Fix: multi-segment short video - each key point shown with its own image

ai_ext.py

@@ -547,14 +547,22 @@ def _generate_tts_gtts(text: str, out_path: str):
     gTTS(text, lang="vi").save(out_path)
 
 
-# ===== SHORT VIDEO GENERATION =====
+# ===== SHORT VIDEO GENERATION (multi-segment: each key point with its own image) =====
 def _download_image(url, fallback_topic, out_path):
+    """Download an image (un-proxying our own /api/proxy/img). Falls back to generated image."""
     if url:
+        u = url
+        m = re.search(r'/api/proxy/img\?url=(.+)$', u)
+        if m:
+            from urllib.parse import unquote
+            u = unquote(m.group(1))
         try:
-            r = requests.get(url, headers=HEADERS, timeout=15)
+            r = requests.get(u, headers={**HEADERS, "Referer": "https://dantri.com.vn/"}, timeout=15)
             if r.status_code == 200 and len(r.content) > 1000:
                 with open(out_path, "wb") as f:
                     f.write(r.content)
+                if Image:
+                    Image.open(out_path).verify()
                 return out_path
         except Exception:
             pass
@@ -568,116 +576,196 @@ def _download_image(url, fallback_topic, out_path):
     except Exception:
         pass
     if Image:
-        Image.new("RGB", (1080, 860), (30, 55, 42)).save(out_path)
+        Image.new("RGB", (1080, 980), (30, 55, 42)).save(out_path)
         return out_path
     raise RuntimeError("Không tạo được ảnh")
 
