From 63ca4237eb43d64911af9b4a0e3ab7501c25560f Mon Sep 17 00:00:00 2001
From: Classic298 <27028174+Classic298@users.noreply.github.com>
Date: Thu, 29 Oct 2026 09:24:47 +0200
Subject: [PATCH] feat: keep tool images away from models with Vision turned
 off (#33345)

When a tool returned an image, for example Open Terminal reading a PNG, a native tool or an MCP screenshot tool, the image was sent to the model even when the model has Vision turned off. A provider serving a text-only model rejects that request, so the reply failed with an error like "At most 0 image(s) may be provided in one prompt", and every later message in the chat failed the same way. With Vision off the model now gets a short "[image]" note in place of each tool image, including images already saved in the chat, so stuck chats work again. The images stay saved, so they reach the model again once Vision is turned on.

Fixes #32720
---
 backend/open_webui/utils/middleware.py | 26 +++++++++++++++++++++++---
 1 file changed, 23 insertions(+), 3 deletions(-)

diff --git a/backend/open_webui/utils/middleware.py b/backend/open_webui/utils/middleware.py
index 934a078f2c..7b27832b08 100644
--- a/backend/open_webui/utils/middleware.py
+++ b/backend/open_webui/utils/middleware.py
@@ -1939,6 +1939,22 @@ def strip_reasoning_details(output: list) -> list:
     ]
 
 
+def strip_tool_images(output: list, model: dict) -> list:
+    """Replace tool result images with a text note when the model has Vision turned off."""
+    if (model.get('info', {}).get('meta', {}).get('capabilities') or {}).get('vision', True):
+        return output
+    stripped_output = []
+    for item in output:
+        if isinstance(item, dict) and item.get('type') == 'function_call_output':
+            parts = [
+                {'type': 'input_text', 'text': '\n[image]'} if part.get('type') == 'input_image' else part
+                for part in item.get('output', [])
+            ]
+            item = {**item, 'output': parts}
+        stripped_output.append(item)
+    return stripped_output
+
+
 def process_messages_with_output(
     messages: list[dict],
     reasoning_format: str | None = None,
@@ -2192,6 +2208,8 @@ async def process_chat_payload(request, form_data, user, metadata, model):
         # for output produced by the same model.
         if message.get('role') == 'assistant' and message.get('model') != model['id'] and isinstance(output, list):
             message['output'] = strip_reasoning_details(output)
+        if message.get('role') == 'assistant' and isinstance(output, list):
+            message['output'] = strip_tool_images(message['output'], model)
 
     form_data['messages'] = process_messages_with_output(
         form_data.get('messages', []),
@@ -2990,6 +3008,8 @@ async def process_chat_payload(request, form_data, user, metadata, model):
         assistant_message = dict(pending_assistant_message)
         if assistant_message.get('model') != model['id'] and isinstance(assistant_message.get('output'), list):
             assistant_message['output'] = strip_reasoning_details(assistant_message['output'])
+        if isinstance(assistant_message.get('output'), list):
+            assistant_message['output'] = strip_tool_images(assistant_message['output'], model)
         form_data['messages'].extend(
             process_messages_with_output([assistant_message], reasoning_format=get_reasoning_format(model))
         )
@@ -5911,12 +5931,12 @@ async def streaming_chat_response_handler(response, ctx):
                             new_form_data['messages'] = (
                                 [system_message] if system_message else []
                             ) + convert_output_to_messages(
-                                output, raw=True, reasoning_format=get_reasoning_format(model)
+                                strip_tool_images(output, model), raw=True, reasoning_format=get_reasoning_format(model)
                             )
                             new_form_data['previous_response_id'] = last_response_id
                         else:
                             tool_messages = convert_output_to_messages(
-                                output,
+                                strip_tool_images(output, model),
                                 raw=True,
                                 reasoning_format=get_reasoning_format(model),
                                 flatten_tool_images=True,
@@ -6241,7 +6261,7 @@ async def streaming_chat_response_handler(response, ctx):
                                 'messages': [
                                     *messages,
                                     *convert_output_to_messages(
-                                        output,
+                                        strip_tool_images(output, model),
                                         raw=True,
                                         reasoning_format=get_reasoning_format(model),
                                         flatten_tool_images=True,