im2 = im2.crop((left, top, left+W, top+H))
-        bg.paste(im2, (0,0))
-        # dark overlay
-        overlay = Image.new('RGB',(W,H),(0,0,0))
-        bg = Image.blend(bg, overlay, 0.48)
-        # hero image block top
-        hero_h = 720
-        ratio = im.width / max(1, im.height)
-        target_ratio = W / hero_h
-        if ratio > target_ratio:
-            nh = hero_h; nw = int(nh * ratio)
-        else:
-            nw = W; nh = int(nw / ratio)
-        hero = im.resize((nw, nh))
-        left=(nw-W)//2; top=(nh-hero_h)//2
-        hero = hero.crop((left, top, left+W, top+hero_h))
-        bg.paste(hero,(0,0))
-    except Exception:
-        pass
-    draw = ImageDraw.Draw(bg)
-    try:
-        font_brand = ImageFont.truetype('/usr/share/fonts/truetype/dejavu/DejaVuSans-Bold.ttf', 34)
-        font_label = ImageFont.truetype('/usr/share/fonts/truetype/dejavu/DejaVuSans-Bold.ttf', 32)
-        font_seg = ImageFont.truetype('/usr/share/fonts/truetype/dejavu/DejaVuSans-Bold.ttf', 58)
-        font_title = ImageFont.truetype('/usr/share/fonts/truetype/dejavu/DejaVuSans.ttf', 34)
-        font_small = ImageFont.truetype('/usr/share/fonts/truetype/dejavu/DejaVuSans.ttf', 28)
-    except Exception:
-        font_brand = font_label = font_seg = font_title = font_small = None
-    # gradient/dark content panel
-    draw.rectangle((0,680,W,H), fill=(12,12,12))
-    # progress dots
-    dot_y=742; dot_x=48
-    for i in range(total):
-        fill=(92,184,122) if i==idx else (70,70,70)
-        draw.rounded_rectangle((dot_x+i*38,dot_y,dot_x+i*38+24,dot_y+10), radius=5, fill=fill)
-    draw.text((48, 780), 'VNEWS AI SHORT', fill=(110, 231, 143), font=font_brand)
-    topic = emotion or 'Tin tức'
-    draw.rounded_rectangle((48, 834, 260, 880), radius=20, fill=(28,70,45))
-    draw.text((66, 842), f'Đoạn {idx+1}/{total}', fill=(235,235,235), font=font_small)
-    y = 940
-    maxw = W - 96
-    for ln in _wrap_text_px(draw, segment, font_seg, maxw, 8):
-        draw.text((48, y), ln, fill=(255,255,255), font=font_seg)
-        y += 74
-        if y > 1500: break
-    # Title at bottom as small context only, not full article
-    title = post.get('title','')
-    y2 = 1640
-    draw.line((48,y2-22,W-48,y2-22), fill=(70,70,70), width=2)
-    for ln in _wrap_text_px(draw, title, font_title, maxw, 3):
-        draw.text((48, y2), ln, fill=(220,220,220), font=font_title)
-        y2 += 46
-    bg.save(out_path, quality=92)
-
-
-def _estimate_audio_duration(path, fallback=4.0):
-    try:
-        pr = subprocess.run(['ffprobe','-v','error','-show_entries','format=duration','-of','default=noprint_wrappers=1:nokey=1',path], stdout=subprocess.PIPE, stderr=subprocess.PIPE, timeout=20)
-        return max(1.5, float((pr.stdout or b'').decode().strip() or fallback))
-    except Exception:
-        return fallback
-
-
 @app.post('/api/ai/short/{post_id}')
 async def patched_ai_short(post_id: str, request: Request):
-    """Create 9:16 AI short as multiple timed scenes.
-    Each summary paragraph/bullet becomes its own visual scene.
-    No subtitle burn-in; only scene text is rendered on frames.