-def _make_short_frame(post, img_path, out_path):
+
+def _split_keypoint_sentences(text, max_points=6):
+    """Split summary text into key points: prefer bullet markers, else sentences."""
+    text = _clean_text(text)
+    parts = re.split(r'\s*•\s*', text)
+    good = [p.strip() for p in parts if len(p.strip()) > 20]
+    if len(good) >= 2:
+        # Explicit bullet points: keep each one as-is (never merge).
+        return good[:max_points]
+    # Fallback: split into sentences and merge orphan short fragments.
+    pts = [p.strip() for p in re.split(r'(?<=[.!?])\s+', text) if len(p.strip()) > 20]
+    out = []
+    for p in pts:
+        if out and len(p) < 40:
+            out[-1] = (out[-1] + " " + p).strip()
+        else:
+            out.append(p)
+    return out[:max_points] if out else ([text] if text else [])
+
+
+def _build_keypoints(post, max_points=6):
+    """Return [{text, image}] pairing each key point with its own image."""
+    slides = post.get("slides") or []
+    images = post.get("images") or ([post.get("img")] if post.get("img") else [])
+    images = [i for i in images if i]
+    if slides:
+        kps = []
+        for i, s in enumerate(slides[:max_points]):
+            t = _clean_text(s.get("text", ""))
+            img = s.get("image") or (images[i] if i < len(images) else (images[-1] if images else ""))
+            if t:
+                kps.append({"text": t, "image": img})
+        if kps:
+            return kps
+    points = _split_keypoint_sentences(post.get("text", ""), max_points)
+    kps = []
+    for i, t in enumerate(points):
+        img = images[i] if i < len(images) else (images[-1] if images else "")
+        kps.append({"text": t, "image": img})
+    if not kps:
+        kps = [{"text": _clean_text(post.get("title", "")) or "VNEWS", "image": images[0] if images else ""}]
+    return kps
+
+
+def _wrap_text(draw, text, font, max_w):
+    words = text.split()
+    lines, cur = [], ""
+    for w in words:
+        test = (cur + " " + w).strip()
+        if draw.textlength(test, font=font) <= max_w:
+            cur = test
+        else:
+            if cur:
+                lines.append(cur)
+            cur = w
+    if cur:
+        lines.append(cur)
+    return lines
+
+
+def _make_segment_frame(title, point_text, img_path, idx, total, out_path):
+    """Render a 1080x1920 vertical frame: image on top, key point text below."""
     if Image is None:
         raise RuntimeError("Pillow chưa sẵn sàng")
     W, H = 1080, 1920
-    bg = Image.new("RGB", (W, H), (14, 14, 14))
+    IMG_H = 980
+    bg = Image.new("RGB", (W, H), (12, 14, 18))
     try:
         im = Image.open(img_path).convert("RGB")
-        target = (1080, 860)
-        im_ratio = im.width / im.height
-        target_ratio = target[0] / target[1]
-        if im_ratio > target_ratio:
-            new_h = target[1]; new_w = int(new_h * im_ratio)
+        tr = W / IMG_H
+        ir = im.width / im.height
+        if ir > tr:
+            nh = IMG_H; nw = int(nh * ir)
         else:
-            new_w = target[0]; new_h = int(new_w / im_ratio)
-        im = im.resize((new_w, new_h))
-        left = (new_w - target[0]) // 2; top = (new_h - target[1]) // 2
-        im = im.crop((left, top, left + target[0], top + target[1]))
+            nw = W; nh = int(nw / ir)
+        im = im.resize((nw, nh))
+        left = (nw - W) // 2; top = (nh - IMG_H) // 2
+        im = im.crop((left, top, left + W, top + IMG_H))
         bg.paste(im, (0, 0))
     except Exception:
         pass
     draw = ImageDraw.Draw(bg)
+    draw.rectangle((0, IMG_H, W, H), fill=(12, 14, 18))
     try:
-        font_title = ImageFont.truetype("/usr/share/fonts/truetype/dejavu/DejaVuSans-Bold.ttf", 52)
-        font_body = ImageFont.truetype("/usr/share/fonts/truetype/dejavu/DejaVuSans.ttf", 40)
-        font_label = ImageFont.truetype("/usr/share/fonts/truetype/dejavu/DejaVuSans-Bold.ttf", 32)
+        f_label = ImageFont.truetype("/usr/share/fonts/truetype/dejavu/DejaVuSans-Bold.ttf", 34)
+        f_title = ImageFont.truetype("/usr/share/fonts/truetype/dejavu/DejaVuSans-Bold.ttf", 46)
+        f_point = ImageFont.truetype("/usr/share/fonts/truetype/dejavu/DejaVuSans.ttf", 52)
     except Exception:
-        font_title = font_body = font_label = None
-    draw.rectangle((0, 780, W, H), fill=(14, 14, 14))
-    draw.text((54, 830), "VNEWS · Tường AI", fill=(92, 184, 122), font=font_label)
-    title = post.get("title", "")
-    words = title.split(); lines_t = []; cur = ""
-    for w in words:
-        if len(cur) + len(w) + 1 <= 24:
-            cur = (cur + " " + w).strip()
-        else:
-            if cur: lines_t.append(cur)
-            cur = w
-    if cur: lines_t.append(cur)
-    draw.multiline_text((54, 900), "\n".join(lines_t[:3]), fill=(255, 255, 255), font=font_title, spacing=10)
-    body_text = post.get("text", "")
-    words_b = body_text.split(); lines_b = []; cur_b = ""
-    for w in words_b:
-        if len(cur_b) + len(w) + 1 <= 34:
-            cur_b = (cur_b + " " + w).strip()
-        else:
-            if cur_b: lines_b.append(cur_b)
-            cur_b = w
-        if len(lines_b) >= 10:
-            break
-    if cur_b and len(lines_b) < 10: lines_b.append(cur_b)
-    draw.multiline_text((54, 1120), "\n".join(lines_b), fill=(220, 220, 220), font=font_body, spacing=12)
+        f_label = f_title = f_point = ImageFont.load_default()
+    draw.text((54, IMG_H + 24), "VNEWS · Tường AI", fill=(92, 184, 122), font=f_label)
+    cnt = f"{idx + 1}/{total}"
+    draw.text((W - 54 - draw.textlength(cnt, font=f_label), IMG_H + 24), cnt, fill=(240, 192, 64), font=f_label)
+    y = IMG_H + 86
+    for ln in _wrap_text(draw, _clean_text(title), f_title, W - 108)[:2]:
+        draw.text((54, y), ln, fill=(255, 255, 255), font=f_title)
+        y += 56
+    y += 16
+    for ln in _wrap_text(draw, _clean_text(point_text), f_point, W - 108)[:11]:
+        draw.text((54, y), ln, fill=(225, 230, 235), font=f_point)
+        y += 64
     bg.save(out_path, quality=92)
+    return out_path
+
+
+def _ffmpeg_bin():
+    return os.environ.get("FFMPEG_BIN", "ffmpeg")
+
+
+def _audio_duration(path):
+    try:
+        out = subprocess.run([_ffmpeg_bin(), "-i", path], capture_output=True, text=True, timeout=30).stderr
+        m = re.search(r"Duration:\s*(\d+):(\d+):(\d+\.\d+)", out)
+        if m:
+            h, mi, s = m.groups()
+            return int(h) * 3600 + int(mi) * 60 + float(s)
+    except Exception:
+        pass
+    return 0.0
 
-def _short_script(post, max_chars=700):
-    """Build TTS script: title + text (concise, within char limit)."""
-    txt = _clean_text(post.get("text", ""))
-    if len(txt) > max_chars:
-        txt = txt[:max_chars].rsplit(" ", 1)[0] + "."
-    title = _clean_text(post.get("title", ""))
-    return f"{title}. {txt}"
 
 async def _generate_short_video(post, post_id: str, voice_id: str = None, speed: float = None) -> str:
-    """Generate MP4 short video with TTS. Returns video URL or empty string on failure."""
+    """Generate a multi-segment MP4 short: each key point shown with its OWN image + narration."""
     try:
         os.makedirs(SHORTS_DIR, exist_ok=True)
         out_mp4 = os.path.join(SHORTS_DIR, _safe_name(post_id) + ".mp4")
-        if os.path.exists(out_mp4):
+        if os.path.exists(out_mp4) and voice_id is None and speed is None:
             return "/api/ai/short-file/" + post_id
 
