KIE Gemini (LLM) [Experimental]:
KIE Gemini (LLM) is an experimental node designed to generate text using the Gemini 2.5/3 Pro/Flash models. This node is particularly useful for creating dynamic and contextually relevant text outputs by leveraging advanced language models. It allows you to input a variety of media types, such as images, video, and audio, alongside text prompts to enrich the content generation process. The node is versatile, supporting different roles for message prompts and offering options for streaming responses, including reasoning content, and even integrating Google Search for enhanced information retrieval. Its primary goal is to provide a flexible and powerful tool for generating text that can be tailored to specific needs, making it an invaluable asset for AI artists looking to incorporate sophisticated language capabilities into their projects.
KIE Gemini (LLM) [Experimental] Input Parameters:
model
This parameter allows you to select the Gemini model version you wish to use, such as 2.5, 3 Pro, or Flash. The choice of model can impact the quality and style of the generated text, with each version offering different capabilities and performance characteristics.
prompt
The prompt is a required text input that serves as the initial seed for text generation. It is essential when messages_json is not provided, as it guides the model in producing relevant and coherent responses based on the given context.
role
This parameter defines the role of the message within the conversation, such as developer, system, user, assistant, or tool. The role influences how the model interprets the prompt and generates responses, allowing for more nuanced and context-aware interactions.
images
An optional input that allows you to include a batch of images as media content. This can enhance the context of the text generation by providing visual references that the model can incorporate into its responses.
video
Similar to images, this optional input lets you include video content to enrich the media context. Videos can provide dynamic visual cues that the model can use to generate more contextually relevant text.
audio
This optional input allows you to include audio content, adding another layer of media context that can influence the text generation process. Audio inputs can be particularly useful for creating responses that are sensitive to auditory cues.
messages_json
An optional JSON array of message objects that can override the prompt, role, and media inputs. This parameter provides a structured way to input complex conversation histories or predefined message sequences for the model to process.
stream
A boolean parameter that, when enabled, allows the node to stream responses using Server-Sent Events (SSE). This means that the output is returned incrementally as it is generated, rather than waiting for the entire response to be completed.
include_thoughts
A boolean parameter that, when enabled, includes reasoning content in the output. This can provide insights into the model's thought process and enhance the interpretability of the generated text.
reasoning_effort
This parameter can be set to either low or high, determining the level of effort the model should put into reasoning. A higher setting may result in more detailed and thoughtful responses, but could also increase processing time.
enable_google_search
A boolean parameter that, when enabled, allows the node to use the Google Search tool to retrieve additional information. This option is mutually exclusive with response_format_json, meaning only one can be active at a time.
response_format_json
An optional parameter that specifies a JSON schema for the output format. This is mutually exclusive with enable_google_search, allowing you to define a structured format for the generated responses.
log
A boolean parameter that controls console logging. When enabled, it provides detailed logs of the node's operations, which can be useful for debugging and understanding the model's behavior.
KIE Gemini (LLM) [Experimental] Output Parameters:
STRING: Assistant response text
This output provides the main text generated by the assistant, based on the input parameters and media content. It is the primary result of the node's text generation process.
STRING: Reasoning text
If include_thoughts is enabled, this output contains the reasoning text generated by the model. It offers insights into the model's decision-making process and can be empty if include_thoughts is false.
STRING: data
This output contains formatted JSON from the last response chunk, providing a structured representation of the generated content. It is particularly useful when a specific output format is required.
KIE Gemini (LLM) [Experimental] Usage Tips:
- To achieve the best results, carefully craft your prompt to provide clear and specific guidance to the model. This will help in generating more relevant and coherent responses.
- Experiment with different model versions to find the one that best suits your needs. Each version may offer unique strengths in terms of style and performance.
- Utilize the
include_thoughtsparameter to gain insights into the model's reasoning, which can be valuable for understanding and refining the text generation process.
KIE Gemini (LLM) [Experimental] Common Errors and Solutions:
Failed to call chat completions endpoint
- Explanation: This error occurs when the node fails to connect to the chat completions endpoint, possibly due to network issues or incorrect endpoint configuration.
- Solution: Check your network connection and ensure that the endpoint URL is correctly configured. Retry the operation after verifying these settings.
chat completions returned HTTP 429
- Explanation: This error indicates that the request rate limit has been exceeded, resulting in a temporary block from the server.
- Solution: Wait for a while before retrying the request. Consider implementing rate limiting in your application to prevent this error in the future.
chat completions returned HTTP 500
- Explanation: This error signifies a server-side issue, which could be due to temporary server overload or internal errors.
- Solution: Retry the request after some time. If the issue persists, contact support for further assistance.
