fix(image-gen): chunk base64 SSE to avoid 131072 byte aiter_lines limit

Author: ahmedmarzougui-enclaive

backend/privacy_proxy/app/main.py

@@ -716,8 +716,14 @@ async def proxy(request: Request, path: str):
             b64 = data.get("b64_json", "")
             img_url = f"data:image/png;base64,{b64}" if b64 else ""
         content = f"![image]({img_url})"
-        sse = f'data: {{"choices":[{{"delta":{{"role":"assistant","content":{orjson.dumps(content).decode()}}},"index":0}}]}}\n\ndata: [DONE]\n\n'
-        return StreamingResponse(iter([sse.encode()]), media_type="text/event-stream")
+        CHUNK = 60000
+        parts = [content[i:i+CHUNK] for i in range(0, len(content), CHUNK)]
+        sse_events = []
+        for i, part in enumerate(parts):
+            delta = {"role": "assistant", "content": part} if i == 0 else {"content": part}
+            sse_events.append(f'data: {{"choices":[{{"delta":{orjson.dumps(delta).decode()},"index":0}}]}}\n\n'.encode())
+        sse_events.append(b"data: [DONE]\n\n")
+        return StreamingResponse(iter(sse_events), media_type="text/event-stream")
 
     if is_chat and body and not is_openai:
         async with httpx.AsyncClient() as client: