Spaces:
Running
Running
Summaries only and add AI shorts video generation from wall posts
Browse files
ai_ext.py
CHANGED
|
@@ -1,10 +1,10 @@
|
|
| 1 |
-
import os, re, json, time, random, html as html_lib
|
| 2 |
from urllib.parse import quote_plus, quote, urlparse
|
| 3 |
from typing import Optional
|
| 4 |
import requests
|
| 5 |
from bs4 import BeautifulSoup
|
| 6 |
from fastapi import Request
|
| 7 |
-
from fastapi.responses import HTMLResponse, JSONResponse
|
| 8 |
|
| 9 |
from main import app
|
| 10 |
|
|
@@ -12,8 +12,16 @@ try:
|
|
| 12 |
from huggingface_hub import AsyncInferenceClient
|
| 13 |
except Exception:
|
| 14 |
AsyncInferenceClient = None
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 15 |
|
| 16 |
-
# Accept all common HF token env names used by Spaces/Hub examples.
|
| 17 |
def _hf_token():
|
| 18 |
for k in ("HF_TOKEN", "HUGGINGFACEHUB_API_TOKEN", "HUGGING_FACE_HUB_TOKEN", "HF_API_TOKEN"):
|
| 19 |
v = os.getenv(k, "").strip()
|
|
@@ -23,10 +31,13 @@ def _hf_token():
|
|
| 23 |
|
| 24 |
HF_TOKEN = _hf_token()
|
| 25 |
QWEN_VL_MODEL = os.getenv("QWEN_VL_MODEL", "Qwen/Qwen2.5-VL-7B-Instruct")
|
| 26 |
-
|
|
|
|
|
|
|
| 27 |
HEADERS = {"User-Agent":"Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36","Accept-Language":"vi-VN,vi;q=0.9,en;q=0.8"}
|
| 28 |
LAST_QWEN_ERROR = ""
|
| 29 |
|
|
|
|
| 30 |
def _load_ai_wall():
|
| 31 |
try:
|
| 32 |
if os.path.exists(WALL_FILE):
|
|
@@ -36,6 +47,7 @@ def _load_ai_wall():
|
|
| 36 |
pass
|
| 37 |
return []
|
| 38 |
|
|
|
|
| 39 |
def _save_ai_wall(posts):
|
| 40 |
try:
|
| 41 |
os.makedirs(os.path.dirname(WALL_FILE), exist_ok=True)
|
|
@@ -46,26 +58,31 @@ def _save_ai_wall(posts):
|
|
| 46 |
except Exception:
|
| 47 |
pass
|
| 48 |
|
|
|
|
| 49 |
def _clean_text(s: str) -> str:
|
| 50 |
s = html_lib.unescape(s or "")
|
| 51 |
return re.sub(r"\s+", " ", s).strip()
|
| 52 |
|
|
|
|
| 53 |
def _domain(u):
|
| 54 |
try:
|
| 55 |
return urlparse(u).netloc.replace("www.", "")
|
| 56 |
except Exception:
|
| 57 |
return ""
|
| 58 |
|
|
|
|
| 59 |
def _reader_url(target_url: str) -> str:
|
| 60 |
safe = quote(target_url, safe=":/?#[]@!$&'()*+,;=%")
|
| 61 |
return "https://r.jina.ai/http://" + safe
|
| 62 |
|
|
|
|
| 63 |
def jina_reader_markdown(url: str) -> str:
|
| 64 |
jr = _reader_url(url)
|
| 65 |
r = requests.get(jr, headers={"Accept":"text/markdown,text/plain,*/*", "X-Return-Format":"markdown", "User-Agent":"Mozilla/5.0"}, timeout=35)
|
| 66 |
r.raise_for_status()
|
| 67 |
return r.text or ""
|
| 68 |
|
|
|
|
| 69 |
def _parse_jina_markdown(md: str, url: str):
|
| 70 |
lines = [x.rstrip() for x in (md or "").splitlines()]
|
| 71 |
title = ""; image = ""; content_lines = []; in_content = False
|
|
@@ -92,6 +109,7 @@ def _parse_jina_markdown(md: str, url: str):
|
|
| 92 |
if not title and paras: title = paras[0][:90]
|
| 93 |
return {"url": url, "title": title or url, "summary": paras[0] if paras else "", "text": "\n".join(paras), "image": image, "images": [image] if image else [], "via":"jina"}
|
| 94 |
|
|
|
|
| 95 |
def _best_content_block(soup):
|
| 96 |
best, best_score = None, 0
|
| 97 |
for el in soup.find_all(["article", "main", "section", "div"]):
|
|
@@ -103,6 +121,7 @@ def _best_content_block(soup):
|
|
| 103 |
if score > best_score: best, best_score = el, score
|
| 104 |
return best
|
| 105 |
|
|
|
|
| 106 |
def scrape_any_url_direct(url: str):
|
| 107 |
r = requests.get(url, headers=HEADERS, timeout=18)
|
| 108 |
if r.status_code in {401,403,406,409,429,451,503}: raise RuntimeError(f"blocked status {r.status_code}")
|
|
@@ -135,6 +154,7 @@ def scrape_any_url_direct(url: str):
|
|
| 135 |
if len(data["text"]) < 160: raise RuntimeError("direct scrape too short")
|
| 136 |
return data
|
| 137 |
|
|
|
|
| 138 |
def scrape_any_url(url: str):
|
| 139 |
try:
|
| 140 |
return scrape_any_url_direct(url)
|
|
@@ -145,10 +165,12 @@ def scrape_any_url(url: str):
|
|
| 145 |
raise RuntimeError(f"Jina Reader also returned too little content; direct={direct_error}")
|
| 146 |
return data
|
