T8mars/minimax-h3-prompt-skill-T8

Creative DNA video prompt cases and installable Skills for MiniMax H3 and Seedance 2.0, with an Electron video viewer.

This is a curated library of reusable video-generation prompts (called "Creative DNA") for two AI video models: MiniMax H3 and Seedance 2.0. It collects real, working prompt patterns — not tutorials or benchmarks — and packages them three ways: as a browsable catalog, as installable "Skills" for AI agents, and as a desktop app that shows actual video examples with sound.

What it contains

  • catalog/ — A browsable database of cases. Each case explains the creative mechanism, gives prompts for both H3 and Seedance, credits the original author/platform, and links back to the original post. On GitHub, previews are lightweight looping GIFs.
  • skills/ — The same mechanisms, packaged as installable Skills so an AI agent can reuse them directly. Each Skill references the same case and source.
  • catalog/official-skills/ — A fixed index of the 9 Skills from MiniMax's official H3 repository (1 generic prompt-writing skill + 8 style-production skills). The repo does not copy the Skill text; it stores commit hashes, file SHA-256 hashes, install commands, and links. The repo also ships its own original "companion Skills" so Seedance 2.0 can use the same 9 mechanisms. These 9 entries explicitly set comfyui_import=false since the target ComfyUI nodes already cover that capability.
  • catalog/community-skills/ — Two user-contributed, non-official Skills (clearly labeled as such). These were reverse-engineered from user-provided videos and prompt fragments, rebuilt from scratch, and verified against the actual footage — fixing mismatches between text and what the video actually shows.
  • Electron desktop viewer — A downloadable app with real video clips (including sound and offline playback), bilingual search, filtering, sorting, favorites, collections, history, side-by-side comparison, and copy options (per-section or full case). It also links to the original posts.

Key distinctions

  • Only reusable creative mechanisms — explicitly not tutorials, performance benchmarks, model comparisons, or workflow posts.
  • GIFs are previews, not the real video — the desktop app carries the actual media.
  • Attribution is preserved — every case retains author, platform, and original post URL.
  • Official index vs. community content are kept separate — clearly marked "official" or "unofficial · user contributed."

Related project

A companion repo — ComfyUI MiniMax H3 Prompt Enhancer T8 — is linked but not detailed here.

License

Dual-licensed: MIT + CC BY 4.0.

Related

  • Project
    apiframe-ai/seedance-2.0-api

    An API implementation and example set for ByteDance's Seedance 2.0, enabling text, image, and multimodal reference-to-video generation with synchronized audio.

  • Project
    gulucaptain/Camera-Transformer-1

    CT-1 is a Vision-Language-Camera model that estimates precise camera trajectories from images and text prompts to enable spatially aware, controllable video generation.

  • Project
    AkshitIreddy/AI-Powered-Video-Tutorial-Generator

    AI‑Powered Video Tutorial Generator is a Windows‑only desktop app that turns a question or topic into a fully editable video lesson. It handles research, script writing, storyboard creation, narration, illustration, presenter animation and final export, using either local diffusion models (ComfyUI/SDXL) or cloud LLM/image providers. The open‑source stack combines a React/Tauri UI, a Rust core, and a Python pipeline, and it ships with bundled runtimes (Python, Node, Chromium, FFmpeg).

  • Project
    jamichss/Stream-DiffVSR

    Stream‑DiffVSR is a diffusion‑based video‑super‑resolution system designed for low‑latency, online use. It runs a 4‑step distilled denoiser, an auto‑regressive temporal guidance module, and a lightweight temporal decoder to upscale 720p video frames in ~0.33 s on an RTX 4090, far faster than prior diffusion VSR methods. The repo provides conda environments, CLI and Gradio UI for inference, optional TensorRT acceleration, Docker support, and full training scripts using REDS and Vimeo‑90K datasets.