WeChat Message Content Format Reference
Message Type Codes (Complete)
| Type | Meaning | Content Field | imgPath Field |
|---|---|---|---|
| 1 | Text | Plain text, XML, or special format | NULL |
| 3 | Image | XML with <img> + CDN metadata |
THUMBNAIL_DIRPATH://th_<hash> |
| 34 | Voice | XML with <voicemsg> + CDN metadata |
Voice filename hash (no extension) |
| 43 | Video | XML with <videomsg> |
Video hash |
| 47 | Emoji | XML with <emoji> |
NULL |
| 49 | Article/Link/Mini-program | XML with <appmsg> containing <url>, <title> |
NULL |
| 10000 | System | Plain text | NULL |
| 754974769 | Unsupported (newer client message) | "当前微信版本不支持展示该内容" | NULL |
| 822083633 | Referenced appmsg (reply with quote) | XML with <appmsg> containing <refermsg> + <title> |
NULL |
Type 822083633: This is a "reply with reference" message. The XML structure contains
<refermsg>pointing to the original message being quoted, embedded inside an<appmsg>. Extract<title>from the outer appmsg for display. Example:\n>xml\n> <msg><appmsg><title>原来是你</title>...<refermsg><type>3</type><displayname>发条猴子</displayname>...\n>\n> Display as:[引用发条猴子的消息] 原来是你or just the title.\n\n> Type 754974769 (Finder/视频号): Although title says \"当前微信版本不支持展示该内容\", the XML contains a full<finderFeed>block with usable metadata. CRITICAL: All text fields inside finderFeed are wrapped in<![CDATA[...]]>— you MUST strip CDATA before regex extraction:\n>python\n> cdata_clean = re.sub(r\"<!\\[CDATA\\[(.*?)\\]\\]>\", r\"\\1\", content)\n> nick = re.search(r\"<nickname>(.*?)</nickname>\", cdata_clean)\n>\n> Extractable fields:<nickname>(账号名),<desc>(视频描述),<avatar>(CDN头像),<coverUrl>(封面图, HTTP可直接访问),<media><url>(视频下载链接),<videoPlayDuration>(秒),<width>/<height>(分辨率).
XML Message Parsing Patterns
Pattern 1: Clean text (type=1)
display_text = content.split("\n", 1)[-1] # strip sender prefix
if "<msg>" not in display_text:
return display_text.strip() # plain text, done
Pattern 2: appmsg with title (types 49, 822083633, and others)
title = re.search(r"<title>(.*?)</title>", content)
url = re.search(r"<url>(.*?)</url>", content)
if title:
if url:
return {"articles": [{"url": url, "title": title}]}
return title # text only
Pattern 3: refermsg (quoted reply)
if "<refermsg>" in content:
ref_from = re.search(r"<displayname>(.*?)</displayname>", content)
title = re.search(r"<title>(.*?)</title>", content)
if title:
return f"[引用{ref_from}的消息] {title}"
return "[引用消息]"
Pattern 4: Garbage suppression
After stripping XML tags (re.sub(r"<[^>]+>", "")), if the remaining text is >200 chars, it's likely metadata garbage — truncate to 150 chars + "...".
Pattern 5: finderFeed with CDATA (type=754974769)
# Step 1: strip CDATA wrappers (ESSENTIAL — regex won't match otherwise)
cdata_clean = re.sub(r"<!\[CDATA\[(.*?)\]\]>", r"\1", content)
# Step 2: extract fields
nick = re.search(r"<nickname>(.*?)</nickname>", cdata_clean)
desc = re.search(r"<desc>(.*?)</desc>", cdata_clean)
cover = re.search(r"<coverUrl>(.*?)</coverUrl>", cdata_clean)
avatar = re.search(r"<avatar>(.*?)</avatar>", cdata_clean)
video_url = re.search(r"<media>.*?<url>(.*?)</url>", cdata_clean, re.DOTALL)
duration = int(re.search(r"<videoPlayDuration>(\d+)</videoPlayDuration>", cdata_clean).group(1))
# The cover URL is an HTTP CDN URL — directly accessible from browser
# e.g. http://wxapp.tc.qq.com/251/20304/stodownload?encfilekey=...
Chatroom Message Content Format
Chatroom messages have a sender:\n prefix that MUST be stripped before display:
Raw content: "yibooo:\nok 我来也整一个 快速看齐"
Display text: "ok 我来也整一个 快速看齐"
Raw content: "hw090909:\n<?xml version=\"1.0\"?>..."
