Skip the 2000-word prompt describing camera moves; spend five minutes building a 3D animatic instead.

On August 20, MiniMax officially launched MiniMax Design, a multimodal creative Agent workbench built around H3, the natively multimodal video model open-sourced on July 31. The standout feature is called the "3D Director Console" — you can position characters, adjust poses, and set camera angles in a simplified 3D space, confirm composition and spatial relationships from multiple viewpoints, then feed that 3D previsualization as a reference to H3 for final video generation. For the first time, AI video creation has a professional "rehearse before shooting" workflow.

Anyone familiar with AI video knows the pain: you write a hyper-specific prompt down to focal length and aperture, but the generated video has clipping characters, random camera jumps, composition nothing like what you wanted — and your only option is to regenerate repeatedly like pulling a slot machine. Get lucky, one in ten is usable; get unlucky, dozens of attempts go nowhere. The 3D Director Console is conceptually simple: lock down the most error-prone spatial relationships in 3D first, so the AI only needs to render and fill detail within a defined framework rather than guessing at camera and composition from scratch.

From "Model" to "Harness": Toolchain Evolution in AI Video

MiniMax calls Design the first official Harness for H3 since open-sourcing — and "harness" is the key word. In AI, a harness is the engineering mechanism wrapped around a model: task planning, context management, tool invocation, asset organization, multi-step orchestration, error correction — essentially, how you organize raw model capability into a production-ready tool.

The past six months of AI video competition have largely been at the "model" layer: who generates sharper video, smoother motion, longer clips, better consistency. Sora, Kling, Hunyuan Video, H3 — everyone chases each other on model quality, but the creator tool experience universally remains at the primitive "prompt in, wait for generation" stage. Professional creators don't need a model that's better at guessing prompts; they need a complete production tool that plugs into their existing workflow.

Design takes an Agent-native approach. You don't write complex prompts; describe your creative goal in natural language, and the Agent automatically interprets requirements, decomposes tasks, calls text/image/video/audio models to process assets, then handles editing, cutting, and subtitling in post. For pro creators already using ComfyUI for node-based workflows, Design supports importing existing local workflows, running in cloud or local mode, with the Agent assisting via conversation to adjust node parameters.

How Far Until AI Video's "Blender Moment"?

QbitAI's coverage offered a sharp framing: AI video creation is entering the "previsualization era." That's accurate. The 3D Director Console solves AI video's biggest uncertainty — space and composition — turning "lottery generation" into "plannable creation."

But this is only the beginning. Real professional video production is extraordinarily complex: screenwriting, storyboarding, character design, set building, animation, lighting, rendering, editing, color grading, sound design, scoring — AI can currently participate in a limited subset of stages. The 3D Director Console effectively upgrades "storyboarding" from text to 3D previs, but there's a long road ahead.

The good news is the direction is clear. Just as Blender integrated 3D modeling, animation, rendering, and post into a single open platform, AI video will see its own "Blender moment" — a unified, open, extensible creation platform that seamlessly combines model capability with professional workflows. MiniMax Design is a step in that direction, but it's a closed SaaS product; the real "Blender of AI video" will likely emerge from open source.

Design currently covers scenarios including KOC content, UGC short video, e-commerce seeding, shoppable video, educational content, and creative shorts. These are all cost-sensitive, fast-turnaround use cases that match AI video's current capability envelope. A MiniMax representative stated that Design aims to further organize H3's model capabilities into commercial content production workflows while continuously incorporating open-source community workflow patterns.

H3 is a full-modal generative model released July 31 that supports unified understanding and generation of text, images, video, and audio, with output including native dual-channel audio. From model release to official creative workbench: MiniMax did it in 20 days — faster than many expected. As model capabilities continue to evolve and toolchains mature, the inflection point where AI video shifts from "toy" to "production tool" may arrive sooner than most expect.

See you tomorrow.