Kling Image O1 is a multimodal image generation and editing model developed by Kuaishou Technology as part of the Kling Omni ecosystem. Released on December 1, 2025, the model is designed to unify text, image, and video processing within a single engine. It distinguishes itself from traditional generators by emphasizing semantic reasoning, allowing users to perform complex compositions and edits using natural language instructions rather than manual brushes or masks.
The model is built on a Multimodal Visual Language (MVL) architecture, which treats visual elements—such as subjects, styles, and lighting—as tokens within a unified semantic space. This framework enables the model to understand contextual relationships between multiple inputs, supporting commands like "place the subject from reference image A into the setting of reference image B" while maintaining perspective and lighting consistency.
Key Capabilities
- Multi-Reference Support: Users can upload up to 10 reference images to guide a single generation. This feature is specifically optimized for maintaining character consistency in serial content or preserving brand identity across different visual assets.
- Text-Based Image Editing: The model supports precise local modifications via text prompts. Users can add, remove, or transform specific elements (e.g., "change the orange juice to blue soda" or "remove the person in the background") while the model intelligently preserves the original image's texture and style.
- High-Resolution Output: Kling Image O1 supports generation and editing at resolutions up to 2K, with flexible aspect ratio support and batch generation capabilities.
Creators are encouraged to use a numbered reference system (e.g., @Image1, @Image2) within their prompts to explicitly define how various source images should interact. The model excels in workflows requiring high feature retention, such as IP character design, virtual try-ons for e-commerce, and professional storyboarding.