-    """
     try:
         body = await request.json()
     except Exception:
@@ -645,28 +534,31 @@ async def patched_ai_short(post_id: str, request: Request):
     if not post:
         return JSONResponse({'error': 'post not found'}, status_code=404)
 
-    segments = _summary_segments_from_post(post, max_segments=7)
-    seg_hash = hashlib.md5(('|'.join(segments)+voice+emotion+str(speed)).encode('utf-8')).hexdigest()[:8]
     os.makedirs(base.SHORTS_DIR, exist_ok=True)
-    suffix = f"_{voice}_{emotion}_{str(speed).replace('.', 'p')}_{seg_hash}_scenes_nosub"
+    suffix = f"_{voice}_{emotion}_{str(speed).replace('.', 'p')}"
     out_mp4 = os.path.join(base.SHORTS_DIR, base._safe_name(post_id + suffix) + '.mp4')
     if os.path.exists(out_mp4):
         post['video'] = '/api/ai/short-file/' + post_id + suffix
         post['short_voice'] = voice
         post['short_emotion'] = emotion
         post['short_speed'] = speed
-        post['short_segments'] = segments
-        post['short_subtitles'] = False
         base._save_ai_wall(posts)
-        return JSONResponse({'video': post['video'], 'voice': voice, 'emotion': emotion, 'speed': speed, 'subtitles': False, 'segments': segments})
+        return JSONResponse({'video': post['video'], 'voice': voice, 'emotion': emotion, 'speed': speed})
     if base.gTTS is None:
         return JSONResponse({'error': 'gTTS chưa sẵn sàng'}, status_code=503)
 
     work = os.path.join(base.SHORTS_DIR, base._safe_name(post_id + suffix))
     os.makedirs(work, exist_ok=True)
     img = os.path.join(work, 'image.jpg')
+    frame = os.path.join(work, 'frame.jpg')
+    audio = os.path.join(work, 'voice.mp3')
+    audio_fast = os.path.join(work, 'voice_fast.mp3')
+    srt = os.path.join(work, 'subtitles.srt')
     try:
         base._download_image(post.get('img'), post.get('title', 'AI news'), img)
+        _make_short_frame_full(post, img, frame)
+        script = _tts_script_smart(post, emotion)
+        # True selectable voice via edge-tts if installed. Fallback to gTTS.
         edge_voice = {
             'nam': 'vi-VN-NamMinhNeural',
             'male': 'vi-VN-NamMinhNeural',
@@ -674,42 +566,36 @@ async def patched_ai_short(post_id: str, request: Request):
             'female': 'vi-VN-HoaiMyNeural',
             'mien-nam': 'vi-VN-HoaiMyNeural',
         }.get(voice, 'vi-VN-HoaiMyNeural')
-        part_files=[]
-        for idx, seg in enumerate(segments):
-            frame = os.path.join(work, f'frame_{idx:02d}.jpg')
-            aud = os.path.join(work, f'voice_{idx:02d}.mp3')
-            aud_fast = os.path.join(work, f'voice_{idx:02d}_fast.mp3')
-            part = os.path.join(work, f'part_{idx:02d}.mp4')
-            _make_scene_frame(post, seg, idx, len(segments), img, frame, emotion=emotion)
-            spoken = _emotion_script(seg, emotion)
+        try:
+            subprocess.run(['python', '-m', 'edge_tts', '--voice', edge_voice, '--text', script, '--write-media', audio], check=True, stdout=subprocess.PIPE, stderr=subprocess.PIPE, timeout=160)
+        except Exception:
+            tld = 'com.vn' if voice in ('nu', 'female', 'mien-nam') else 'com'
             try:
-                subprocess.run(['python','-m','edge_tts','--voice',edge_voice,'--text',spoken,'--write-media',aud], check=True, stdout=subprocess.PIPE, stderr=subprocess.PIPE, timeout=120)
-            except Exception:
-                tld = 'com.vn' if voice in ('nu','female','mien-nam') else 'com'
-                try:
-                    base.gTTS(spoken, lang='vi', tld=tld, slow=False).save(aud)
-                except TypeError:
-                    base.gTTS(spoken, lang='vi', slow=False).save(aud)
-            subprocess.run(['ffmpeg','-y','-i',aud,'-filter:a',f'atempo={speed}','-vn',aud_fast], check=True, stdout=subprocess.PIPE, stderr=subprocess.PIPE, timeout=90)
-            dur = _estimate_audio_duration(aud_fast, fallback=4.0) + 0.35
-            # No subtitles. Text is already part of the scene frame.
