Although launching the video model in one day with Seedance 2.5 might have led to its displacement, MiniMax's H3 model was an exception. After it was released in early August, users demonstrated enormous interest, particularly using the Chinese Wonderland series. Users generated high-quality reference images using Midjourney and then uploaded them to H3. Users noted that videos created by H3 feature advanced visual taste, smooth motion, and character stability.
MiniMax has now embodied this model functionality into a ready-made product—MiniMax Design. Similar to how GPT-5.6 Sol plus Codex or DeepSeek V4 plus DeepSeek Harness unlock the true agentic potential of a model, MiniMax Design integrates H3 into a workflow that supports continuous modification, collaboration, and final project delivery.
The interface looks professional, offering basic elements in the style of ComfyUI and a full canvas, but the main work is performed by a team of agents. Users do not need to understand the canvas, camera positions in the 3D director's cabin, or ComfyUI connection schemes; they simply describe the desired outcome, and MiniMax Design handles the entire production chain, utilizing skills, models, and agents to parse and execute the request. Its main task is to structure professional capabilities into executable nodes that activate H3, image, music, and voice models, bringing AI video creation closer to full production.
Instead of writing text prompts for AI video and copying them from ChatGPT into Hailuo or other web tools, users can switch to MiniMax Design: a simple requirement is enough, and the system itself creates the canvas, generates scripts, storyboards, images, music, dialogues, editing, subtitles, and the video itself. Authors reproduced a scene from the film Niu Coming by inputting a prompt to stage the scene in a director's cabin, where the heroine and hero walk in a meadow, meet by a wide river, she jumps across it, and he walks along the bank. Creating this scene was complex but only required uploading materials and communication.
The 3D director's cabin allows users or agents to place character models and control camera positions and poses in three-dimensional space, creating references for storyboarding and reducing trial-and-error costs. The complexity lay in the spatial relationships between characters; MiniMax Design first generated a camera movement video from the cabin and then created the precise final video using this reference. The value of the cabin lies in transforming vague requirements into specific spatial and cinematic benchmarks: if the action is incorrect, users adjust the characters and cameras, after which they regenerate the frame. In addition to such complex scenes, MiniMax Design easily handles various spectacular videos.