langchain - 💡(How to fix) Fix fix(ollama): multimodal message text content incorrectly prefixed with \n

Submission checklist

• This is a bug, not a usage question.
• I added a clear and descriptive title that summarizes this issue.
• I used the GitHub search to find a similar question and didn't find it.
• I am sure that this is a bug in LangChain rather than my code.
• The bug is not resolved by updating to the latest stable version of LangChain (or the specific integration package).
• This is not related to the langchain-community package.
• I posted a self-contained, minimal, reproducible example. A maintainer can copy it and run it AS IS.

Package (Required)

• langchain
• langchain-openai
• langchain-anthropic
• langchain-classic
• langchain-core
• langchain-model-profiles
• langchain-tests
• langchain-text-splitters
• langchain-chroma
• langchain-deepseek
• langchain-exa
• langchain-fireworks
• langchain-groq
• langchain-huggingface
• langchain-mistralai
• langchain-nomic
• langchain-ollama
• langchain-openrouter
• langchain-perplexity
• langchain-qdrant
• langchain-xai
• Other / not sure / general

Related Issues / PRs

When a HumanMessage (or any message) has list-format content — which is required for multimodal inputs containing both text and an image — _convert_messages_to_ollama_messages unconditionally prepends \n to every text segment, including the very first one:

# langchain_ollama/chat_models.py, inside _convert_messages_to_ollama_messages
content = ""
for content_part in message.content:
    ...
    elif content_part.get("type") == "text":
        content += f"\n{content_part['text']}"   # ← \n always prepended

A call like:

HumanMessage(content=[
    {"type": "text", "text": "Extract all visible text from this image."},
    {"type": "image_url", "image_url": "data:image/png;base64,<data>"},
])

produces the Ollama message

content = "\nExtract all visible text from this image."

Impact

Vision models that are sensitive to exact prompt formatting return empty responses without any error or warning. One confirmed case: deepseek-ocr, which requires single-line prompts and silently produces an empty response when the content starts with \n. The failure is invisible to the user and very hard to debug.

Reproduction Steps / Example Code (Python)

from langchain_core.messages import HumanMessage
from langchain_ollama import ChatOllama

# A typical multimodal message: text instruction + base64 image
message = HumanMessage(
    content=[
        {"type": "text", "text": "Extract all text from this image."},
        {
            "type": "image_url",
            # Minimal valid base64 PNG (1×1 transparent pixel)
            "image_url": {
                "url": "data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAADUlEQVR42mNk+M9QDwADhgGAWjR9awAAAABJRU5ErkJggg=="
            },
        },
    ]
)

# Instantiate with any model name — no network call is made here
model = ChatOllama(model="any")

ollama_messages = model._convert_messages_to_ollama_messages([message])
content = ollama_messages[0]["content"]

Error Message and Stack Trace (if applicable)

Description

I´m trying to use deepseek-ocr running on my local hosted ollama to do an ocr of a document

System Info

System Information

OS: Linux OS Version: #1 SMP PREEMPT_DYNAMIC Thu Jun 5 18:30:46 UTC 2025 Python Version: 3.12.3 (main, Mar 23 2026, 19:04:32) [GCC 13.3.0]

Package Information

langchain_core: 1.4.0 langsmith: 0.8.5 langchain_ollama: 1.1.0 langchain_protocol: 0.0.15