Back to directory

H3 Prompt Journal

A journal of case studies documenting prompt-engineering experiments for the MiniMax H3 video model, recording what failed and which final prompts worked.

H3 Prompt Journal

H3 Prompt Journal is a collection of case studies from prompt-engineering experiments with the MiniMax H3 video generation model. Each entry documents a real attempt: the failure modes the author ran into, the insight that resolved them, and the final prompt that worked, laid out as a fixed pair of files per case, a README with the context and rationale and a prompt file meant to be pasted directly into H3.

The journal argues that most public H3 prompts fail because they describe what should appear in each shot rather than the continuous physical logic the camera must follow, and the collected entries are meant to demonstrate prompts that captured that logic, covering techniques such as multi-person choreography, camera anchoring, and beat-synced editing.

It is a companion to a separate repository that reverse-engineers H3 prompts from example video.

More in Generative media

See all 146
  • Blender Copilot

    github.com

    A Copilot-style AI chat panel inside Blender that runs its agent loop in Blender's own Python process, executes the model's generated code against the live scene, and wraps every conversational turn in a single undo step.

  • A ComfyUI add-on that patches ComfyUI-GGUF so Qwen-Image-2.1 models can run fully as GGUF files, fixing a text-encoder mismatch and missing architecture detection until upstream support arrives.

  • A ComfyUI node that expands a short prompt into a full styled prompt, negative prompt and aspect ratio for Qwen-Image-2.1, using one of 17 photographic styles.