A Photo, A Blockbuster: MiniMax(Hailuo AI) Multimodal Generation Technology Innovates Again

Introduction of Hailuo AI

Everyone harbors a dream of the movies—whether it’s stepping into different roles to experience life on screen, becoming a director framing every shot, or a screenwriter creating endless possibilities in parallel universes.

Hailuo AI acts as a dream machine, offering everyone a movie-like experience. At the start of the new year, Hailuo AI unveils a new creative assistant for global users: Subject Reference.

MiniMax’s latest self-developed S2V-01 video model enables precise visual detail restoration through a single-image subject reference framework.

With less than 1% of the input and computing cost of traditional solutions.

Simply upload an image, and high-quality video generation begins instantly, with highly accurate subject consistency and creative freedom.

Currently, the Subject Reference feature is available globally.

Users can try it immediately on the Hailuo AI video creation platform.

A Photo, A Blockbuster with Hailuo AI

In the AI video generation field, maintaining the realism and stability of character faces from multiple angles in dynamic videos, and ensuring consistency when stitching continuous clips together, have always been challenges.

Through our S2V-01 video model, we offer users an optimal solution.

After selecting the “Subject Reference” feature in Hailuo AI, users only need to upload a single image, and the system recognizes and locks the subject character.

By entering prompt keywords, a high-quality video is generated immediately, maintaining creative consistency.

The S2V-01 model accurately identifies facial features like gender, age, skin color, and facial structure, ensuring stability and coherence across frames.

prompt:A close-up of a young lady in a dimly lit room, his eyes fixed on the glowing screen of a gaming console. The camera is positioned slightly above eye level, focusing on his concentrated expression as his fingers nimbly manipulate the controller. A game character appears, breaking free from the screen’s confines.

Subject reference+Prompt：A male officer opened the door and got out of the police car. The camera followed the man and stayed in close-up, focusing on the man’s face. The man was wearing a police uniform. The man’s expression changes from calm to menacing. The city is surrounded by a night scene, and there are several police cars with flashing lights around.

It excels at facial expression control for the main character, while maintaining high-quality visuals for non-subject scenes.

Currently, Hailuo AI supports single-character references, requiring identifiable facial features as input. Future updates will expand this capability to include multiple subjects, objects, and scenes.

Lower cost, computing expenses reduce, better experience with Hailuo AI

Since its early development, MiniMax(Hailuo AI) has been exploring image-based references for roles, styles, and more.

After extensive technical research, we believe image reference solutions for subject consistency offer high effectiveness and scalability, surpassing fine-tuned LoRA solutions in some cases.

We aim to provide technology that serves a broad user base while solving real-world problems.

The subject reference solution only requires one image for input, with minimal computation and waiting time.

This drastically reduces both user input costs and computation time, providing a superior user experience. The computing expenses will be reduced to below 1%.

Prompt: A woman in an elaborate gown and a pair of white gloves walks through a corridor in a medieval castle. She runs with her back to the camera, then looks back to the camera, her expression changing from calm to horror. The end of the corridor is dimly lit. The camera follows the woman as she pushes closer and the view changes from medium to close-up, focusing on the woman’s face.

To ensure the video retains only essential visual information (like facial features) and avoids distractions from posture, expression, or lighting, MiniMax continually optimizes its data structures and model architecture.

The S2V-01 model achieves key effects such as precise restoration of visual details and high creative freedom, allowing characters to express any pose or expression and fit naturally into any environment.

With the subject reference technology, users can focus on content creation instead of worrying about consistency, thus dramatically improving the efficiency of long-video production.

Your character is inherently consistent.

New era of AI co-creation with Hailuo AI

AI technology has already brought ease to industries like microfilms, advertisements, variety shows, and CG effects.

However, the biggest challenge in video generation is the instability of subjects, which often leads to disjointed or inflexible results.

The Subject Reference feature offers professional creators high consistency and flexibility, bringing a disruptive innovation to video industries like short-form content and advertising.

MiniMax’s platform now includes this feature as an API service, with plans to extend it to multi-subject references.

Since launching its video models, Hailuo AI has been a focal point in the industry.

In December 2024, MiniMax’s I2V-01-Live image-to-video model received widespread praise, and Hailuo AI’s overseas visits exceeded 27 million, setting a new record and topping the global AI video product rankings.

Human interaction with the world is inherently multimodal, and multimodal understanding and generation are critical to advancing toward AGI.