How can we achieve the coexistence of images and text?

[liuzhuosin]

I’ve noticed that when I use the manual mode to parse PDF documents, ragflow now uses a vision model to parse images in the documents, but the results aren’t quite ideal. Is there any way to output images via links? This would make it easier for others to understand.
Previously, I thought of converting PDFs to markdown and using image links in the markdown to output images, and I successfully did this on Dify, but I haven’t found a good way to do it in ragflow yet.
Would anyone who has managed to output images be kind enough to share some advice?

[dosubot[bot]]

RAGFlow can extract images from PDFs and associate them with text chunks during parsing. When you query the API (for example, /v1/chunk/list), each chunk may include an img_id if it contains an image. You can then construct a link to the image using /v1/document/image/<img_id>, which serves the image binary directly—this lets you output images as links in your own markdown or other formats if you want to present both text and images together reference.

Currently, RAGFlow does not natively output markdown with image links; images are rendered inline in chat/search responses if present, but not as markdown links. If you want markdown output with image links, you can post-process the chunked output: for each chunk with an img_id, insert a markdown image link like ![](/v1/document/image/<img_id>) at the appropriate spot in your text.

Compared to Dify, which supports markdown image links out of the box, RAGFlow requires this manual step. There’s no config option to change this behavior yet reference.

If you need a script or example for converting chunked output to markdown with image links, let me know!

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}    

[Magicbook1108]

Are you meaning you prefer to have the image url in chunks so that you can share this url with others?

[liuzhuosin]

I’m afraid I didn’t quite catch your meaning, but I just tested how Dify handles images without links for you—it directly uses OCR to scan the images and extract text information from them.
My approach is to convert the PDF to Markdown. Markdown generates a link address for the image, which is then stored in the knowledge base. During Q&A, the original image can be retrieved by keeping the link unchanged.

[Magicbook1108]

Now I understand — you mean that during the conversation, the image itself should be referenced right?

By the way, have you joined our WeChat group yet?

[liuzhuosin]

Yes, images themselves contain a wealth of information, and even outputting them as they are would be very helpful.
I’ve sent a friend request to RAGFlow Assistant, but they haven’t accepted it yet. Would you mind adding me first and then inviting me to the WeChat group?

[Magicbook1108]

The image description prompt is stored in rag/prompts/vision_llm_figure_describe_prompt.md. You can modify it to adjust the focus as needed.

Currently, the VLLM is instructed to extract any text that appears in an image, and it performs quite well in that regard. However, it does tend to lose information for non-text images. This is a known long-term issue, and we are continuously working to improve it.

Send me your wechat id via [email protected], I'll check the applications in "RagFlow-Assistant" and invite you to the group.

[liuzhuosin]

Thank you very much