chore: import upstream snapshot with attribution

servers/fastapi/utils/template_vision_errors.py

@@ -0,0 +1,106 @@
+"""Heuristics for LLM errors when image + text (vision) is required but the model rejects it."""
+
+from __future__ import annotations
+
+import re
+
+# Shown in API/job error text; frontend uses this for a clearer toast title.
+VISION_LAYOUT_ERROR_MARKER = "TEMPLATE_VISION_MODEL_REQUIRED"
+
+VISION_LAYOUT_USER_MESSAGE = (
+    f"{VISION_LAYOUT_ERROR_MARKER}\n\n"
+    "Generating a custom template layout sends a slide screenshot to your text model, "
+    "so the model must support vision (image + text).\n\n"
+    "The current text model does not accept image inputs. Open Settings, pick a "
+    "vision-capable model for your provider (for example GPT-4o, Claude 3.5 Sonnet or "
+    "newer, or Gemini with image support), save, and try again."
+)
+
+
+def _collect_exception_text(exc: BaseException, *, max_chain: int = 10) -> str:
+    parts: list[str] = []
+    seen: set[int] = set()
+    cur: BaseException | None = exc
+    depth = 0
+    while cur is not None and depth < max_chain and id(cur) not in seen:
+        seen.add(id(cur))
+        parts.append(str(cur))
+        msg = getattr(cur, "message", None)
+        if isinstance(msg, str) and msg.strip():
+            parts.append(msg)
+        body = getattr(cur, "body", None)
+        if body is not None:
+            parts.append(repr(body))
+        for attr in ("response", "error", "detail"):
+            obj = getattr(cur, attr, None)
+            if isinstance(obj, str) and obj.strip():
+                parts.append(obj)
+            elif isinstance(obj, dict):
+                parts.append(repr(obj))
+        cur = cur.__cause__ or cur.__context__
+        depth += 1
+    return " ".join(parts).lower()
+
+
+def is_likely_vision_capability_error(exc: BaseException) -> bool:
+    """
+    Best-effort detection across OpenAI-compatible, Anthropic, Gemini, and LiteLLM errors
+    when the model rejects multimodal / image content.
+    """
+    blob = _collect_exception_text(exc)
+    if not blob.strip():
+        return False
+
+    strong = (
+        "image input",
+        "image inputs",
+        "image_url",
+        "input_image",
+        "inline_data",
+        "multimodal",
+        "multi-modal",
+        "vision is not",
+        "does not support images",
+        "does not support image",
+        "cannot accept image",
+        "image content is not supported",
+        "unsupported content type",
+        "model is not multimodal",
+        "text-only model",
+        "text only model",
+        "this model only supports text",
+        "only supports text",
+        "invalid image",
+        "badimage",
+        "no image support",
+        "images are not supported",
+        "image parts",
+        "content blocks of type image",
+        "type 'image'",
+        "modality image",
+    )
+    if any(s in blob for s in strong):
+        return True
+
+    if "image" in blob or "picture" in blob or "screenshot" in blob:
+        weak = (
+            "not supported",
+            "unsupported",
+            "not allowed",
+            "invalid",
+            "cannot",
+            "does not support",
+            "not available",
+            "not enabled",
+            "not accept",
+            "rejected",
+            "forbidden",
+            "bad request",
+        )
+        if any(w in blob for w in weak):
+            return True
+
+    if re.search(r"\bimage_url\b", blob) or re.search(r"\binput_image\b", blob):
+        return True
+
+    return False