-            subprocess.run(['ffmpeg','-y','-loop','1','-t',str(dur),'-i',frame,'-i',aud_fast,'-shortest','-c:v','libx264','-tune','stillimage','-pix_fmt','yuv420p','-c:a','aac','-b:a','128k',part], check=True, stdout=subprocess.PIPE, stderr=subprocess.PIPE, timeout=150)
-            part_files.append(part)
-        concat = os.path.join(work,'concat.txt')
-        with open(concat,'w',encoding='utf-8') as f:
-            for p in part_files:
-                f.write("file '" + p.replace("'", "'\\''") + "'\n")
-        subprocess.run(['ffmpeg','-y','-f','concat','-safe','0','-i',concat,'-c','copy',out_mp4], check=True, stdout=subprocess.PIPE, stderr=subprocess.PIPE, timeout=180)
+                base.gTTS(script, lang='vi', tld=tld, slow=False).save(audio)
+            except TypeError:
+                base.gTTS(script, lang='vi', slow=False).save(audio)
+        subprocess.run(['ffmpeg', '-y', '-i', audio, '-filter:a', f'atempo={speed}', '-vn', audio_fast], check=True, stdout=subprocess.PIPE, stderr=subprocess.PIPE, timeout=120)
+        # Estimate duration for subtitle timing.
+        duration = 30.0
+        try:
+            pr = subprocess.run(['ffprobe', '-v', 'error', '-show_entries', 'format=duration', '-of', 'default=noprint_wrappers=1:nokey=1', audio_fast], stdout=subprocess.PIPE, stderr=subprocess.PIPE, timeout=20)
+            duration = float((pr.stdout or b'30').decode().strip() or 30)
+        except Exception:
+            pass
+        _write_srt(script, srt, duration)
+        # Burn subtitles at bottom, with readable style.
+        vf = "scale=1080:1920,subtitles='{}':force_style='FontName=DejaVu Sans,FontSize=34,PrimaryColour=&H00FFFFFF,OutlineColour=&HAA000000,BorderStyle=1,Outline=3,Shadow=1,Alignment=2,MarginV=90'".format(srt.replace("'", "\\'"))
+        cmd = ['ffmpeg', '-y', '-loop', '1', '-i', frame, '-i', audio_fast, '-shortest', '-c:v', 'libx264', '-tune', 'stillimage', '-pix_fmt', 'yuv420p', '-c:a', 'aac', '-b:a', '128k', '-vf', vf, out_mp4]
+        subprocess.run(cmd, check=True, stdout=subprocess.PIPE, stderr=subprocess.PIPE, timeout=220)
         post['video'] = '/api/ai/short-file/' + post_id + suffix
         post['short_voice'] = voice
         post['short_emotion'] = emotion
         post['short_speed'] = speed
-        post['short_segments'] = segments
-        post['short_subtitles'] = False
+        post['short_subtitles'] = True
         base._save_ai_wall(posts)
-        return JSONResponse({'video': post['video'], 'voice': voice, 'emotion': emotion, 'speed': speed, 'subtitles': False, 'segments': segments})
+        return JSONResponse({'video': post['video'], 'voice': voice, 'emotion': emotion, 'speed': speed, 'subtitles': True})
     except Exception as e:
-        return JSONResponse({'error': 'Không tạo được shorts: ' + str(e)[:220]}, status_code=500)
+        return JSONResponse({'error': 'Không tạo được shorts: ' + str(e)[:180]}, status_code=500)
 
 
 @app.get('/api/ai/short-file/{file_id}')