MiniMax H3 is the first truly omni-modal video generation model Craze offers. It understands all your inputs at once: text, images, videos, and audio. That unified understanding lets you reference a character from an image, motion from a clip, and mood from an audio track, all in the same prompt. The result is coherent video that reflects the relationships between your references.

Why creators choose MiniMax H3

MiniMax H3 is the first truly omni-modal video model Craze offers. It understands all your inputs at once: text, images, videos, and audio. That unified understanding lets you reference a character from an image, motion from a clip, and mood from an audio track, all in the same prompt. For a faster variant, see MiniMax H3 Max.

Unified multimodal understanding

H3 doesn't treat text, image, and video as separate inputs with separate weights or processing paths. It reads your prompt and your references together, understanding how they relate. That coherence shows in the output. When you pair a character image with a motion reference video and an audio mood track, H3 synthesizes all of them into a single coherent generation instead of balancing disparate input types.

Accurate with detail, text, and brand elements

MiniMax says H3 excels at instruction following, accurate brand rendering, and text in the image. That accuracy matters for logos, labels, and legible design elements. In commercial work where a brand's visual identity is non-negotiable, H3's accuracy with text and color is a significant advantage over competitors.

Native stereo sound always included

H3 always generates stereo sound together with the video. You can silence it in the edit if your spot carries voiceover or music, but having sound baked in saves a step. The audio is synchronized with the video in a single generation, reducing the friction of audio post-production.

Recognizable public figures with accuracy

Craze uses MiniMax H3 when you need video with recognizable public figures' likenesses to stay accurate. It's the tool for on-model celebrity and influencer work. The combination of multimodal understanding and facial accuracy makes it the right choice for work where the person on screen is as important as the scene around them.

Up to 4K resolution

At up to 4K, H3 supports high-resolution output suitable for commercial work. The per-second rate increases with the chosen resolution.

How to use MiniMax H3 in Craze

You combine all your references in one brief and let H3 understand them together.

1. Describe what you want. Add reference images for look, video clips for motion, and even audio tracks for mood or voiceover reference.
2. Name MiniMax H3 or let Craze show you available options and prices before planning.
3. Craze generates the clip with all your references in mind.
4. Review and regenerate if needed. Try different reference combinations to dial in the look.

Specific direction helps: "Keep the character's expression curious" or "Match the motion of the reference video" tells the model what to prioritize. Craze carries your direction into the generation.

MiniMax H3 pricing

You pay per second of video. The rate scales across 480p, 768p, 2K, and 4K, starting at $0.053 per second of video. All clips include sound. The reference-to-video rate also has a surcharge at nine references; check the calculator for your options.

Who MiniMax H3 is for

Celebrity and public figure work

H3's accuracy with faces makes it the tool for recognizable likenesses. Use it when the person on screen matters as much as the setting. The multimodal input support means you can provide a photo of the person, a motion reference showing the desired movement style, and audio reference capturing the mood, and H3 synthesizes all of them.

Brand films with consistent detail and accuracy

H3's accuracy with text and logos matters for product work where brand consistency is important. Choose up to 4K resolution when fine detail matters.

Multimodal creative direction

If you're working with multiple references (a character look, a motion style, and a mood), H3 lets you express all of that in one prompt instead of guessing which inputs the model prioritizes. You don't need to choose between referencing motion and referencing audio mood; you can do both.

Quick turnaround at high quality

H3 generates fast. That speed, combined with its accuracy and resolution, makes it practical for campaigns and commercial work on deadline. The unified multimodal processing means fewer iterations needed to dial in the right aesthetic.

MiniMax H3 vs MiniMax H3 Max

Base H3 is more accurate, especially with facial features. H3 Max generates faster and costs less, but supports 1080p instead of H3's 4K ceiling. Choose H3 for accuracy with faces. Choose H3 Max when iterating fast and 1080p is sufficient.

MiniMax H3 vs MiniMax Hailuo 02 Pro

MiniMax H3 generates up to 15 seconds with always-on stereo sound, at multiple resolutions. Hailuo 02 Pro generates 6 or 10 seconds silent at 1080p only. Choose H3 for longer clips with sound included. Choose Hailuo for 1080p silent video and precise audio post-production.