[Bug]: AI summary embeds base64 images in Markdown, causing large files and slow loading

Existing issues

• I searched the existing issues and did not find a report of the same problem.

What happened?

The AI summary function generated Markdown node files containing images as inline data:image/...;base64,... payloads. This makes the authored Markdown unexpectedly large and causes nodes to open slowly, especially on poor network connections.

Observed in two AI-generated summary nodes:

More than 99.9% of each file is therefore attributable to lines containing embedded image data. The image payloads and private node content are intentionally omitted from this report.

What did you expect?

AI-generated summaries should store images as managed Huabu assets and reference them by URL/path in Markdown, rather than embedding large base64 data URLs. Opening a summary node should not require downloading and parsing roughly 10.8 MB of Markdown.

Steps to reproduce

1. Use the AI summary function on source content containing multiple images or slides.
2. Allow the generated summary node to be applied/saved.
3. Inspect the generated node Markdown file.
4. Observe multiple inline data:image payloads and a file size of roughly 10.8 MB.
5. Open the generated node over a slow network and observe delayed loading.

How often does it happen?

Observed in both inspected AI-generated summary files (2/2).

System information

Not collected; this is a persisted-content issue confirmed directly in the generated Markdown files.

AI-related details

• Feature: AI summary
• Output: Markdown node containing inline base64 image data
• Provider/model: Unknown
• Proposed changes were applied and persisted to the node files

Sensitive information

• I reviewed this report and removed node content, embedded image data, and sensitive information.

No source files or tests are named in the report. Start by locating the AI summary function and the code that persists generated Markdown and images; trace how image content becomes node Markdown. Done means generated summaries store images as managed Huabu assets and reference them by URL or path, without large data URLs, and the reported reproduction no longer produces roughly 10.8 MB files.