Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add SlavaSexton/ComfyUI-Agent-Kit --skill minimax-h3git clone --depth 1 https://github.com/SlavaSexton/ComfyUI-Agent-KitWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/slavasexton/comfyui-agent-kit/minimax-h3)<a href="https://agentmods.dev/skills/slavasexton/comfyui-agent-kit/minimax-h3"><img src="https://agentmods.dev/badge/skills/slavasexton/comfyui-agent-kit/minimax-h3/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/slavasexton/comfyui-agent-kit/minimax-h3"><img src="https://agentmods.dev/badge/skills/slavasexton/comfyui-agent-kit/minimax-h3.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector pass
What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00102 | $0.02475 |
| Opus 5 | $0.00051 | $0.01238 |
| Sonnet 5 | $0.00020 | $0.00495 |
| Haiku 4.5 | $0.00010 | $0.00248 |
Grade A, and why
minimax-h3 scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 10d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 161 lines — stays where its author put it; the contents beside it link to each section on GitHub.
MiniMax H3 (Hailuo 3)
H3 generates video and synchronised stereo audio jointly, from text, images, reference video, reference audio, or any mix. It is the same model family whether you hit the hosted API or run the open weights, but the two paths are wired completely differently and only one of them is free.
Two paths, do not confuse them.
- Hosted API, partner nodes
MinimaxHailuo03TextToVideoNode/...FirstLastFrameNode/...ReferenceNode, categorypartner/video/MiniMax. 2K output, priced per second, no local weights. - Local open weights, core nodes
MiniMaxH3ImageToVideoandMiniMaxH3ReferenceToVideofromcomfy_extras/nodes_minimax_h3.py. 768p, free, and everything below is about this path.
Who owns what, so you open one file, not three. THIS file owns the prompt format and the operating rules.
reference.md next to it owns weights, quant sizes, acceleration packs and their wiring. The MiniMax entry in
the kit's MODELS.md owns the node-level graph (every node and socket) and the licence. When they disagree,
the node code wins and the discrepancy is a bug worth reporting.
The prompt format is not free-form prose
H3-Base consumes the output of a hosted prompt refiner (H3-Context-IR) that is not in the open release.
So locally you write in the refiner's output shape yourself. From MiniMax's own
VIDEO_PROMPT_WRITING_GUIDE_base_en.md, that is an optional instruction line, a blank line, then three fields:
integrated_multimodal_description: [Shot 1] ... [Shot 2] At 00:04.500, ...
overall_soundscape: ...
non_diegetic_music: ...
integrated_multimodal_descriptioncarries visuals, action, shots, speakers, dialogue and diegetic sound along the timeline.overall_soundscapesums ambience and physical-action sound.non_diegetic_musicis score the characters cannot hear.- Image modes need a fixed first line. I2V:
For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.First-and-last-frame uses the alignment sentence naming both pictures and the second each lands on, to two decimals.
What ships with it
1 file beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 10d ago First seen · 161 lines · 102 tokens per session scan A 6f0203b47e89
minimax-h3 is a skill published in the GitHub repository SlavaSexton/ComfyUI-Agent-Kit (94 stars, last pushed 6d ago), licensed Apache-2.0. It adds 102 tokens to every session and 2,475 once invoked, about $0.0005 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
cinematic-frame-director
Turn a user's idea, scene description, or mood into ONE production-grade cinematic still-image prompt in English for text-to-image models (Seedream, Midjourney, Flux, DALL-E, or any image generator). Use this skill whenever the user asks for an image prompt, a cinematic frame, a "movie still" look, a photorealistic…
flux-txt2img
Build Flux txt2img workflows with Flux.1 Dev (SRPO), Flux 2 Klein 9B, Turbo LoRAs, FluxGuidance, and DualCLIPLoader patterns.
prompt-engineering
ComfyUI prompt engineering knowledge covering CLIP text encoding syntax, weight modifiers, model-specific prompting strategies, and best practices.
higgsfield-seedance
Rewrites scene descriptions using professional cinematography language, structures prompts with a six-slot formula (camera + subject + action + setting + style + lighting), and diagnoses content filter rejections via a preflight linter. Use whenever the user asks for a Seedance 2.0 / Seedance Pro prompt, describes a…
higgsfield-prompt
Use when building, writing, refining, or structuring a Higgsfield AI prompt. Covers the MCSLA formula, prompt structure, narrative vs. timestamped formats, and how to write for both text-to-video and image-to-video workflows.
higgsfield-seedance-2-5
Seedance 2.5 prompt director — the omni-reference dialect. Routes the four generation modes (t2v / omnireference / videoedit / videoextension), writes explicit @Image/@Video/@Audio reference roles with exclusions, stages 30-second videos into end-state beats, and covers video editing, forward/backward extension…