Can Rovo Analyze Images in Confluence? [Champion Slack Insider]

[#1]

This question from @Fabio Genovese _ArtigianoDelSoftware_  comes up more often than you’d think—especially for teams with visual-heavy documentation:

“Can Rovo analyze images in KB articles and describe diagrams for users?”

Short answer: Not reliably (yet).

What Rovo can do today

Rovo can:

• Reference KB articles that contain images
• Surface those articles in responses
• Preserve images for users to view in the original content

So users can still:

• See manuals
• Follow diagrams
• Access visual documentation

What Rovo cannot do (today)

Rovo does not reliably interpret image content. That means it cannot:

• Read diagrams
• Understand flow arrows or visual relationships
• Extract meaning from labels inside images
• Accurately describe what’s happening in a visual

If it does describe an image, it’s usually based on:

• Surrounding text
• Captions
• Context clues

Not true image understanding

Why this matters for your use case

In Fabio’s scenario:

• Users rely on diagrams
• Images carry key meaning
• The goal is to explain those visuals

Right now:

• Rovo can point users to the right place
• But it cannot replace the interpretation of the image itself

What customers are doing as a workaround

Teams are getting good results by making images more “AI-readable”:

1. Add descriptive captions

• Explain what the image shows
• Include key steps or relationships

2. Use alt text strategically

• Treat it as searchable, meaningful content
• Not just accessibility filler

3. Pair diagrams with short explanations

• Even 1–2 sentences makes a big difference

4. Store critical visuals with supporting text

• Ensure Rovo has something to index

[Fabio Genovese _ArtigianoDelSoftware_]

@Dr Valeri Colon _Connect Centric_  An excellent article. If this is the result of a simple chat, I can’t imagine what we might write if we were to meet in person or via videocall. 

[Ryan Boyd]

I've seen pretty good outputs wrt image recognition. We're using it to evaluate server rooms with the use case written up on LinkedIn here.

[Dr Valeri Colon _Connect Centric_]

@Ryan Boyd  you’ve essentially added a 5th workaround—attaching images directly in chat. In your use case, you’re leveraging Rovo with image-capable models in a specific workflow (explicit image input → model analyzes it), rather than Rovo passively analyzing images in Confluence pages (retrieving indexed content that includes images). When Rovo works from indexed content, it doesn’t reliably interpret images yet—so the capability exists, just not consistently in the KB retrieval experience.

[Dr Valeri Colon _Connect Centric_]

@Fabio Genovese _ArtigianoDelSoftware_ we’d make magic, no doubt. The Champion Slack channel always pushes me to think a level deeper, and this was definitely one of those moments. Imagine what we’d come up with in a live session or whiteboard moment—might have to make that happen...

[Ryan Boyd]

Ok...