Showcases

MiniMax H3 Scene Storyboarding: From Text to Image Transformation Example

A two-stage "text → concept image → dynamic clip" demonstration of MiniMax H3's multimodal reference capability. The work first uses a concept storyboard image to establish composition, color, and character, then feeds that image into H3 as reference material to generate a 2K landscape short film, verifying how the reference image constrains visual style and subject consistency. At the schema level it deliberately takes the "multimodal reference video generation" branch: image_urls carries the concept image, prompt describes the camera language and pacing, and first_frame_image/last_frame_image first/last frame control is not used, avoiding conflicts with the reference media fields. duration is set to 6 seconds, aspect_ratio is fixed at 16:9, resolution uses the currently only available option 2K, and watermark remains off. Use cases: storyboard proofing for short videos and ads, animating illustrations, batch shot previsualization requiring character or style consistency, and reusing open-weight video generation pipelines in local workflows such as ComfyUI. Developers can treat this entry as the minimal runnable sample of the "concept image → reference video" input combination; replacing the reference image in image_urls reproduces similar compositions.

Style Scene storyboard · reference-to-video · Cinematic

Model MiniMax H3

Reference inputs

Reference images

Try it yourself

Prompt

Try this model

A 6-second cinematic scene adapted from the reference concept image: early morning light rakes across rain-soaked city rooftops as a lone courier pauses by a railing, coat fluttering in the wind. The shot opens with a wide establishing view and slowly pushes in, then transitions smoothly into a gentle lateral tracking move as the figure turns toward the skyline; shallow depth of field keeps the silhouette crisp against softly blurred high-rises and drifting clouds. Consistent with the reference image's soft teal-and-amber palette, gentle volumetric haze, and fine film grain. Natural 24fps motion, slight handheld micro-shake, no cuts, no text overlays.

Copy the full prompt and adapt it for your own generation.