Make vision analysis detailed for downstream AI consumption

- Expanded system prompt with structured 6-point analysis
- Increased max_output_tokens from 300 to 1000
- Description now covers game, actions, UI, camera, overlays, text

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

analyzer.py

@@ -4,19 +4,29 @@ from google import genai
 from google.genai import types
 
 SYSTEM_PROMPT_RU = (
-    "Ты анализируешь кадры с Twitch-стрима. "
-    "Кратко опиши что происходит на экране: игра, действия стримера, "
-    "интерфейс, чат, оверлеи. Будь лаконичен (2-3 предложения). "
-    "Если ничего не изменилось по сравнению с предыдущим описанием, "
-    "скажи 'Без изменений' и уточни только новые детали."
+    "Ты анализируешь кадры с Twitch-стрима. Дай подробное описание всего, что видишь на экране:\n"
+    "1. Какая игра/приложение на экране, жанр, сеттинг\n"
+    "2. Что конкретно происходит: действия персонажа, ситуация в игре, этап (меню, геймплей, катсцена, лобби)\n"
+    "3. Элементы интерфейса: HUD, здоровье, инвентарь, мини-карта, счёт, таймеры\n"
+    "4. Камера стримера: что видно, эмоции, жесты (если есть)\n"
+    "5. Оверлеи: донаты, алерты, виджеты, чат\n"
+    "6. Текст на экране: любой читаемый текст, названия, никнеймы\n"
+    "Пиши развёрнуто (5-10 предложений). Описание должно быть достаточно детальным, "
+    "чтобы другая AI-модель могла полностью понять контекст происходящего без просмотра изображения.\n"
+    "Если сцена похожа на предыдущую, опиши только изменения, но подробно."
 )
 
 SYSTEM_PROMPT_EN = (
-    "You are analyzing frames from a Twitch stream. "
-    "Briefly describe what's happening on screen: game, streamer actions, "
-    "UI, chat, overlays. Be concise (2-3 sentences). "
-    "If nothing changed compared to the previous description, "
-    "say 'No changes' and only note new details."
+    "You are analyzing frames from a Twitch stream. Give a detailed description of everything on screen:\n"
+    "1. What game/application is shown, genre, setting\n"
+    "2. What exactly is happening: character actions, game situation, stage (menu, gameplay, cutscene, lobby)\n"
+    "3. UI elements: HUD, health, inventory, minimap, score, timers\n"
+    "4. Streamer camera: what's visible, emotions, gestures (if present)\n"
+    "5. Overlays: donations, alerts, widgets, chat\n"
+    "6. On-screen text: any readable text, names, nicknames\n"
+    "Write in detail (5-10 sentences). The description must be detailed enough "
+    "for another AI model to fully understand the context without seeing the image.\n"
+    "If the scene is similar to the previous one, describe only the changes, but in detail."
 )
 
 
@@ -67,7 +77,7 @@ class VisionAnalyzer:
             contents=contents,
             config=types.GenerateContentConfig(
                 system_instruction=self.system_prompt,
-                max_output_tokens=300,
+                max_output_tokens=1000,
                 temperature=0.3,
             ),
         )

capture.py

@@ -1,6 +1,10 @@
 import asyncio
 from collections.abc import AsyncIterator
 
+from rich.console import Console
+
+console = Console()
+
 
 async def _pipe_stream(source: asyncio.StreamReader, dest: asyncio.StreamWriter):
     """Forward data from streamlink stdout to ffmpeg stdin."""
@@ -17,6 +21,17 @@ async def _pipe_stream(source: asyncio.StreamReader, dest: asyncio.StreamWriter)
         dest.close()
 