| 147 |
|
|
|
|
| 148 |
def pollinations_image_url(topic: str, width=1024, height=576):
|
| 149 |
prompt = f"editorial illustration, topic: {topic}, modern clean cinematic news image, high quality, no text, no watermark"
|
| 150 |
return f"https://image.pollinations.ai/prompt/{quote_plus(prompt)}?width={width}&height={height}&seed={random.randint(1,999999)}&nologo=true&private=true"
|
| 151 |
|
|
|
|
| 152 |
def jina_search(query: str, limit=6):
|
| 153 |
for accept in ("application/json", "text/plain"):
|
| 154 |
try:
|
|
@@ -174,6 +196,7 @@ def jina_search(query: str, limit=6):
|
|
| 174 |
pass
|
| 175 |
return []
|
| 176 |
|
|
|
|
| 177 |
def duck_context(topic: str, limit=6):
|
| 178 |
try:
|
| 179 |
url = "https://duckduckgo.com/html/?q=" + quote_plus(topic + " tin tức phân tích bối cảnh")
|
|
@@ -187,6 +210,7 @@ def duck_context(topic: str, limit=6):
|
|
| 187 |
return results
|
| 188 |
except Exception: return []
|
| 189 |
|
|
|
|
| 190 |
def web_context(topic: str, limit=6):
|
| 191 |
queries = [topic + " tin tức phân tích bối cảnh", topic]
|
| 192 |
results=[]
|
|
@@ -206,6 +230,7 @@ def web_context(topic: str, limit=6):
|
|
| 206 |
context = "\n".join([f"- {x.get('title') or _domain(x.get('url',''))} ({_domain(x.get('url',''))}): {x.get('excerpt') or x.get('description','')}" for x in enriched if x.get("excerpt") or x.get("title")])
|
| 207 |
return context, enriched
|
| 208 |
|
|
|
|
| 209 |
async def qwen_generate(prompt: str, image_url: Optional[str] = None, max_tokens: int = 1200):
|
| 210 |
global LAST_QWEN_ERROR, HF_TOKEN
|
| 211 |
HF_TOKEN = _hf_token()
|
|
@@ -215,8 +240,7 @@ async def qwen_generate(prompt: str, image_url: Optional[str] = None, max_tokens
|
|
| 215 |
if not AsyncInferenceClient:
|
| 216 |
LAST_QWEN_ERROR = "Thiếu thư viện huggingface_hub hoặc import lỗi."
|
| 217 |
return None
|
| 218 |
-
errors=[]
|
| 219 |
-
models=[]
|
| 220 |
for m in [QWEN_VL_MODEL, "Qwen/Qwen2.5-VL-7B-Instruct", "Qwen/Qwen2.5-VL-3B-Instruct"]:
|
| 221 |
if m and m not in models: models.append(m)
|
| 222 |
for model in models:
|
|
@@ -229,38 +253,41 @@ async def qwen_generate(prompt: str, image_url: Optional[str] = None, max_tokens
|
|
| 229 |
resp=await client.chat_completion(model=model, messages=messages, max_tokens=max_tokens, temperature=0.45, top_p=0.85)
|
| 230 |
txt=(resp.choices[0].message.content or "").strip()
|
| 231 |
if txt:
|
| 232 |
-
LAST_QWEN_ERROR=""
|
| 233 |
-
return txt
|
| 234 |
except Exception as e:
|
| 235 |
errors.append(f"{model}: {type(e).__name__}: {str(e)[:220]}")
|
| 236 |
LAST_QWEN_ERROR = " | ".join(errors) or "Qwen không trả nội dung."
|
| 237 |
print("[qwen errors]", LAST_QWEN_ERROR)
|
| 238 |
return None
|
| 239 |
|
|
|
|
| 240 |
def make_post(title, text, image, source_url, kind, sources=None):
|
| 241 |
-
return {"id":str(int(time.time()*1000))+str(random.randint(100,999)),"title":title,"text":text,"img":image,"url":source_url,"kind":kind,"sources":sources or [],"ts":int(time.time())}
|
|
|
|
| 242 |
|
| 243 |
@app.get("/api/ai_wall")
|
| 244 |
def api_ai_wall(): return JSONResponse({"posts":_load_ai_wall()[:80]})
|
| 245 |
|
|
|
|
| 246 |
@app.get("/api/ai/status")
|
| 247 |
def api_ai_status():
|
| 248 |
-
return JSONResponse({"has_token":bool(_hf_token()),"client_imported":AsyncInferenceClient is not None,"model":QWEN_VL_MODEL,"last_error":LAST_QWEN_ERROR})
|
|
|
|
| 249 |
|
| 250 |
@app.post("/api/ai/topic")
|
| 251 |
async def api_ai_topic(request:Request):
|
| 252 |
-
body=await request.json();topic=_clean_text(body.get("topic",""))
|
| 253 |
if not topic: return JSONResponse({"error":"missing topic"},status_code=400)
|
| 254 |
ctx,sources=web_context(topic)
|
| 255 |
if not ctx: return JSONResponse({"error":"Không lấy được dữ liệu internet cho chủ đề này. Hãy thử chủ đề cụ thể hơn."},status_code=422)
|
| 256 |
image=pollinations_image_url(topic)
|
| 257 |
-
prompt=f"""
|
| 258 |
|
| 259 |
Chủ đề: {topic}
|
| 260 |
|
| 261 |
Yêu cầu:
|
| 262 |
- Chỉ tóm tắt các ý chính dựa trên nguồn internet bên dưới.
|
| 263 |
-
- Không viết thành bài dài, không
|
| 264 |
- 1 tiêu đề ngắn.
|
| 265 |
- 3-5 gạch đầu dòng, mỗi dòng tối đa 1 câu.
|
| 266 |
- 1 câu kết luận ngắn.
|
|
@@ -270,12 +297,13 @@ Nguồn/bối cảnh internet:
|
|
| 270 |
{ctx}"""
|
| 271 |
text=await qwen_generate(prompt,image_url=image,max_tokens=900)
|
| 272 |
if not text: return JSONResponse({"error":"Qwen2.5-VL chưa sẵn sàng: "+LAST_QWEN_ERROR},status_code=503)
|
| 273 |
-
post=make_post(topic,text,image,"","topic",sources=sources[:5]);posts=_load_ai_wall();posts.insert(0,post);_save_ai_wall(posts)
|
| 274 |
return JSONResponse({"post":post})
|
| 275 |
|
|
|
|
| 276 |
@app.post("/api/ai/url")
|
| 277 |
async def api_ai_url(request:Request):
|
| 278 |
-
body=await request.json();url=_clean_text(body.get("url",""))
|
| 279 |
if not url.startswith("http"): return JSONResponse({"error":"missing url"},status_code=400)
|
| 280 |
try: data=scrape_any_url(url)
|
| 281 |
except Exception as e: return JSONResponse({"error":"Không scrape được URL: "+str(e)[:180]},status_code=422)
|
|
@@ -298,12 +326,13 @@ Nội dung gốc:
|
|
| 298 |
text=await qwen_generate(prompt,image_url=data.get("image") or None,max_tokens=900)
|
| 299 |
if not text: return JSONResponse({"error":"Qwen2.5-VL chưa sẵn sàng: "+LAST_QWEN_ERROR},status_code=503)
|
| 300 |
post=make_post(data.get("title") or "Bài viết",text,data.get("image") or "",url,"url",sources=[{"title":data.get("title"),"url":url,"excerpt":raw[:500],"via":data.get("via")}])
|
| 301 |
-
posts=_load_ai_wall();posts.insert(0,post);_save_ai_wall(posts)
|
| 302 |
return JSONResponse({"post":post})
|
| 303 |
|
|
|
|
| 304 |
@app.post("/api/rewrite_share")
|
| 305 |
async def api_rewrite_share(request:Request):
|
| 306 |
-
body=await request.json();url=_clean_text(body.get("url",""))
|
| 307 |
if not url.startswith("http"): return JSONResponse({"error":"missing url"},status_code=400)
|
| 308 |
try: data=scrape_any_url(url)
|
| 309 |
except Exception as e: return JSONResponse({"error":"Không đọc được bài viết: "+str(e)[:180]},status_code=422)
|
|
@@ -322,13 +351,128 @@ Nội dung:
|
|
| 322 |
text=await qwen_generate(prompt,image_url=data.get("image") or None,max_tokens=900)
|
| 323 |
if not text: return JSONResponse({"error":"Qwen2.5-VL chưa sẵn sàng: "+LAST_QWEN_ERROR},status_code=503)
|
| 324 |
post=make_post(data.get("title") or "Bài viết",text,data.get("image") or "",url,"rewrite")
|
| 325 |
-
posts=_load_ai_wall();posts.insert(0,post);_save_ai_wall(posts)
|
| 326 |
return JSONResponse({"post":post})
|
| 327 |
|
| 328 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 329 |
app.router.routes=[r for r in app.router.routes if not (getattr(r,'path',None)=='/' and 'GET' in getattr(r,'methods',set()))]
|
| 330 |
AI_INJECT=r'''
|
| 331 |
-
<style>.ai-wall-extra{margin:6px 4px;background:#1a1a1a;border:1px solid #2a2a2a;border-radius:8px;overflow:hidden}.ai-wall-card{flex:0 0 250px;background:#141414;border:1px solid #2b2b2b;border-radius:10px;padding:8px}.ai-wall-img{width:100%;aspect-ratio:16/9;background:#222;border-radius:8px;overflow:hidden;margin-bottom:6px}.ai-wall-img img{width:100%;height:100%;object-fit:cover}.ai-wall-title{font-size:12px;color:#5cb87a;font-weight:800;line-height:1.3;margin-bottom:4px}.ai-wall-text{font-size:11px;color:#bbb;line-height:1.45;white-space:pre-wrap;display:-webkit-box;-webkit-line-clamp:5;-webkit-box-orient:vertical;overflow:hidden}.ai-wall-
|
| 332 |
@app.get('/')
|
| 333 |
async def index_ai():
|
| 334 |
with open('/app/static/index.html','r',encoding='utf-8') as f: html=f.read()
|
|
|
|
| 1 |
+
import os, re, json, time, random, html as html_lib, subprocess, uuid
|
| 2 |
from urllib.parse import quote_plus, quote, urlparse
|
| 3 |
from typing import Optional
|
| 4 |
import requests
|
| 5 |
from bs4 import BeautifulSoup
|
| 6 |
from fastapi import Request
|
| 7 |
+
from fastapi.responses import HTMLResponse, JSONResponse, FileResponse
|
| 8 |
|
| 9 |
from main import app
|
| 10 |
|
|
|
|
| 12 |
from huggingface_hub import AsyncInferenceClient
|
| 13 |
except Exception:
|
| 14 |
AsyncInferenceClient = None
|
| 15 |
+
try:
|
| 16 |
+
from gtts import gTTS
|
| 17 |
+
except Exception:
|
| 18 |
+
gTTS = None
|
| 19 |
+
try:
|
| 20 |
+
from PIL import Image, ImageDraw, ImageFont
|
| 21 |
+
except Exception:
|
| 22 |
+
Image = ImageDraw = ImageFont = None
|
| 23 |
+
|
| 24 |
|
|
|
|
| 25 |
def _hf_token():
|
| 26 |
for k in ("HF_TOKEN", "HUGGINGFACEHUB_API_TOKEN", "HUGGING_FACE_HUB_TOKEN", "HF_API_TOKEN"):
|
| 27 |
v = os.getenv(k, "").strip()
|
|
|
|
| 31 |
|
| 32 |
HF_TOKEN = _hf_token()
|
| 33 |
QWEN_VL_MODEL = os.getenv("QWEN_VL_MODEL", "Qwen/Qwen2.5-VL-7B-Instruct")
|
| 34 |
+
DATA_DIR = "/data" if os.path.isdir("/data") else "/app/data"
|
| 35 |
+
WALL_FILE = os.path.join(DATA_DIR, "ai_wall_posts.json")
|
| 36 |
+
SHORTS_DIR = os.path.join(DATA_DIR, "ai_shorts")
|
| 37 |
HEADERS = {"User-Agent":"Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36","Accept-Language":"vi-VN,vi;q=0.9,en;q=0.8"}
|
| 38 |
LAST_QWEN_ERROR = ""
|
| 39 |
|
| 40 |
+
|
| 41 |
def _load_ai_wall():
|
| 42 |
try:
|
| 43 |
if os.path.exists(WALL_FILE):
|
|
|
|
| 47 |
pass
|
| 48 |
return []
|
| 49 |
|
| 50 |
+
|
| 51 |
def _save_ai_wall(posts):
|
| 52 |
try:
|
| 53 |
os.makedirs(os.path.dirname(WALL_FILE), exist_ok=True)
|
|
|
|
| 58 |
except Exception:
|
| 59 |
pass
|
| 60 |
|
| 61 |
+
|
| 62 |
def _clean_text(s: str) -> str:
|
| 63 |
s = html_lib.unescape(s or "")
|
| 64 |
return re.sub(r"\s+", " ", s).strip()
|
| 65 |
|
| 66 |
+
|
| 67 |
def _domain(u):
|
| 68 |
try:
|
| 69 |
return urlparse(u).netloc.replace("www.", "")
|
| 70 |
except Exception:
|
| 71 |
return ""
|
| 72 |
|
| 73 |
+
|
| 74 |
def _reader_url(target_url: str) -> str:
|
| 75 |
safe = quote(target_url, safe=":/?#[]@!$&'()*+,;=%")
|
| 76 |
return "https://r.jina.ai/http://" + safe
|
| 77 |
|
| 78 |
+
|
| 79 |
def jina_reader_markdown(url: str) -> str:
|
| 80 |
jr = _reader_url(url)
|
| 81 |
r = requests.get(jr, headers={"Accept":"text/markdown,text/plain,*/*", "X-Return-Format":"markdown", "User-Agent":"Mozilla/5.0"}, timeout=35)
|
| 82 |
r.raise_for_status()
|
| 83 |
return r.text or ""
|
| 84 |
|
| 85 |
+
|
| 86 |
def _parse_jina_markdown(md: str, url: str):
|
| 87 |
lines = [x.rstrip() for x in (md or "").splitlines()]
|
| 88 |
title = ""; image = ""; content_lines = []; in_content = False
|
|
|
|
| 109 |
if not title and paras: title = paras[0][:90]
|
| 110 |
return {"url": url, "title": title or url, "summary": paras[0] if paras else "", "text": "\n".join(paras), "image": image, "images": [image] if image else [], "via":"jina"}
|
| 111 |
|
| 112 |
+
|
| 113 |
def _best_content_block(soup):
|
| 114 |
best, best_score = None, 0
|
| 115 |
for el in soup.find_all(["article", "main", "section", "div"]):
|
|
|
|
| 121 |
if score > best_score: best, best_score = el, score
|
| 122 |
return best
|
| 123 |
|
| 124 |
+
|
| 125 |
def scrape_any_url_direct(url: str):
|
| 126 |
r = requests.get(url, headers=HEADERS, timeout=18)
|
| 127 |
if r.status_code in {401,403,406,409,429,451,503}: raise RuntimeError(f"blocked status {r.status_code}")
|
|
|
|
| 154 |
if len(data["text"]) < 160: raise RuntimeError("direct scrape too short")
|
| 155 |
return data
|
| 156 |
|
| 157 |
+
|
| 158 |
def scrape_any_url(url: str):
|
| 159 |
try:
|
| 160 |
return scrape_any_url_direct(url)
|
|
|
|
| 165 |
raise RuntimeError(f"Jina Reader also returned too little content; direct={direct_error}")
|
| 166 |
return data
|
| 167 |
|
| 168 |
+
|
| 169 |
def pollinations_image_url(topic: str, width=1024, height=576):
|
| 170 |
prompt = f"editorial illustration, topic: {topic}, modern clean cinematic news image, high quality, no text, no watermark"
|
| 171 |
return f"https://image.pollinations.ai/prompt/{quote_plus(prompt)}?width={width}&height={height}&seed={random.randint(1,999999)}&nologo=true&private=true"
|
| 172 |
|
| 173 |
+
|
| 174 |
def jina_search(query: str, limit=6):
|
| 175 |
for accept in ("application/json", "text/plain"):
|
| 176 |
try:
|
|
|
|
| 196 |
pass
|
| 197 |
return []
|
| 198 |
|
| 199 |
+
|
| 200 |
def duck_context(topic: str, limit=6):
|
| 201 |
try:
|
| 202 |
url = "https://duckduckgo.com/html/?q=" + quote_plus(topic + " tin tức phân tích bối cảnh")
|
|
|
|
| 210 |
return results
|
| 211 |
except Exception: return []
|
| 212 |
|
| 213 |
+
|
| 214 |
def web_context(topic: str, limit=6):
|
| 215 |
queries = [topic + " tin tức phân tích bối cảnh", topic]
|
| 216 |
results=[]
|
|
|
|
| 230 |
context = "\n".join([f"- {x.get('title') or _domain(x.get('url',''))} ({_domain(x.get('url',''))}): {x.get('excerpt') or x.get('description','')}" for x in enriched if x.get("excerpt") or x.get("title")])
|
| 231 |
return context, enriched
|
| 232 |
|
| 233 |
+
|
| 234 |
async def qwen_generate(prompt: str, image_url: Optional[str] = None, max_tokens: int = 1200):
|
| 235 |
global LAST_QWEN_ERROR, HF_TOKEN
|
| 236 |
HF_TOKEN = _hf_token()
|
|
|
|
| 240 |
if not AsyncInferenceClient:
|
| 241 |
LAST_QWEN_ERROR = "Thiếu thư viện huggingface_hub hoặc import lỗi."
|
| 242 |
return None
|
| 243 |
+
errors=[]; models=[]
|
|
|
|
| 244 |
for m in [QWEN_VL_MODEL, "Qwen/Qwen2.5-VL-7B-Instruct", "Qwen/Qwen2.5-VL-3B-Instruct"]:
|
| 245 |
if m and m not in models: models.append(m)
|
| 246 |
for model in models:
|
|
|
|
| 253 |
resp=await client.chat_completion(model=model, messages=messages, max_tokens=max_tokens, temperature=0.45, top_p=0.85)
|
| 254 |
txt=(resp.choices[0].message.content or "").strip()
|
| 255 |
if txt:
|
| 256 |
+
LAST_QWEN_ERROR=""; return txt
|
|
|
|
| 257 |
except Exception as e:
|
| 258 |
errors.append(f"{model}: {type(e).__name__}: {str(e)[:220]}")
|
| 259 |
LAST_QWEN_ERROR = " | ".join(errors) or "Qwen không trả nội dung."
|
| 260 |
print("[qwen errors]", LAST_QWEN_ERROR)
|
| 261 |
return None
|
| 262 |
|
| 263 |
+
|
| 264 |
def make_post(title, text, image, source_url, kind, sources=None):
|
| 265 |
+
return {"id":str(int(time.time()*1000))+str(random.randint(100,999)),"title":title,"text":text,"img":image,"url":source_url,"kind":kind,"sources":sources or [],"video":"","ts":int(time.time())}
|
| 266 |
+
|
| 267 |
|
| 268 |
@app.get("/api/ai_wall")
|
| 269 |
def api_ai_wall(): return JSONResponse({"posts":_load_ai_wall()[:80]})
|
| 270 |
|
| 271 |
+
|
| 272 |
@app.get("/api/ai/status")
|
| 273 |
def api_ai_status():
|
| 274 |
+
return JSONResponse({"has_token":bool(_hf_token()),"client_imported":AsyncInferenceClient is not None,"model":QWEN_VL_MODEL,"last_error":LAST_QWEN_ERROR,"tts_ready":gTTS is not None})
|
| 275 |
+
|
| 276 |
|
| 277 |
@app.post("/api/ai/topic")
|
| 278 |
async def api_ai_topic(request:Request):
|
| 279 |
+
body=await request.json(); topic=_clean_text(body.get("topic",""))
|
| 280 |
if not topic: return JSONResponse({"error":"missing topic"},status_code=400)
|
| 281 |
ctx,sources=web_context(topic)
|
| 282 |
if not ctx: return JSONResponse({"error":"Không lấy được dữ liệu internet cho chủ đề này. Hãy thử chủ đề cụ thể hơn."},status_code=422)
|
| 283 |
image=pollinations_image_url(topic)
|
| 284 |
+
prompt=f"""Tóm tắt chủ đề để đăng Tường AI.
|
| 285 |
|
| 286 |
Chủ đề: {topic}
|
| 287 |
|
| 288 |
Yêu cầu:
|
| 289 |
- Chỉ tóm tắt các ý chính dựa trên nguồn internet bên dưới.
|
| 290 |
+
- Không viết lại thành bài dài, không lan man, không lặp lại.
|
| 291 |
- 1 tiêu đề ngắn.
|
| 292 |
- 3-5 gạch đầu dòng, mỗi dòng tối đa 1 câu.
