improvement(media): 提升图片/表情资源解密稳定性并自愈缓存

- 增加图片有效性校验并优化 JPEG/WebP 识别，降低 XOR 解密误判
- 图片资源按变体(b/h/c/t)与大小排序尝试多个候选，提高命中率
- 发现损坏缓存文件时自动删除并重建（单张获取与批量解密均生效）
- 表情本地解密失败时支持从 emoticon.db 安全回退下载，并支持 AES-CBC 解密

src/wechat_decrypt_tool/media_helpers.py

@@ -2,6 +2,7 @@ import ctypes
 import datetime
 import glob
 import hashlib
+import ipaddress
 import json
 import mimetypes
 import os
@@ -11,6 +12,7 @@ import struct
 from functools import lru_cache
 from pathlib import Path
 from typing import Any, Optional
+from urllib.parse import urlparse
 
 from fastapi import HTTPException
 
@@ -74,15 +76,391 @@ def _detect_image_media_type(data: bytes) -> str:
 
     if data.startswith(b"\x89PNG\r\n\x1a\n"):
         return "image/png"
-    if data.startswith(b"\xff\xd8\xff"):
+    if data.startswith(b"\xff\xd8\xff") and len(data) >= 4:
+        marker = data[3]
+        # Most JPEG marker types are in 0xC0..0xFE (APP, SOF, DQT, DHT, SOS, COM, etc.).
+        # This avoids false positives where random bytes start with 0xFFD8FF.
+        if marker not in (0x00, 0xFF) and marker >= 0xC0:
             return "image/jpeg"
     if data.startswith(b"GIF87a") or data.startswith(b"GIF89a"):
         return "image/gif"
-    if data.startswith(b"RIFF") and data[8:12] == b"WEBP":
+    if data.startswith(b"RIFF") and len(data) >= 12 and data[8:12] == b"WEBP":
         return "image/webp"
     return "application/octet-stream"
 