 
+async def _log_stderr(proc_name: str, stderr: asyncio.StreamReader):
+    """Read and display stderr from a subprocess."""
+    while True:
+        line = await stderr.readline()
+        if not line:
+            break
+        text = line.decode("utf-8", errors="replace").rstrip()
+        if text:
+            console.print(f"[dim red][{proc_name}] {text}[/dim red]")
+
+
 async def capture_frames(
     channel: str, quality: str, interval: int
 ) -> AsyncIterator[bytes]:
@@ -33,6 +48,7 @@ async def capture_frames(
 
     ffmpeg_cmd = [
         "ffmpeg",
+        "-loglevel", "warning",
         "-i", "pipe:0",
         "-vf", f"fps=1/{interval}",
         "-f", "image2pipe",
@@ -41,24 +57,43 @@ async def capture_frames(
         "pipe:1",
     ]
 
+    console.print("[dim]Starting streamlink...[/dim]")
     streamlink_proc = await asyncio.create_subprocess_exec(
         *streamlink_cmd,
         stdout=asyncio.subprocess.PIPE,
-        stderr=asyncio.subprocess.DEVNULL,
+        stderr=asyncio.subprocess.PIPE,
     )
 
+    # Wait a moment and check if streamlink started OK
+    await asyncio.sleep(2)
+    if streamlink_proc.returncode is not None:
+        stderr_out = await streamlink_proc.stderr.read()
+        raise RuntimeError(
+            f"streamlink exited with code {streamlink_proc.returncode}: "
+            f"{stderr_out.decode('utf-8', errors='replace')}"
+        )
+
+    console.print("[dim]Starting ffmpeg...[/dim]")
     ffmpeg_proc = await asyncio.create_subprocess_exec(
         *ffmpeg_cmd,
         stdin=asyncio.subprocess.PIPE,
         stdout=asyncio.subprocess.PIPE,
-        stderr=asyncio.subprocess.DEVNULL,
+        stderr=asyncio.subprocess.PIPE,
     )
 
+    # Log stderr from both processes
+    stderr_tasks = [
+        asyncio.create_task(_log_stderr("streamlink", streamlink_proc.stderr)),
+        asyncio.create_task(_log_stderr("ffmpeg", ffmpeg_proc.stderr)),
+    ]
+
     # Forward streamlink → ffmpeg in background
     pipe_task = asyncio.create_task(
         _pipe_stream(streamlink_proc.stdout, ffmpeg_proc.stdin)
     )
 
+    console.print("[dim]Pipeline running, waiting for first frame...[/dim]")
+
     try:
         buf = b""
         while True:
@@ -82,6 +117,8 @@ async def capture_frames(
                 yield frame
     finally:
         pipe_task.cancel()
+        for t in stderr_tasks:
+            t.cancel()
         for proc in (ffmpeg_proc, streamlink_proc):
             try:
                 proc.terminate()

main.py

@@ -28,20 +28,26 @@ async def run(config) -> None:
 
     frame_number = 0
 
-    async for frame_data in capture_frames(
-        config.channel, config.quality, config.interval
-    ):
-        frame_number += 1
-        console.print(f"[dim]Captured frame #{frame_number}, analyzing...[/dim]")
+    try:
+        async for frame_data in capture_frames(
+            config.channel, config.quality, config.interval
+        ):
+            frame_number += 1
+            console.print(f"[dim]Captured frame #{frame_number}, analyzing...[/dim]")
 
-        try:
-            description = await analyzer.analyze_frame(frame_data)
-        except Exception as e:
-            console.print(f"[bold red]Analysis error:[/bold red] {e}")
-            continue
+            try:
+                description = await analyzer.analyze_frame(frame_data)
+            except Exception as e:
+                console.print(f"[bold red]Analysis error:[/bold red] {e}")
+                continue
 
-        print_description(description, frame_number)
-        await log_description(config.log_file, description, frame_number)
+            print_description(description, frame_number)
+            await log_description(config.log_file, description, frame_number)
+    except RuntimeError as e:
+        console.print(f"[bold red]Error:[/bold red] {e}")
+    finally:
+        if frame_number == 0:
+            console.print("[bold yellow]No frames were captured.[/bold yellow]")
 
 
 def main() -> None: