Image / File / Multimedia rendering in dataset items using mediaId

[ArthurMor4is]

Describe the feature or potential improvement

Currently, when adding a dataset item that includes an image or a file (such as type: file with @@langfuseMedia:id=...), there is no visual rendering of the media in the UI — only the JSON appears. This makes it difficult to quickly verify if the correct file was attached or to inspect its content visually.

It would be extremely helpful if Langfuse could render these media assets directly within the dataset item view. For example, when an image or PDF file is attached using the mediaId, the interface could display a thumbnail or a preview, as done in the Multi-Modality tracing view.

This would significantly improve the developer experience and dataset curation workflow, especially for teams working with OCR, document understanding, or image-based LLM pipelines.

Additional information

The current dataset view does show the file_data structure with embedded media references, such as @@langfuseMedia:id=..., but it does not visualize the content. A similar mechanism is already implemented in tracing, as documented here, where images and audio are rendered in context.

Enhancing the dataset UI to support basic rendering of image/PDF previews — when applicable — would bring consistency with other Langfuse features and provide immediate value for multi-modal debugging and annotation workflows.

[xxchan]

link feature request here https://github.com/orgs/langfuse/discussions/8860
Beyond rendering media, we hope langfuse can support:

• upload media directly in dataset (UI/SDK)
• do not restrict attachment type

---

For reference, braintrust support uploading attachments in any editable fields in datasets. And it does not have limits on attachment types.

https://www.braintrust.dev/docs/guides/attachments#in-the-ui

[haronka1999]

Hello LangFuse team, is this still on plan, shall we expect this feature? Or is there any other way to render images in datasets?

Thank you!

[Hjalmar-Ohman]

+1 👀