 
+def _is_probably_valid_image(data: bytes, media_type: str) -> bool:
+    """Heuristic validation to reduce false positives when guessing XOR keys.
+
+    We keep it lightweight (no full parsing), only checking common trailers.
+    """
+    if not data:
+        return False
+
+    mt = str(media_type or "").strip().lower()
+    if not mt.startswith("image/"):
+        return False
+
+    if mt == "image/jpeg":
+        if _detect_image_media_type(data[:32]) != "image/jpeg":
+            return False
+        trimmed = data.rstrip(b"\x00")
+        if len(trimmed) < 4 or not trimmed.startswith(b"\xff\xd8\xff"):
+            return False
+        if trimmed.endswith(b"\xff\xd9"):
+            return True
+        tail = trimmed[-4096:] if len(trimmed) > 4096 else trimmed
+        i = tail.rfind(b"\xff\xd9")
+        return i >= 0 and i >= len(tail) - 64 - 2
+
+    if mt == "image/png":
+        if not data.startswith(b"\x89PNG\r\n\x1a\n"):
+            return False
+        trailer = b"\x00\x00\x00\x00IEND\xaeB`\x82"
+        trimmed = data.rstrip(b"\x00")
+        if trimmed.endswith(trailer):
+            return True
+        tail = trimmed[-256:] if len(trimmed) > 256 else trimmed
+        i = tail.rfind(trailer)
+        return i >= 0 and i >= len(tail) - 64 - len(trailer)
+
+    if mt == "image/gif":
+        if not (data.startswith(b"GIF87a") or data.startswith(b"GIF89a")):
+            return False
+        trimmed = data.rstrip(b"\x00")
+        if trimmed.endswith(b"\x3B"):
+            return True
+        tail = trimmed[-256:] if len(trimmed) > 256 else trimmed
+        i = tail.rfind(b"\x3B")
+        return i >= 0 and i >= len(tail) - 16 - 1
+
+    if mt == "image/webp":
+        if len(data) < 12:
+            return False
+        return bool(data.startswith(b"RIFF") and data[8:12] == b"WEBP")
+
+    # Unknown image types: fall back to header-only check.
+    return _detect_image_media_type(data[:32]) != "application/octet-stream"
+
+
+def _normalize_variant_basename(name: str) -> str:
+    """Normalize a media filename stem by stripping common variant suffixes.
+
+    Mirrors echotrace's idea of normalizing `.t/.h/.b/.c` and `_t/_h/_b/_c`.
+    """
+    v = str(name or "").strip()
+    if not v:
+        return ""
+    lower = v.lower()
+    for suf in ("_b", "_h", "_c", "_t", ".b", ".h", ".c", ".t"):
+        if lower.endswith(suf) and len(lower) > len(suf):
+            return lower[: -len(suf)]
+    return lower
+
+
+def _variant_rank(name: str) -> int:
+    """Ordering used when trying multiple candidate resources.
+
+    Prefer: big > high > original > cache > thumb.
+    """
+    n = str(name or "").lower()
+    if n.endswith(("_b", ".b")):
+        return 0
+    if n.endswith(("_h", ".h")):
+        return 1
+    if n.endswith(("_c", ".c")):
+        return 3
+    if n.endswith(("_t", ".t")):
+        return 4
+    return 2
+
+
+def _iter_media_source_candidates(source: Path, *, limit: int = 30) -> list[Path]:
+    """Yield sibling variant files around a resolved source path.
+
+    This is a lightweight approximation of echotrace's \"search many .dat variants then try them\".
+    """
+    if not source:
+        return []
+
+    try:
+        if not source.exists():
+            return []
+    except Exception:
+        return []
+
+    try:
+        if source.is_dir():
+            return []
+    except Exception:
+        return []
+
+    out: list[Path] = []
+    try:
+        out.append(source.resolve())
+    except Exception:
+        out.append(source)
+
+    parent = source.parent
+    stem = str(source.stem or "")
+    base = _normalize_variant_basename(stem)
+    if not base:
+        return out
+
+    preferred_names = [
+        f"{base}_b.dat",
+        f"{base}_h.dat",
+        f"{base}.dat",
+        f"{base}_c.dat",
+        f"{base}_t.dat",
+        f"{base}.b.dat",
+        f"{base}.h.dat",
+        f"{base}.c.dat",
+        f"{base}.t.dat",
+        f"{base}.gif",
+        f"{base}.webp",
+        f"{base}.png",
+        f"{base}.jpg",
+        f"{base}.jpeg",
+    ]
+
+    for name in preferred_names:
+        p = parent / name
+        try:
+            if p.exists() and p.is_file():
+                out.append(p.resolve())
+        except Exception:
+            continue
+
+    # Add any other local .dat siblings with the same normalized base (limit to avoid explosion).
+    try:
+        for p in parent.glob(f"{base}*.dat"):
+            try:
+                if p.exists() and p.is_file():
+                    out.append(p.resolve())
+            except Exception:
+                continue
+            if len(out) >= int(limit):
+                break
+    except Exception:
+        pass
+
+    # De-dup while keeping order.
+    seen: set[str] = set()
+    uniq: list[Path] = []
+    for p in out:
+        try:
+            k = str(p.resolve())
+        except Exception:
+            k = str(p)
+        if k in seen:
+            continue
+        seen.add(k)
+        uniq.append(p)
+    return uniq
+
+
+def _order_media_candidates(paths: list[Path]) -> list[Path]:
+    """Sort candidate files similar to echotrace's variant preference + size heuristic."""
+    def _stat(p: Path) -> tuple[int, float]:
+        try:
+            st = p.stat()
+            return int(st.st_size), float(st.st_mtime)
+        except Exception:
+            return 0, 0.0
+
+    def key(p: Path) -> tuple[int, int, int, float, str]:
+        name = str(p.stem or "").lower()
+        rank = _variant_rank(name)
+        ext = str(p.suffix or "").lower()
+        # Prefer already-decoded formats (non-.dat) within the same variant rank.
+        ext_penalty = 1 if ext == ".dat" else 0
+        size, mtime = _stat(p)
+        return (rank, ext_penalty, -size, -mtime, str(p))
+
+    try:
+        return sorted(list(paths or []), key=key)
+    except Exception:
+        return list(paths or [])
+
+
+def _is_safe_http_url(url: str) -> bool:
+    u = str(url or "").strip()
