fixes

lora-trigger-sheet

@@ -172,7 +172,56 @@ def _fetch_json(url, timeout=10):
         return json.loads(resp.read().decode())
 
 
-def lookup_source(filepath, filename, cache):
+# URL pattern for extracting embedded links from free-text fields
+_URL_RE = re.compile(r'https?://\S+', re.IGNORECASE)
+
+# Metadata fields that may contain a source URL, checked in order
+_META_URL_FIELDS = [
+    "modelspec.source",
+    "source",
+    "url",
+    "model_url",
+    "ss_training_comment",
+    "note",
+    "description",
+    "comment",
+]
+
+# Map hostname → display label
+_SOURCE_LABELS = {
+    "civitai.com": "CivitAI",
+    "huggingface.co": "HuggingFace",
+    "modelscope.cn": "ModelScope",
+    "tensor.art": "Tensor.Art",
+}
+
+
+def _label_for_url(url):
+    from urllib.parse import urlparse
+    host = urlparse(url).netloc.lstrip("www.")
+    for domain, label in _SOURCE_LABELS.items():
+        if host == domain or host.endswith("." + domain):
+            return label
+    return host or "Link"
+
+
+def _extract_meta_url(meta):
+    """Return (url, label) from embedded metadata fields, or (None, None)."""
+    if not meta:
+        return None, None
+    for field in _META_URL_FIELDS:
+        val = meta.get(field)
+        if not val or not isinstance(val, str):
+            continue
+        # Field may be a bare URL or free text containing one
+        m = _URL_RE.search(val.strip())
+        if m:
+            url = m.group(0).rstrip(".,;)")
+            return url, _label_for_url(url)
+    return None, None
+
+
+def lookup_source(filepath, filename, cache, meta=None):
     now = int(time.time())
     cached = cache.get(filename)
     if cached:
@@ -182,6 +231,12 @@ def lookup_source(filepath, filename, cache):
         if (now - last) < 86400 * 30:
             return (url, label) if url else (None, None)
 
+    # 0. Embedded URL in safetensors __metadata__ (free, no network call)
+    url, label = _extract_meta_url(meta)
+    if url:
+        cache[filename] = {"url": url, "label": label, "last_checked": now}
+        return url, label
+
     # CivitAI model-version lookup by BLAKE2b file hash
     try:
         h = file_blake2b_hash(filepath)
@@ -227,6 +282,30 @@ def lookup_source(filepath, filename, cache):
     except (HTTPError, URLError, json.JSONDecodeError, OSError):
         pass
 
+    # Fallback: ModelScope model search by filename
+    try:
+        q = (
+            filename.replace(".safetensors", "")
+            .replace("_", " ")
+            .replace("-", " ")
+        )
+        data = _fetch_json(
+            f"https://www.modelscope.cn/api/v1/models?Name={quote(q)}&PageSize=3&PageNumber=1"
+        )
+        models = (data or {}).get("Data", {}).get("Models") or []
+        if models:
+            m = models[0]
+            model_id = m.get("Path") or m.get("Name", "")
+            if model_id:
+                url = f"https://www.modelscope.cn/models/{model_id}"
+                cache[filename] = {
+                    "url": url, "label": "ModelScope",
+                    "model_name": model_id, "last_checked": now,
+                }
+                return url, "ModelScope"
+    except (HTTPError, URLError, json.JSONDecodeError, OSError):
+        pass
+
     cache[filename] = {"url": None, "label": None, "last_checked": now}
     return None, None
 
@@ -472,9 +551,9 @@ def main():
     cache = load_source_cache()
     sources = {}
     total = len(loras)
-    for i, (fname, _) in enumerate(loras, 1):
+    for i, (fname, meta) in enumerate(loras, 1):
         fpath = os.path.join(target, fname)
-        url, label = lookup_source(fpath, fname, cache)
+        url, label = lookup_source(fpath, fname, cache, meta)
         if url:
             sources[fname] = (url, label)
         if i % 10 == 0 or i == total: