2.8 KiB
2.8 KiB
KIE Gemini 3 Pro (LLM) [Experimental]
Generate text using Gemini 3 Pro. This node is experimental.
Inputs
- prompt (STRING, required): Main text prompt. Used when
messages_jsonis empty. - role (COMBO, required): Message role for the prompt. Options:
developer,system,user,assistant,tool. - images (IMAGE, optional): One or more images (batch) to include as media content.
- video (VIDEO, optional): A single video input to include as media content.
- messages_json (STRING, optional): Full JSON array of message objects. If provided, it overrides
prompt,role, and media inputs. - stream (BOOLEAN, optional): Use streaming responses (SSE). Output is returned after completion.
- include_thoughts (BOOLEAN, optional): When enabled, reasoning content is returned in the reasoning output.
- reasoning_effort (COMBO, optional): Controls reasoning depth:
loworhigh. - tools_json (STRING, optional): JSON array of tool definitions. Mutually exclusive with
response_format_json. - response_format_json (STRING, optional): JSON schema output format. Mutually exclusive with
tools_json. - log (BOOLEAN, optional): Enable console logging for uploads and response status.
Node UI Reference (What each widget expects)
- Prompt (text box): A natural language instruction or question. Leave empty only if you provide
messages_jsonor media. - Role (dropdown): Sets the
rolefield in the message object (default:user). - Images (input socket): Connect a ComfyUI IMAGE or batch. All images are uploaded and appended to message content.
- Video (input socket): Connect a ComfyUI VIDEO. The video is uploaded and appended to message content.
- Messages JSON (text box): Raw JSON array of messages (advanced). Overrides prompt/role/media entirely.
- Stream (toggle): Enables SSE streaming under the hood; output still returns when complete.
- Include Thoughts (toggle): When on, reasoning text is captured and returned in the second output.
- Reasoning Effort (dropdown):
lowis faster,highis more thorough. - Tools JSON (text box): Raw JSON array of tool definitions (advanced).
- Response Format JSON (text box): Raw JSON schema definition (advanced).
- Log (toggle): Emits upload and completion messages to the console.
Outputs
- STRING: Assistant response text.
- STRING: Reasoning text (empty if include_thoughts is false).
- STRING: Raw JSON from the last response chunk.
Helper behavior
- Validates
reasoning_effort. - Enforces mutual exclusivity for tools vs response_format.
- If
messages_jsonis provided, it is used as the message array and media inputs are rejected. - Media inputs are uploaded and added to the message content using the unified
image_urlformat.