+    if not u:
+        return False
+    try:
+        p = urlparse(u)
+    except Exception:
+        return False
+    if p.scheme not in ("http", "https"):
+        return False
+    host = (p.hostname or "").strip()
+    if not host:
+        return False
+    if host in {"localhost"}:
+        return False
+    try:
+        ip = ipaddress.ip_address(host)
+        if ip.is_private or ip.is_loopback or ip.is_link_local:
+            return False
+    except Exception:
+        pass
+    return True
+
+
+def _download_http_bytes(url: str, *, timeout: int = 20, max_bytes: int = 30 * 1024 * 1024) -> bytes:
+    if not _is_safe_http_url(url):
+        raise HTTPException(status_code=400, detail="Unsafe URL.")
+
+    try:
+        import requests
+    except Exception as e:
+        raise HTTPException(status_code=500, detail=f"requests not available: {e}")
+
+    try:
+        with requests.get(url, stream=True, timeout=timeout) as r:
+            r.raise_for_status()
+            try:
+                cl = int(r.headers.get("content-length") or 0)
+                if cl and cl > int(max_bytes):
+                    raise HTTPException(status_code=413, detail="Remote file too large.")
+            except HTTPException:
+                raise
+            except Exception:
+                pass
+
+            chunks: list[bytes] = []
+            total = 0
+            for chunk in r.iter_content(chunk_size=256 * 1024):
+                if not chunk:
+                    continue
+                chunks.append(chunk)
+                total += len(chunk)
+                if total > int(max_bytes):
+                    raise HTTPException(status_code=413, detail="Remote file too large.")
+            return b"".join(chunks)
+    except HTTPException:
+        raise
+    except Exception as e:
+        raise HTTPException(status_code=502, detail=f"Download failed: {e}")
+
+
+def _decrypt_emoticon_aes_cbc(data: bytes, aes_key_hex: str) -> Optional[bytes]:
+    """Decrypt WeChat emoticon payload from kNonStoreEmoticonTable.encrypt_url.
+
+    Observed scheme (WeChat 4.x):
+    - key = bytes.fromhex(aes_key_hex)  (16 bytes)
+    - iv  = key
+    - cipher = AES-128-CBC
+    - padding = PKCS7
+    """
+    if not data:
+        return None
+    if len(data) % 16 != 0:
+        return None
+
+    khex = str(aes_key_hex or "").strip().lower()
+    if not re.fullmatch(r"[0-9a-f]{32}", khex):
+        return None
+
+    try:
+        key = bytes.fromhex(khex)
+        if len(key) != 16:
+            return None
+    except Exception:
+        return None
+
+    try:
+        from Crypto.Cipher import AES
+        from Crypto.Util import Padding
+
+        pt_padded = AES.new(key, AES.MODE_CBC, iv=key).decrypt(data)
+        pt = Padding.unpad(pt_padded, AES.block_size)
+        return pt
+    except Exception:
+        return None
+
+
+@lru_cache(maxsize=2048)
+def _lookup_emoticon_info(account_dir_str: str, md5: str) -> dict[str, str]:
+    account_dir = Path(account_dir_str)
+    md5s = str(md5 or "").strip().lower()
+    if not md5s:
+        return {}
+
+    db_path = account_dir / "emoticon.db"
+    if not db_path.exists():
+        return {}
+
+    conn = sqlite3.connect(str(db_path))
+    conn.row_factory = sqlite3.Row
+    try:
+        row = conn.execute(
+            "SELECT md5, aes_key, cdn_url, encrypt_url, extern_url, thumb_url, tp_url "
+            "FROM kNonStoreEmoticonTable WHERE lower(md5) = lower(?) LIMIT 1",
+            (md5s,),
+        ).fetchone()
+        if not row:
+            return {}
+        return {k: str(row[k] or "") for k in row.keys()}
+    except Exception:
+        return {}
+    finally:
+        try:
+            conn.close()
+        except Exception:
+            pass
+
+
+def _try_fetch_emoticon_from_remote(account_dir: Path, md5: str) -> tuple[Optional[bytes], Optional[str]]:
+    info = _lookup_emoticon_info(str(account_dir), str(md5 or "").lower())
+    if not info:
+        return None, None
+
+    aes_key_hex = str(info.get("aes_key") or "").strip()
+    urls: list[str] = []
+    # Prefer plain CDN URL first; fall back to encrypt_url (needs AES-CBC decrypt).
+    for k in ("cdn_url", "extern_url", "thumb_url", "tp_url", "encrypt_url"):
+        u = str(info.get(k) or "").strip()
+        if u and _is_safe_http_url(u):
+            urls.append(u)
+
+    for url in urls:
+        try:
+            payload = _download_http_bytes(url)
+        except Exception:
+            continue
+
+        candidates: list[bytes] = [payload]
+        dec = _decrypt_emoticon_aes_cbc(payload, aes_key_hex)
+        if dec is not None:
+            candidates.insert(0, dec)
+
+        for data in candidates:
+            if not data:
+                continue
+            try:
+                data2, mt = _try_strip_media_prefix(data)
+            except Exception:
+                data2, mt = data, "application/octet-stream"
+
+            if mt == "application/octet-stream":
+                mt = _detect_image_media_type(data2[:32])
+            if mt == "application/octet-stream":
+                try:
+                    if len(data2) >= 8 and data2[4:8] == b"ftyp":
+                        mt = "video/mp4"
+                except Exception:
+                    pass
+
+            if mt.startswith("image/") and (not _is_probably_valid_image(data2, mt)):
+                continue
+            if mt != "application/octet-stream":
+                return data2, mt
+
+    return None, None
+
+
 class _WxAMConfig(ctypes.Structure):
     _fields_ = [
         ("mode", ctypes.c_int),
@@ -191,7 +569,7 @@ def _try_strip_media_prefix(data: bytes) -> tuple[bytes, str]:
         if j >= 0 and j <= 128 * 1024:
             sliced = data[j:]
             mt2 = _detect_image_media_type(sliced[:32])
-            if mt2 != "application/octet-stream":
+            if mt2 != "application/octet-stream" and _is_probably_valid_image(sliced, mt2):
                 return sliced, mt2
 