Display text: stripped XML content
Parsing rule: split on first \n. If the first segment contains :, it's the sender prefix — drop it.
Image Path Resolution
The imgPath field for type=3 messages stores a thumbnail path:
Raw imgPath: THUMBNAIL_DIRPATH://th_505c5070300c2473cdd5c672ee08d581
Stripped: th_505c5070300c2473cdd5c672ee08d581
Hash: 505c5070300c2473cdd5c672ee08d581
Actual path: <WECHAT_DATA_ROOT>/image2/50/5c/th_505c5070300c2473cdd5c672ee08d581
Path construction rule:
- Strip
THUMBNAIL_DIRPATH://orTHUMBNAIL_DIR_PATH://prefix - Remove
th_prefix to get the hash - Subdirectory:
{hash[0:2]}/{hash[2:4]} - Try both
th_{hash}and{hash}(original may not exist locally) - Full path:
<WECHAT_DATA_ROOT>/image2/{sub1}/{sub2}/{filename}
Permissions: The WeChat data directory (/data/data/com.tencent.mm/MicroMsg/) is owned by the WeChat UID (u0_a336), NOT Termux. Direct file access from Flask/Python running as Termux user will raise PermissionError: [Errno 13]. Use su -c for:
- Checking file existence:
su -c "test -f <path> && echo yes" - Copying files to a readable location:
su -c "cp <src> <dst> && chmod 644 <dst>" - Finding files:
su -c "find <dir> -name '*<pattern>*'"
Serving pattern for Flask: Copy the file to /tmp/ via su, then serve from /tmp/.
Voice Path Resolution
Voice files have a complex directory structure NOT based on simple hash prefix:
imgPath from DB: 05142107062628ef8e19f29104
Actual file: voice2/66/33/msg_05142107062628ef8e19f29104.amr
^^^^^^ — subdirs 66/33 do NOT match hash prefix 05/14
Resolution method: Use find via su because the subdirectory pattern is non-trivial:
su -c "find <WECHAT_DATA_ROOT>/voice2 -name '*<imgPath>*' 2>/dev/null | head -1"
Files are named msg_{hash}.amr, stored in voice2/{XX}/{YY}/ subdirectories.
Playback in browser: AMR is not browser-supported. Convert to MP3 via ffmpeg:
# Must copy to temp first (permission issue)
su -c "cp <voice_path> /tmp/voice_src && chmod 644 /tmp/voice_src"
ffmpeg -y -i /tmp/voice_src -acodec libmp3lame -ar 22050 -ac 1 /tmp/voice.mp3
SQLCipher CLI Quirks
When using sqlcipher CLI from Python subprocess:
"ok\n" prefix: sqlcipher prints
ok\nbefore JSON output in some versions. Strip it:if output.startswith("ok"): output = output[2:].strip()Multi-line JSON arrays:
.mode jsonproduces pretty-printed arrays with line breaks inside the array.json.loads()handles this fine AFTER stripping "ok\n".Required pragmas for WeChat EnMicroMsg.db:
PRAGMA key=<key>; PRAGMA cipher_compatibility=1; PRAGMA cipher_page_size=1024; PRAGMA query_only=true; # Recommended for read-only accessCommand construction: Must use
su -cwrapper since DB is root-owned:cmd = ["su", "-c", " ".join([ "/path/to/sqlcipher", db_path, '-cmd "PRAGMA key=<key>;"', '-cmd "PRAGMA cipher_compatibility=1;"', ... ])]
Group Name Resolution
For @chatroom talkers, the custom group name lives in rcontact.nickname, NOT in chatroom.displayname:
rcontact.nickname → "宜创AI全员群" (custom name if set, else empty)
chatroom.displayname → "火星人、丁鹏、小伙子..." (member nickname list)
SQL pattern for session list:
CASE
WHEN m.talker LIKE '%@chatroom' THEN
COALESCE(NULLIF(r.nickname, ''), c.displayname, m.talker)
ELSE
COALESCE(NULLIF(r.conRemark, ''), r.nickname, m.talker)
END AS display_name
For personal contacts, prefer rcontact.conRemark (remark name) over rcontact.nickname.