Multimodal parameters

[kingWYZ]

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

When I select a multimodal model, after turning on the visual switch, what is the purpose of the variable values that need to be input? Are the images obtained from here then fed to the model?

[dosubot[bot]]

Yes — those variable fields tell the LLM node where to pull images from so they can be sent to the multimodal model.

When you enable the vision switch, you configure a variable selector that points to the source of image files in your workflow. The most common options are:

1. sys.files (default) — images uploaded by the user through the chat interface's file-upload button [1].
2. Output from a previous node — e.g., a file returned by an HTTP request node, a Code node, or another tool.
3. Context files — images retrieved from a Knowledge Base via a Knowledge Retrieval node [2].

Once the variable is resolved at runtime, the system converts those files into image prompt messages and appends them to the prompt that is sent to the model [3]. You can also choose a detail level (high/low) which maps to the API's image quality parameter [4].

So in short: the variable value you input is simply a pointer to the image source within your workflow, and the images obtained from that source are indeed fed to the model along with the text prompt.

To reply, just mention @dosu.

---

Share context across your team and agents. Try Dosu.

[kingWYZ]

The value I have checked now is {'title': 'Large Model', 'desc': None, 'version': '1', 'error_strategy': None, 'default_value': None, 'retry_config': {'max_retries': 0, 'retry_interval': 0, 'retry_enabled': False}, 'model': {'provider': 'lcjx/openai_api_like/openai_api_like', 'name': 'custom-multimodal-user', 'mode': 'chat', 'completion_params': {}}, 'prompt_template': [{'text': 'Hello', 'role': 'system', 'edition_type': None, 'jinja2_text': None}, {'text': 'Hello', 'role': 'user', 'edition_type': None, 'jinja2_text': None}], 'prompt_config': {'jinja2_variables': []}, 'memory': None, 'context': {'enabled': False, 'variable_selector': []}, 'vision': {'enabled': True, 'configs': {'variable_selector': ['1779680684548', 'a'], 'detail': 'high'}}, 'structured_output': None, 'structured_output_switch_on': False, 'reasoning_format': 'tagged'}}, how can I get the file or url?

[dosubot[bot]]

The variable_selector: ['1779680684548', 'a'] in your vision config is a reference path into the workflow's variable pool:

• 1779680684548 — the node ID of an upstream node in your workflow
• 'a' — the specific output variable name from that node

At runtime, when the LLM node executes, it calls variable_pool.get(['1779680684548', 'a']) to retrieve the file(s) that upstream node produced [1]. The returned value is a FileSegment or ArrayFileSegment containing File objects that hold either a remote_url (for URL-based images) or a storage_key (for locally uploaded files) [2].

So to answer your question: the upstream node with ID 1779680684548 must output a file-type variable named a. That node could be a file-upload start variable, an HTTP Request node that downloads an image, a Code node that constructs a File object, etc. The LLM node then automatically extracts the URL or file data from it and passes it to the multimodal model.

If you're building this programmatically, make sure the upstream node's output variable a is of type File or ArrayFile — otherwise fetch_files() will return an empty list and no image will be sent to the model [3].

To reply, just mention @dosu.

---

Docs are dead. Just use Dosu.