-        work = os.path.join(SHORTS_DIR, _safe_name(post_id))
+        work = os.path.join(SHORTS_DIR, _safe_name(post_id) + "_work")
         os.makedirs(work, exist_ok=True)
-        img_path = os.path.join(work, "image.jpg")
-        frame_path = os.path.join(work, "frame.jpg")
-        audio_path = os.path.join(work, "voice.mp3")
 
-        _download_image(post.get("img"), post.get("title", "AI news"), img_path)
-        _make_short_frame(post, img_path, frame_path)
-        script = _short_script(post)
+        title = _clean_text(post.get("title", "")) or "VNEWS"
+        kps = _build_keypoints(post)
 
-        # Auto-detect voice if not specified
+        # Resolve voice
         if voice_id is None:
             voice_id = _detect_voice_for_topic(post.get("title", ""), post.get("text", ""))
         if speed is None:
             speed = TTS_DEFAULT_SPEED
+        vcfg = TTS_VOICES.get(voice_id, TTS_VOICES[TTS_DEFAULT_VOICE])
+
+        seg_files = []
+        ff = _ffmpeg_bin()
+        for i, kp in enumerate(kps):
+            img_path = os.path.join(work, f"img{i}.jpg")
+            frame_path = os.path.join(work, f"frame{i}.jpg")
+            audio_path = os.path.join(work, f"voice{i}.mp3")
+            seg_mp4 = os.path.join(work, f"seg{i}.mp4")
+            _download_image(kp.get("image", ""), title, img_path)
+            _make_segment_frame(title, kp["text"], img_path, i, len(kps), frame_path)
+            narration = (title + ". " + kp["text"]) if i == 0 else kp["text"]
+            try:
+                await _generate_tts_edge(narration, voice_id, speed, audio_path)
+            except Exception as e:
+                print(f"[TTS edge-tts error] {e}, falling back to gTTS")
+                if gTTS:
+                    _generate_tts_gtts(narration, audio_path)
+                else:
+                    return ""
+            dur = _audio_duration(audio_path)
+            if dur < 1.0:
+                dur = 2.0
+            cmd = [ff, "-y", "-loop", "1", "-i", frame_path, "-i", audio_path,
+                   "-c:v", "libx264", "-tune", "stillimage", "-pix_fmt", "yuv420p",
+                   "-t", f"{dur + 0.4:.2f}", "-c:a", "aac", "-b:a", "128k", "-ar", "44100",
+                   "-vf", "scale=1080:1920", "-r", "25", seg_mp4]
+            subprocess.run(cmd, check=True, stdout=subprocess.PIPE, stderr=subprocess.PIPE, timeout=180)
+            seg_files.append(seg_mp4)
+
+        if not seg_files:
+            return ""
+        if len(seg_files) == 1:
+            os.replace(seg_files[0], out_mp4)
+            return "/api/ai/short-file/" + post_id
 
-        # Generate TTS
-        try:
-            await _generate_tts_edge(script, voice_id, speed, audio_path)
-        except Exception as e:
-            print(f"[TTS edge-tts error] {e}, falling back to gTTS")
-            if gTTS:
-                _generate_tts_gtts(script, audio_path)
-            else:
-                return ""
-
-        # Assemble video
-        cmd = ["ffmpeg", "-y", "-loop", "1", "-i", frame_path, "-i", audio_path,
-               "-shortest", "-c:v", "libx264", "-tune", "stillimage", "-pix_fmt", "yuv420p",
-               "-c:a", "aac", "-b:a", "128k", "-vf", "scale=1080:1920", out_mp4]
-        subprocess.run(cmd, check=True, stdout=subprocess.PIPE, stderr=subprocess.PIPE, timeout=180)
+        listfile = os.path.join(work, "concat.txt")
+        with open(listfile, "w", encoding="utf-8") as f:
+            f.write("\n".join(f"file '{s}'" for s in seg_files))
+        cmd = [ff, "-y", "-f", "concat", "-safe", "0", "-i", listfile,
+               "-c:v", "libx264", "-pix_fmt", "yuv420p", "-c:a", "aac", "-b:a", "128k", out_mp4]
+        subprocess.run(cmd, check=True, stdout=subprocess.PIPE, stderr=subprocess.PIPE, timeout=300)
         return "/api/ai/short-file/" + post_id
     except Exception as e:
         print(f"[short video error] {e}")
         return ""
 
-
 # ===== MAKE POST =====
 def make_post(title, text, image, source_url, kind, sources=None, images=None):
     return {