diff --git a/frontend/app/main.py b/frontend/app/main.py
index 3c21f69..a5c380e 100644
--- a/frontend/app/main.py
+++ b/frontend/app/main.py
@@ -35,6 +35,60 @@ def _api(method: str, path: str, *, token: str | None = None, **kwargs):
     return httpx.request(method, _backend(path), headers=headers, timeout=30, **kwargs)
 
 
+def _model_matches_modality(model: dict, modality: str) -> bool:
+    """Heuristic fallback when backend modality filter returns empty."""
+    model_modality = (model.get("modality") or "").lower()
+    if model_modality == modality:
+        return True
+
+    text = f"{model.get('id', '')} {model.get('name', '')}".lower()
+    keywords = {
+        "image": ["image", "dall-e", "flux", "stable-diffusion", "sdxl", "recraft", "ideogram", "gpt-image"],
+        "video": ["video", "sora", "runway", "veo", "kling", "pika", "luma", "wan"],
+        "audio": ["audio", "speech", "voice", "tts", "transcribe", "whisper"],
+    }
+
+    if modality in keywords:
+        return any(k in text for k in keywords[modality])
+
+    if modality == "text":
+        non_text_hits = any(
+            k in text for k in keywords["image"] + keywords["video"] + keywords["audio"])
+        return not non_text_hits
+
+    return False
+
+
+def _load_models(token: str, modality: str) -> list[dict]:
+    """Load models for modality; fallback to unfiltered cache if needed."""
+    try:
+        models_resp = _api("GET", "/models/", token=token,
+                           params={"modality": modality})
+    except httpx.RequestError:
+        return []
+    if models_resp.status_code == 200:
+        try:
+            models = models_resp.json()
+        except ValueError:
+            models = []
+        if models:
+            return models
+
+    try:
+        all_resp = _api("GET", "/models/", token=token)
+    except httpx.RequestError:
+        return []
+    if all_resp.status_code != 200:
+        return []
+
+    try:
+        all_models = all_resp.json()
+    except ValueError:
+        return []
+    filtered = [m for m in all_models if _model_matches_modality(m, modality)]
+    return filtered or all_models
+
+
 def login_required(view):
     @functools.wraps(view)
     def wrapped(*args, **kwargs):
@@ -163,9 +217,7 @@ def generate_text():
             result = resp.json()
         else:
             error = resp.json().get("detail", "Generation failed.")
-    models_resp = _api("GET", "/models/", token=token,
-                       params={"modality": "text"})
-    models = models_resp.json() if models_resp.status_code == 200 else []
+    models = _load_models(token, "text")
     return render_template("generate_text.html", result=result, error=error, models=models)
 
 
@@ -186,9 +238,7 @@ def generate_image():
             )
             if up_resp.status_code not in (200, 201):
                 error = up_resp.json().get("detail", "Image upload failed.")
-                models_resp = _api("GET", "/models/",
-                                   token=token, params={"modality": "image"})
-                models = models_resp.json() if models_resp.status_code == 200 else []
+                models = _load_models(token, "image")
                 return render_template("generate_image.html", result=result, error=error, models=models)
 
         resp = _api("POST", "/generate/image", token=token, json={
@@ -203,9 +253,7 @@ def generate_image():
             result = resp.json()
         else:
             error = resp.json().get("detail", "Generation failed.")
-    models_resp = _api("GET", "/models/", token=token,
-                       params={"modality": "image"})
-    models = models_resp.json() if models_resp.status_code == 200 else []
+    models = _load_models(token, "image")
     return render_template("generate_image.html", result=result, error=error, models=models)
 
 
@@ -241,9 +289,7 @@ def generate_video():
             result = resp.json()
         else:
             error = resp.json().get("detail", "Generation failed.")
-    models_resp = _api("GET", "/models/", token=token,
-                       params={"modality": "video"})
-    models = models_resp.json() if models_resp.status_code == 200 else []
+    models = _load_models(token, "video")
     return render_template("generate_video.html", result=result, error=error, models=models)