     try:
@@ -363,7 +741,7 @@ def _resolve_media_path_from_hardlink(
             quoted = _quote_ident(table_name)
             try:
                 row = conn.execute(
-                    f"SELECT dir1, dir2, file_name, modify_time FROM {quoted} WHERE md5 = ? ORDER BY modify_time DESC LIMIT 1",
+                    f"SELECT dir1, dir2, file_name, modify_time FROM {quoted} WHERE md5 = ? ORDER BY modify_time DESC, dir1 DESC, rowid DESC LIMIT 1",
                     (md5,),
                 ).fetchone()
             except Exception:
@@ -917,9 +1295,10 @@ def _try_xor_decrypt_by_magic(data: bytes) -> tuple[Optional[bytes], Optional[st
     # (offset, magic, media_type)
     candidates: list[tuple[int, bytes, str]] = [
         (0, b"\x89PNG\r\n\x1a\n", "image/png"),
-        (0, b"\xff\xd8\xff", "image/jpeg"),
         (0, b"GIF87a", "image/gif"),
         (0, b"GIF89a", "image/gif"),
+        (0, b"RIFF", "application/octet-stream"),
+        (4, b"ftyp", "video/mp4"),
         (0, b"wxgf", "application/octet-stream"),
         (1, b"wxgf", "application/octet-stream"),
         (2, b"wxgf", "application/octet-stream"),
@@ -936,8 +1315,8 @@ def _try_xor_decrypt_by_magic(data: bytes) -> tuple[Optional[bytes], Optional[st
         (13, b"wxgf", "application/octet-stream"),
         (14, b"wxgf", "application/octet-stream"),
         (15, b"wxgf", "application/octet-stream"),
-        (0, b"RIFF", "application/octet-stream"),
+        # JPEG magic is short (3 bytes), keep it last to reduce false positives.
-        (4, b"ftyp", "video/mp4"),
+        (0, b"\xff\xd8\xff", "image/jpeg"),
     ]
 
     for offset, magic, mt in candidates:
@@ -968,16 +1347,24 @@ def _try_xor_decrypt_by_magic(data: bytes) -> tuple[Optional[bytes], Optional[st
 
         if offset == 0 and magic == b"RIFF":
             if len(decoded) >= 12 and decoded[8:12] == b"WEBP":
+                if _is_probably_valid_image(decoded, "image/webp"):
                     return decoded, "image/webp"
             continue
 
-        if mt == "application/octet-stream":
+        if mt == "video/mp4":
-            mt2 = _detect_image_media_type(decoded[:32])
+            try:
-            if mt2 != "application/octet-stream":
+                if len(decoded) >= 8 and decoded[4:8] == b"ftyp":
-                return decoded, mt2
+                    return decoded, "video/mp4"
+            except Exception:
+                pass
             continue
 
-        return decoded, mt
+        mt2 = _detect_image_media_type(decoded[:32])
+        if mt2 != mt:
+            continue
+        if not _is_probably_valid_image(decoded, mt2):
+            continue
+        return decoded, mt2
 
     preview_len = 8192
     try:
@@ -1005,6 +1392,8 @@ def _try_xor_decrypt_by_magic(data: bytes) -> tuple[Optional[bytes], Optional[st
                     decoded = bytes(b ^ key for b in data)
                     dec2, mt2 = _try_strip_media_prefix(decoded)
                     if mt2 != "application/octet-stream":
+                        if mt2.startswith("image/") and (not _is_probably_valid_image(dec2, mt2)):
+                            continue
                         return dec2, mt2
             except Exception:
                 continue
@@ -1193,13 +1582,15 @@ def _detect_image_extension(data: bytes) -> str:
     """根据图片数据检测文件扩展名"""
     if not data:
         return "dat"
-    if data.startswith(b"\x89PNG\r\n\x1a\n"):
+    head = data[:32] if len(data) > 32 else data
+    mt = _detect_image_media_type(head)
+    if mt == "image/png":
         return "png"
-    if data.startswith(b"\xff\xd8\xff"):
+    if mt == "image/jpeg":
         return "jpg"
-    if data.startswith(b"GIF87a") or data.startswith(b"GIF89a"):
+    if mt == "image/gif":
         return "gif"
-    if data.startswith(b"RIFF") and len(data) >= 12 and data[8:12] == b"WEBP":
+    if mt == "image/webp":
         return "webp"
     return "dat"
 
@@ -1262,18 +1653,19 @@ def _read_and_maybe_decrypt_media(
 
     try:
         data_pref = path.read_bytes()
+        # Only accept prefix stripping when it looks like a real image/video,
+        # otherwise encrypted/random bytes may trigger false positives.
         stripped, mtp = _try_strip_media_prefix(data_pref)
         if mtp != "application/octet-stream":
+            if mtp.startswith("image/") and (not _is_probably_valid_image(stripped, mtp)):
+                pass
+            else:
                 return stripped, mtp
     except Exception:
         pass
 
     data = path.read_bytes()
 
-    dec, mt2 = _try_xor_decrypt_by_magic(data)
-    if dec is not None and mt2:
-        return dec, mt2
-
     # Try WeChat .dat v1/v2 decrypt.
     version = _detect_wechat_dat_version(data)
     if version in (0, 1, 2):
@@ -1356,8 +1748,23 @@ def _read_and_maybe_decrypt_media(
         except Exception:
             pass
 
+    # Fallback: try guessing XOR key by magic (only after key-based decrypt attempts).
+    # For V4 signature files, XOR guessing is not applicable and may be expensive.
+    if version in (0, -1):
+        dec, mt2 = _try_xor_decrypt_by_magic(data)
+        if dec is not None and mt2:
+            return dec, mt2
+
     # Fallback: return as-is.
     mt3 = _guess_media_type_by_path(path, fallback="application/octet-stream")
+    if mt3.startswith("image/") and (not _is_probably_valid_image(data, mt3)):
+        mt3 = "application/octet-stream"
+    if mt3 == "video/mp4":
+        try:
+            if not (len(data) >= 8 and data[4:8] == b"ftyp"):
+                mt3 = "application/octet-stream"
+        except Exception:
+            mt3 = "application/octet-stream"
     return data, mt3
 
 

src/wechat_decrypt_tool/routers/chat_media.py

@@ -18,6 +18,9 @@ from ..media_helpers import (
     _convert_silk_to_wav,
     _detect_image_extension,
     _detect_image_media_type,
+    _is_probably_valid_image,
+    _iter_media_source_candidates,
+    _order_media_candidates,
     _ensure_decrypted_resource_for_md5,
     _fallback_search_media_by_file_id,
     _fallback_search_media_by_md5,
@@ -31,6 +34,7 @@ from ..media_helpers import (
     _resolve_account_wxid_dir,
     _resolve_media_path_for_kind,
     _resolve_media_path_from_hardlink,
+    _try_fetch_emoticon_from_remote,
     _try_find_decrypted_resource,
     _try_strip_media_prefix,
 )
@@ -255,11 +259,14 @@ async def get_chat_image(
         if decrypted_path:
             data = decrypted_path.read_bytes()
             media_type = _detect_image_media_type(data[:32])
-            if media_type == "application/octet-stream":
+            if media_type != "application/octet-stream" and _is_probably_valid_image(data, media_type):
-                guessed = mimetypes.guess_type(str(decrypted_path))[0]
-                if guessed:
-                    media_type = guessed
                 return Response(content=data, media_type=media_type)
+            # Corrupted cached file (e.g. wrong ext / partial data): remove and regenerate from source.
+            try:
+                if decrypted_path.suffix.lower() in {".jpg", ".jpeg", ".png", ".gif", ".webp"}:
+                    decrypted_path.unlink()
+            except Exception:
+                pass
 
     # 回退：从微信数据目录实时定位并解密
     wxid_dir = _resolve_account_wxid_dir(account_dir)
@@ -283,6 +290,7 @@ async def get_chat_image(
         )
 
     p: Optional[Path] = None
+    candidates: list[Path] = []
 
     if md5:
         p = _resolve_media_path_from_hardlink(
@@ -297,12 +305,25 @@ async def get_chat_image(
             hit = _fallback_search_media_by_md5(str(wxid_dir), str(md5), kind="image")
             if hit:
                 p = Path(hit)
+        # Also add scan-based candidates to improve the chance of finding a usable variant.
+        if wxid_dir:
+            try:
+                hit2 = _fallback_search_media_by_md5(str(wxid_dir), str(md5), kind="image")
+                if hit2:
+                    candidates.extend(_iter_media_source_candidates(Path(hit2)))
+            except Exception:
+                pass
     elif file_id:
         # 一些版本图片消息无 MD5，仅提供 cdnthumburl 等“文件标识”
         for r in [wxid_dir, db_storage_dir]:
             if not r:
                 continue
-            hit = _fallback_search_media_by_file_id(str(r), str(file_id), kind="image", username=str(username or ""))
+            hit = _fallback_search_media_by_file_id(
+                str(r),
+                str(file_id),
+                kind="image",
+                username=str(username or ""),
+            )
             if hit:
                 p = Path(hit)
                 break
@@ -310,9 +331,29 @@ async def get_chat_image(
     if not p:
         raise HTTPException(status_code=404, detail="Image not found.")
 
-    logger.info(f"chat_image: md5={md5} file_id={file_id} resolved_source={p}")
+    candidates.extend(_iter_media_source_candidates(p))
+    candidates = _order_media_candidates(candidates)
 
-    data, media_type = _read_and_maybe_decrypt_media(p, account_dir=account_dir, weixin_root=wxid_dir)
+    logger.info(f"chat_image: md5={md5} file_id={file_id} candidates={len(candidates)} first={p}")
+
+    data = b""
+    media_type = "application/octet-stream"
+    chosen: Optional[Path] = None
+    for src_path in candidates:
+        try:
+            data, media_type = _read_and_maybe_decrypt_media(src_path, account_dir=account_dir, weixin_root=wxid_dir)
+        except Exception:
+            continue
+
+        if media_type.startswith("image/") and (not _is_probably_valid_image(data, media_type)):
+            continue
+
+        if media_type != "application/octet-stream":
+            chosen = src_path
+            break
+
+    if not chosen:
+        raise HTTPException(status_code=422, detail="Image found but failed to decode/decrypt.")
 
