ComfyUI-Gemma4: Integrating Google Gemma 4 Multimodal Large Model into ComfyUI

Title: ComfyUI-Gemma4: Integrating Google Gemma4 Multimodal Large Model into ComfyUI

Original Author/Maintainer: mailzwj Source Platform: GitHub Original Link: https://github.com/mailzwj/ComfyUI-Gemma4 Release/Update Date: 2026-06-14

Core Content: This project is an open-source plugin that integrates Google's newly released Gemma4 multimodal large model into ComfyUI workflows. It supports text generation, image understanding, and video understanding capabilities, breaking the barrier between traditional text models and image generation workflows, and enabling an end-to-end creation process from concept to finished product.

With the rapid development of multimodal large language models, AI image generation workflows are undergoing transformation. Google's Gemma4 series models, released at the end of 2025, possess strong deep understanding capabilities for text, images, and videos, making them an ideal choice for visual creation. As a popular Stable Diffusion graphical tool, ComfyUI has a large community and plugin ecosystem but lacks seamless integration with Gemma4—thus this project came into being.

ComfyUI-Gemma4 is an open-source custom node plugin created and maintained by developer mailzwj. It connects to the Gemma4-12B-it model via the ModelScope platform, achieving native integration of multimodal capabilities in ComfyUI. Its core value lies in allowing users to call Gemma4 capabilities within the ComfyUI interface without switching tools, completing end-to-end creation.

1. Text Generation: Provides dedicated nodes to generate high-quality prompts based on Gemma4, improving the quality and consistency of image generation, which is superior to traditional prompt engineering;
2. Image Understanding: Analyzes generated or reference image content, supporting scenarios such as image moderation optimization, style transfer assistance, batch annotation, and visual question answering;
3. Video Understanding: Analyzes video clips, extracts keyframe descriptions, summarizes themes, and aids in creation tasks like video cover generation.

The plugin adopts a modular node design, where each function corresponds to an independent configurable node; it accesses the model via ModelScope to lower the hardware threshold for local deployment; it follows ComfyUI's standard specifications and is compatible with existing nodes like Stable Diffusion and ControlNet, enabling the construction of complex multimodal generation pipelines.

For AI art creators: Assists in converting vague ideas into precise prompts, and understands the characteristics of generated content to control the direction of creation; For enterprise users: Integrates into automated processes, such as generating marketing copy based on product images in e-commerce scenarios, or generating news summaries based on news images in media scenarios.

ComfyUI-Gemma4 represents an important direction of fusion between multimodal models and creation tools, and we look forward to more cross-modal integration solutions. Users can experience it with a low threshold: no complex deployment is required—just install the plugin and configure the nodes to enjoy the creative innovation brought by multimodal AI.