Phoenix does support displaying images in multimodal traces, but images may not appear if the base64-encoded image exceeds the default size limit or if certain masking settings are enabled. To ensure images display, set the environment variable `OPENINFERENCE_BASE64_IMAGE_MAX_LENGTH` to a higher value (e.g., "1000000") and make sure `OPENINFERENCE_HIDE_INPUT_IMAGES` and `OPENINFERENCE_HIDE_INPUTS` are set to "False" before importing Phoenix or OpenInference. This is a known issue and not a documentation gap鈥攕ee the troubleshooting steps and user confirmations in recent GitHub issues and the official docs for details on configuration and limitations. If images still do not show, check for errors in Phoenix and browser logs, and ensure your instrumentation libraries are up to date and correctly propagating image payloads. According to the documentation, this applies to both base64 and data URL images, and the default limit is 32,000 characters, which often needs to be increased for real images.
Would you like more detail on how to set these environment variables or debug further?
Sources:
- <https://arize.com/docs/phoenix/tracing/how-to-tracing/advanced/multimodal-tracing|Phoenix: Capture Multimodal Traces>
- <https://github.com/Arize-ai/phoenix/issues/10868|GitHub Issue: Self-hosted Phoenix does not render images from base64 strings of Langgraph traces>
- <https://github.com/Arize-ai/phoenix/issues/5564|GitHub Issue: Prompts containing Images with data URLs are not displayed in the UI>