     # 仅在 md5 有效时缓存到 resource 目录；file_id 可能非常长，避免写入超长文件名
     if md5 and media_type.startswith("image/"):
@@ -326,7 +367,9 @@ async def get_chat_image(
         except Exception:
             pass
 
-    logger.info(f"chat_image: md5={md5} file_id={file_id} media_type={media_type} bytes={len(data)}")
+    logger.info(
+        f"chat_image: md5={md5} file_id={file_id} chosen={chosen} media_type={media_type} bytes={len(data)}"
+    )
     return Response(content=data, media_type=media_type)
 
 
@@ -341,18 +384,31 @@ async def get_chat_emoji(md5: str, account: Optional[str] = None, username: Opti
     if decrypted_path:
         data = decrypted_path.read_bytes()
         media_type = _detect_image_media_type(data[:32])
-        if media_type == "application/octet-stream":
+        if media_type != "application/octet-stream" and _is_probably_valid_image(data, media_type):
-            guessed = mimetypes.guess_type(str(decrypted_path))[0]
-            if guessed:
-                media_type = guessed
             return Response(content=data, media_type=media_type)
+        try:
+            if decrypted_path.suffix.lower() in {".jpg", ".jpeg", ".png", ".gif", ".webp"}:
+                decrypted_path.unlink()
+        except Exception:
+            pass
 
     wxid_dir = _resolve_account_wxid_dir(account_dir)
     p = _resolve_media_path_for_kind(account_dir, kind="emoji", md5=str(md5), username=username)
-    if not p:
+
+    data = b""
+    media_type = "application/octet-stream"
+    if p:
+        data, media_type = _read_and_maybe_decrypt_media(p, account_dir=account_dir, weixin_root=wxid_dir)
+
+    if media_type == "application/octet-stream":
+        # Some emojis are stored encrypted (see emoticon.db); try remote fetch as fallback.
+        data2, mt2 = _try_fetch_emoticon_from_remote(account_dir, str(md5).lower())
+        if data2 is not None and mt2:
+            data, media_type = data2, mt2
+
+    if (not p) and media_type == "application/octet-stream":
         raise HTTPException(status_code=404, detail="Emoji not found.")
 
-    data, media_type = _read_and_maybe_decrypt_media(p, account_dir=account_dir, weixin_root=wxid_dir)
     if media_type.startswith("image/"):
         try:
             out_md5 = str(md5).lower()

src/wechat_decrypt_tool/routers/media.py

@@ -11,6 +11,7 @@ from ..media_helpers import (
     _collect_all_dat_files,
     _decrypt_and_save_resource,
     _detect_image_media_type,
+    _is_probably_valid_image,
     _get_resource_dir,
     _load_media_keys,
     _resolve_account_dir,
@@ -315,6 +316,10 @@ async def decrypt_all_media_stream(
                 # 检查是否已解密
                 existing = _try_find_decrypted_resource(account_dir, md5)
                 if existing:
+                    try:
+                        cached = existing.read_bytes()
+                        cached_mt = _detect_image_media_type(cached[:32])
+                        if cached_mt != "application/octet-stream" and _is_probably_valid_image(cached, cached_mt):
                             skip_count += 1
                             # 每100个跳过的文件发送一次进度，减少消息量
                             if skip_count % 100 == 0 or current == total_files:
@@ -323,6 +328,14 @@ async def decrypt_all_media_stream(
                                 )
                                 await asyncio.sleep(0)
                             continue
+                        # Cache exists but looks corrupted: remove and regenerate.
+                        existing.unlink(missing_ok=True)
+                    except Exception:
+                        # If we can't read/validate it, try regenerating.
+                        try:
+                            existing.unlink(missing_ok=True)
+                        except Exception:
+                            pass
 
                 # 解密并保存
                 success, msg = _decrypt_and_save_resource(