JZL - ✨ MiniMax-fl2va提示词预设

MiniMax H3 is the 33B omni-modal video model that generates picture and synced audio together, and it runs on a structured-prompt format rather than a plain sentence. That format is powerful and a pain to write by hand. JZL - ✨ MiniMax-fl2va提示词预设 is the shortcut: you pick from dropdowns, it assembles the whole system prompt for you. No LLM, no API key, no cost per call - the node is pure string assembly, instant and free.

The "fl2va" in the display name is the headline mode - first-and-last-frame to video plus audio - but the node actually covers all four of H3's generation families across eight presets: T2VA (pure text), I2VA (first-frame image), FL2VA (first + last frame), and L2VA (last frame only), each in English and Chinese.

The inputs that actually matter

• 预设模式 - picks the generation family (T2VA / I2VA / FL2VA / L2VA) and output language. Everything downstream is built around this one.
• 视频时长 - 4–15 seconds, H3's supported window. Default 8s.
• 视觉风格 - the fun one, 39 options from Cinematic and Anime to things most presets won't touch: 皮影戏 (shadow puppetry), 青花瓷 (blue-white porcelain), 折纸 (origami), 哥特萝莉 (Gothic Lolita). These aren't single style words. Several are "GLOBAL MATERIAL OVERRIDE" blocks that force the entire world of the clip to become clay, paper, felt, or pixels - characters included - which is the difference between a passing nod to a style and the model actually committing to it.
• 音乐风格 - 14 choices including 禁止音乐 / No Music. The hints are written the way H3's non_diegetic_music field wants them: instrument, tempo, dynamics. Pick Epic Orchestral and it spells out "powerful brass, thundering timpani, swelling strings at a moderate tempo."
• 画面比例 - 8 aspect ratios, each with composition guidance (21:9 pushes for horizontal space; 9:16 stacks vertically).
• 切镜次数 - 0 to 9 cuts, and the node sanity-checks it: ask for 8 cuts in a 4-second clip and it writes a warning into the prompt telling the model to trim.

What comes out and where it goes

One output: system_prompt (STRING). Wire it into the H3 prompt path - the pack's own handwritten-prompt editor or the official MiniMax H3 Text Encode node - and it carries the whole directorial brief: duration, style override, music direction, aspect-ratio composition, and shot timing.

Install

It ships in the JZL MiniMax pack. ComfyUI Manager → search JZL or ComfyUI-JZL-MiniMax-H3, or cd ComfyUI/custom_nodes && git clone https://github.com/wjluoxiao/ComfyUI-JZL-MiniMax-H3 and restart. You need a recent ComfyUI (with comfy_api.latest / the official H3 nodes). No model downloads for this node.

The honest caveats

The labels are Chinese-first with English in brackets, so you'll learn that 电影感 means Cinematic. More important: this node writes prompts for H3, and H3's open weights carry the MiniMax H3 Community License, which excludes the US, EU, UK and South Korea - if you're in an excluded region, the local-weights path isn't licensed for you, and the hosted Hailuo API is the route to keep in mind before you build a whole workflow around this.

If you want a prompt builder for the multi-reference (ref2va) path instead, that's the sibling node - JZL - ✨ MiniMax-ref2va提示词预设.