|
| 293 |
- 1 câu kết luận ngắn.
|
|
|
|
| 297 |
{ctx}"""
|
| 298 |
text=await qwen_generate(prompt,image_url=image,max_tokens=900)
|
| 299 |
if not text: return JSONResponse({"error":"Qwen2.5-VL chưa sẵn sàng: "+LAST_QWEN_ERROR},status_code=503)
|
| 300 |
+
post=make_post(topic,text,image,"","topic",sources=sources[:5]); posts=_load_ai_wall(); posts.insert(0,post); _save_ai_wall(posts)
|
| 301 |
return JSONResponse({"post":post})
|
| 302 |
|
| 303 |
+
|
| 304 |
@app.post("/api/ai/url")
|
| 305 |
async def api_ai_url(request:Request):
|
| 306 |
+
body=await request.json(); url=_clean_text(body.get("url",""))
|
| 307 |
if not url.startswith("http"): return JSONResponse({"error":"missing url"},status_code=400)
|
| 308 |
try: data=scrape_any_url(url)
|
| 309 |
except Exception as e: return JSONResponse({"error":"Không scrape được URL: "+str(e)[:180]},status_code=422)
|
|
|
|
| 326 |
text=await qwen_generate(prompt,image_url=data.get("image") or None,max_tokens=900)
|
| 327 |
if not text: return JSONResponse({"error":"Qwen2.5-VL chưa sẵn sàng: "+LAST_QWEN_ERROR},status_code=503)
|
| 328 |
post=make_post(data.get("title") or "Bài viết",text,data.get("image") or "",url,"url",sources=[{"title":data.get("title"),"url":url,"excerpt":raw[:500],"via":data.get("via")}])
|
| 329 |
+
posts=_load_ai_wall(); posts.insert(0,post); _save_ai_wall(posts)
|
| 330 |
return JSONResponse({"post":post})
|
| 331 |
|
| 332 |
+
|
| 333 |
@app.post("/api/rewrite_share")
|
| 334 |
async def api_rewrite_share(request:Request):
|
| 335 |
+
body=await request.json(); url=_clean_text(body.get("url",""))
|
| 336 |
if not url.startswith("http"): return JSONResponse({"error":"missing url"},status_code=400)
|
| 337 |
try: data=scrape_any_url(url)
|
| 338 |
except Exception as e: return JSONResponse({"error":"Không đọc được bài viết: "+str(e)[:180]},status_code=422)
|
|
|
|
| 351 |
text=await qwen_generate(prompt,image_url=data.get("image") or None,max_tokens=900)
|
| 352 |
if not text: return JSONResponse({"error":"Qwen2.5-VL chưa sẵn sàng: "+LAST_QWEN_ERROR},status_code=503)
|
| 353 |
post=make_post(data.get("title") or "Bài viết",text,data.get("image") or "",url,"rewrite")
|
| 354 |
+
posts=_load_ai_wall(); posts.insert(0,post); _save_ai_wall(posts)
|
| 355 |
return JSONResponse({"post":post})
|
| 356 |
|
| 357 |
+
|
| 358 |
+
def _safe_name(s): return re.sub(r"[^a-zA-Z0-9_-]+", "_", s)[:80]
|
| 359 |
+
|
| 360 |
+
|
| 361 |
+
def _wrap_text(text, width=32, max_lines=10):
|
| 362 |
+
words=_clean_text(text).split(); lines=[]; cur=""
|
| 363 |
+
for w in words:
|
| 364 |
+
if len(cur)+len(w)+1<=width: cur=(cur+" "+w).strip()
|
| 365 |
+
else:
|
| 366 |
+
if cur: lines.append(cur)
|
| 367 |
+
cur=w
|
| 368 |
+
if len(lines)>=max_lines: break
|
| 369 |
+
if cur and len(lines)<max_lines: lines.append(cur)
|
| 370 |
+
return "\n".join(lines)
|
| 371 |
+
|
| 372 |
+
|
| 373 |
+
def _make_short_frame(post, img_path, out_path):
|
| 374 |
+
if Image is None: raise RuntimeError("Pillow chưa sẵn sàng")
|
| 375 |
+
W,H=1080,1920
|
| 376 |
+
bg=Image.new("RGB",(W,H),(14,14,14))
|
| 377 |
+
try:
|
| 378 |
+
im=Image.open(img_path).convert("RGB")
|
| 379 |
+
# cover top area 1080x860
|
| 380 |
+
target=(1080,860)
|
| 381 |
+
im_ratio=im.width/im.height; target_ratio=target[0]/target[1]
|
| 382 |
+
if im_ratio>target_ratio:
|
| 383 |
+
new_h=target[1]; new_w=int(new_h*im_ratio)
|
| 384 |
+
else:
|
| 385 |
+
new_w=target[0]; new_h=int(new_w/im_ratio)
|
| 386 |
+
im=im.resize((new_w,new_h))
|
| 387 |
+
left=(new_w-target[0])//2; top=(new_h-target[1])//2
|
| 388 |
+
im=im.crop((left,top,left+target[0],top+target[1]))
|
| 389 |
+
bg.paste(im,(0,0))
|
| 390 |
+
except Exception:
|
| 391 |
+
pass
|
| 392 |
+
draw=ImageDraw.Draw(bg)
|
| 393 |
+
try:
|
| 394 |
+
font_title=ImageFont.truetype("/usr/share/fonts/truetype/dejavu/DejaVuSans-Bold.ttf",52)
|
| 395 |
+
font_body=ImageFont.truetype("/usr/share/fonts/truetype/dejavu/DejaVuSans.ttf",40)
|
| 396 |
+
