Agent node with multimodal models cannot use vision — is this expected, and can it be supported?

[clongulin]

Self Checks

• I have searched for existing issues search for existing issues, including closed ones.
• I confirm that I am using English to submit this report (我已阅读并同意 Language Policy).
• [FOR CHINESE USERS] 请务必使用英文提交 Issue，否则会被关闭。谢谢！:)
• Please do not modify this template :) and fill in all the required fields.

Content

I’m using a multimodal model in the Agent node, but the Agent seems to operate in text-only mode. In contrast, when using a multimodal model in the LLM node, the vision capability works. I’d like the Agent node to support vision like the LLM node. Are there specific challenges preventing this? If so, what are they?

[dosubot[bot]]

Hi @clongulin! I'm Dosu and I’m helping the dify team.

Yes, it's expected that the Agent node currently operates in text-only mode, even with multimodal models. The LLM node supports vision by explicitly handling multimodal inputs and adapting prompts based on the model's capabilities, but the Agent node is architected for text-based reasoning and tool use, and doesn't forward images or other non-text inputs to the model. This is a recognized limitation and supporting vision in the Agent node would require significant changes to its input schema, workflow logic, and output parsing to handle richer payloads. Multimodal/vision support for Agent nodes is on the roadmap, but not available yet reference.

To reply, just mention @dosu.

---

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

[clongulin]

Have any relevant personnel provided an answer? Thank you.