ImageConverter crashes when LLM API call fails (auth error, rate limit, etc.)

[hanhan761]

Description

ImageConverter.convert() crashes the entire conversion pipeline when the LLM API call in _get_llm_description() fails (e.g. authentication error, rate limit exceeded, network timeout). Unlike PptxConverter which wraps its llm_caption() call in a try/except (lines 119-129), ImageConverter._get_llm_description() makes the OpenAI API call at line 137 without any error handling.

Root Cause

packages/markitdown/src/markitdown/converters/_image_converter.py, line 137:

response = client.chat.completions.create(model=model, messages=messages)

If this call raises any exception (e.g. openai.AuthenticationError, openai.RateLimitError, openai.APITimeoutError), the exception propagates up and fails the entire conversion.

Expected Behavior

A failed LLM caption should be gracefully handled: silently skip the description, and still return the successfully extracted metadata. The image should still be converted to Markdown, just without the LLM-generated caption.

Proposed Fix

Wrap the API call in a try/except block, similar to how PptxConverter handles it:

try:
    response = client.chat.completions.create(model=model, messages=messages)
    return response.choices[0].message.content
except Exception:
    return None

Impact

Anyone using ImageConverter with an LLM client configured will have their entire conversion fail if the API call fails for any reason.

[carter-smalls]

This looks like OpenAI rate limit errors hitting without exponential backoff — requests may be retrying too aggressively.

Worth logging the full HTTP response on a rate limit — check if Retry-After is present and how long the wait is.