font_label=ImageFont.truetype("/usr/share/fonts/truetype/dejavu/DejaVuSans-Bold.ttf",32)
|
| 397 |
+
except Exception:
|
| 398 |
+
font_title=font_body=font_label=None
|
| 399 |
+
draw.rectangle((0,780,W,H),fill=(14,14,14))
|
| 400 |
+
draw.text((54,830),"VNEWS · Tường AI",fill=(92,184,122),font=font_label)
|
| 401 |
+
title=_wrap_text(post.get("title",""),24,3)
|
| 402 |
+
draw.multiline_text((54,900),title,fill=(255,255,255),font=font_title,spacing=10)
|
| 403 |
+
body=_wrap_text(post.get("text",""),34,10)
|
| 404 |
+
draw.multiline_text((54,1120),body,fill=(220,220,220),font=font_body,spacing=12)
|
| 405 |
+
bg.save(out_path,quality=92)
|
| 406 |
+
|
| 407 |
+
|
| 408 |
+
def _download_image(url, fallback_topic, out_path):
|
| 409 |
+
if url:
|
| 410 |
+
try:
|
| 411 |
+
r=requests.get(url,headers=HEADERS,timeout=15)
|
| 412 |
+
if r.status_code==200 and len(r.content)>1000:
|
| 413 |
+
with open(out_path,"wb") as f:f.write(r.content)
|
| 414 |
+
return out_path
|
| 415 |
+
except Exception: pass
|
| 416 |
+
# fallback generated image
|
| 417 |
+
gen=pollinations_image_url(fallback_topic)
|
| 418 |
+
try:
|
| 419 |
+
r=requests.get(gen,headers=HEADERS,timeout=25)
|
| 420 |
+
if r.status_code==200 and len(r.content)>1000:
|
| 421 |
+
with open(out_path,"wb") as f:f.write(r.content)
|
| 422 |
+
return out_path
|
| 423 |
+
except Exception: pass
|
| 424 |
+
if Image:
|
| 425 |
+
Image.new("RGB",(1080,860),(30,55,42)).save(out_path)
|
| 426 |
+
return out_path
|
| 427 |
+
raise RuntimeError("Không tạo được ảnh")
|
| 428 |
+
|
| 429 |
+
|
| 430 |
+
def _short_script(post):
|
| 431 |
+
txt=_clean_text(post.get("text",""))
|
| 432 |
+
# TTS should read concise summary, not entire huge post.
|
| 433 |
+
if len(txt)>900: txt=txt[:900].rsplit(" ",1)[0]+"."
|
| 434 |
+
return f"{post.get('title','')}. {txt}"
|
| 435 |
+
|
| 436 |
+
|
| 437 |
+
@app.post("/api/ai/short/{post_id}")
|
| 438 |
+
def api_ai_short(post_id:str):
|
| 439 |
+
posts=_load_ai_wall(); post=next((p for p in posts if str(p.get("id"))==str(post_id)),None)
|
| 440 |
+
if not post: return JSONResponse({"error":"post not found"},status_code=404)
|
| 441 |
+
os.makedirs(SHORTS_DIR,exist_ok=True)
|
| 442 |
+
out_mp4=os.path.join(SHORTS_DIR,_safe_name(post_id)+".mp4")
|
| 443 |
+
if os.path.exists(out_mp4):
|
| 444 |
+
post["video"]="/api/ai/short-file/"+post_id
|
| 445 |
+
_save_ai_wall(posts)
|
| 446 |
+
return JSONResponse({"video":post["video"]})
|
| 447 |
+
if gTTS is None: return JSONResponse({"error":"gTTS chưa sẵn sàng"},status_code=503)
|
| 448 |
+
work=os.path.join(SHORTS_DIR,_safe_name(post_id))
|
| 449 |
+
os.makedirs(work,exist_ok=True)
|
| 450 |
+
img=os.path.join(work,"image.jpg"); frame=os.path.join(work,"frame.jpg"); audio=os.path.join(work,"voice.mp3")
|
| 451 |
+
try:
|
| 452 |
+
_download_image(post.get("img"), post.get("title","AI news"), img)
|
| 453 |
+
_make_short_frame(post,img,frame)
|
| 454 |
+
script=_short_script(post)
|
| 455 |
+
gTTS(script,lang="vi").save(audio)
|
| 456 |
+
cmd=["ffmpeg","-y","-loop","1","-i",frame,"-i",audio,"-shortest","-c:v","libx264","-tune","stillimage","-pix_fmt","yuv420p","-c:a","aac","-b:a","128k","-vf","scale=1080:1920",out_mp4]
|
| 457 |
+
subprocess.run(cmd,check=True,stdout=subprocess.PIPE,stderr=subprocess.PIPE,timeout=180)
|
| 458 |
+
post["video"]="/api/ai/short-file/"+post_id
|
| 459 |
+
_save_ai_wall(posts)
|
| 460 |
+
return JSONResponse({"video":post["video"]})
|
| 461 |
+
except Exception as e:
|
| 462 |
+
return JSONResponse({"error":"Không tạo được shorts: "+str(e)[:180]},status_code=500)
|
| 463 |
+
|
| 464 |
+
|
| 465 |
+
@app.get("/api/ai/short-file/{post_id}")
|
| 466 |
+
def api_ai_short_file(post_id:str):
|
| 467 |
+
path=os.path.join(SHORTS_DIR,_safe_name(post_id)+".mp4")
|
| 468 |
+
if not os.path.exists(path): return JSONResponse({"error":"not found"},status_code=404)
|
| 469 |
+
return FileResponse(path,media_type="video/mp4",filename=f"vnews-ai-{post_id}.mp4")
|
| 470 |
+
|
| 471 |
+
|
| 472 |
+
# Override index route to show Tường AI slide + shorts generation.
|
| 473 |
app.router.routes=[r for r in app.router.routes if not (getattr(r,'path',None)=='/' and 'GET' in getattr(r,'methods',set()))]
|
| 474 |
AI_INJECT=r'''
|
| 475 |
+
<style>.ai-wall-extra{margin:6px 4px;background:#1a1a1a;border:1px solid #2a2a2a;border-radius:8px;overflow:hidden}.ai-wall-card{flex:0 0 250px;background:#141414;border:1px solid #2b2b2b;border-radius:10px;padding:8px}.ai-wall-img{width:100%;aspect-ratio:16/9;background:#222;border-radius:8px;overflow:hidden;margin-bottom:6px}.ai-wall-img img{width:100%;height:100%;object-fit:cover}.ai-wall-title{font-size:12px;color:#5cb87a;font-weight:800;line-height:1.3;margin-bottom:4px}.ai-wall-text{font-size:11px;color:#bbb;line-height:1.45;white-space:pre-wrap;display:-webkit-box;-webkit-line-clamp:5;-webkit-box-orient:vertical;overflow:hidden}.ai-wall-actions{display:flex;gap:6px;margin-top:8px}.ai-wall-actions button{flex:1;border:1px solid #333;background:#222;color:#ddd;border-radius:14px;padding:6px 8px;font-size:10px}.ai-wall-actions button.primary{background:#2d8659;border-color:#2d8659;color:#fff}</style><script>(function(){function esc(s){return String(s||'').replace(/[&<>"']/g,m=>({'&':'&','<':'<','>':'>','"':'"',"'":'''}[m]));}let aiWall=[];async function refreshAiWall(){try{const r=await fetch('/api/ai_wall');const j=await r.json();aiWall=j.posts||[];renderAiWall();}catch(e){}}function renderAiWall(){const home=document.getElementById('view-home');if(!home)return;let old=document.getElementById('ai-wall-block');if(old)old.remove();if(!aiWall.length)return;let wrap=document.createElement('div');wrap.id='ai-wall-block';wrap.className='ai-wall-extra';let h='<div class="slider-header"><span class="slider-label">🧱 Tường AI</span></div><div class="slider-track">';aiWall.slice(0,30).forEach((p,i)=>{h+=`<div class="ai-wall-card"><div class="ai-wall-img">${p.img?`<img src="${p.img}">`:''}</div><div class="ai-wall-title">${esc(p.title)}</div><div class="ai-wall-text">${esc(p.text)}</div><div class="ai-wall-actions"><button onclick="aiReadWall(${i})">Xem</button><button class="primary" onclick="aiMakeShort(${i})">Shorts</button></div></div>`});h+='</div>';wrap.innerHTML=h;home.prepend(wrap)}window.aiReadWall=function(i){const p=aiWall[i];if(!p)return;showView('view-article');let sources='';if(p.sources&&p.sources.length){sources='<div class="article-summary"><b>Nguồn tham khảo:</b><br>'+p.sources.slice(0,5).map(s=>`• ${esc(s.title||s.url||'Nguồn')} ${s.url?`(${esc(new URL(s.url).hostname.replace('www.',''))})`:''}`).join('<br>')+'</div>'}let h=`<button class="back-btn" onclick="switchCat('home')">← Quay lại</button><div class="article-view"><span class="badge badge-ai">AI</span><h1 class="article-title">${esc(p.title)}</h1>${p.img?`<img class="article-img" src="${p.img}">`:''}${sources}<p class="article-p" style="white-space:pre-wrap">${esc(p.text)}</p>${p.video?`<video class="article-img" src="${p.video}" controls playsinline></video>`:''}<div class="article-actions">${p.url?`<button onclick="window.open('${p.url}','_blank')">🔗 Nguồn</button>`:''}<button onclick="aiMakeShort(${i})">🎬 Tạo video shorts</button></div></div>`;document.getElementById('view-article').innerHTML=h;window.scrollTo(0,0)};window.aiMakeShort=async function(i){const p=aiWall[i];if(!p)return;try{alert('Đang tạo shorts AI, vui lòng chờ...');const r=await fetch('/api/ai/short/'+p.id,{method:'POST'});const j=await r.json();if(!r.ok||j.error)throw new Error(j.error||'Lỗi');p.video=j.video;aiReadWall(i);refreshAiWall();}catch(e){alert(e.message)}};const oldLoad=window.loadHome;if(typeof oldLoad==='function'){window.loadHome=async function(){await oldLoad.apply(this,arguments);setTimeout(refreshAiWall,50);};}setTimeout(refreshAiWall,1200);})();</script>'''
|
| 476 |
@app.get('/')
|
| 477 |
async def index_ai():
|
| 478 |
with open('/app/static/index.html','r',encoding='utf-8') as f: html=f.read()
|