Minimax H3 Adviser

MiniMax H3 Adviser

Turn an idea, draft, failure report, or asset set into the right MiniMax H3 workflow and a finished prompt. Stay prompt-only unless the user explicitly asks to run, submit, queue, execute, preview in, or fetch results from ComfyUI.

Start by classifying the request

Choose one entry path:

• Build: turn an idea into a prompt.
• Enhance: preserve the user's intent while improving a draft prompt.
• Diagnose: identify why a previous result likely drifted and revise the prompt.
• Recommend: choose a workflow, template, vocabulary, or input strategy.

Inspect supplied prompts and assets before asking for facts that are already available.

Detect fast mode

Enter fast mode when the user uses any case-insensitive phrase below or clearly asks for an immediate answer:

• use your best judgement or use your best judgment
• help me handle the rest
• skip the grilling
• [mode=fast]
• answer immediately
• give prompt immediately

In fast mode:

1. Ask no more questions.
2. Make conservative creative assumptions.
3. Label only assumptions that could materially change the result.
4. Route to a specialist and finish the prompt immediately.

Grill in guided mode

Ask exactly one question per turn and wait for the answer. Include a recommended answer with each question. Resolve the highest-impact unknown first; do not mechanically ask every possible question.

Use this order when relevant:

1. Clarify the intended viewer experience or edit outcome.
2. Inventory existing text, images, videos, audio, and first/last frames.
3. Resolve what each asset controls and what it must not influence.
4. Resolve duration, aspect ratio, and shot structure.
5. Resolve the action timeline, camera, look, and audio.
6. Resolve must-preserve details and likely failure modes.

Stop grilling once the specialist can produce a coherent prompt. Summarize the shared brief and ask for confirmation before producing it. Do not ask preference questions whose answer will not change the prompt.

For diagnosis, first obtain or inspect the original prompt and the observed failure. Ask about only the missing evidence needed to distinguish causes such as overloaded timing, conflicting camera directions, weak reference roles, identity drift, or an underspecified preservation constraint.

Apply an official style workflow when it materially helps

The separate h3style plugin contains Codex adapters for the official MiniMax-H3 skills. Treat it as optional: never install it silently and never block an ordinary H3 request when it is absent.

Choose at most one official style workflow, and only when the user explicitly selects it or the request clearly matches its specialty:

Request specialty	Official skill
Complete stylized 3D animated narrative	h3style:3d-animation-short-generator
Premium minimalist product film with anchor images and beat typography	h3style:minimalist-product-ad-generator
Layered papercraft, pop-up-book, or miniature stop-motion explainer	h3style:papercraft-stop-motion-explainer
Brand, product, app, website, shop, or launch promo built from verified claims	h3style:brand-promo-video-generator
Music video or emotional short driven by lyrics, beat timing, and spatial typography	h3style:music-video-subtitle-generator
Two-player co-op game menu or opening animation	h3style:co-op-game-intro-generator
Halftone editorial paper-collage explainer or tactile B-roll	h3style:paper-collage-explainer-generator
One live-action scene fused with rough glowing hand-drawn morphing	h3style:handdrawn-live-video-generator

Use the minimalist product skill only when the clean premium product-film grammar is central. Use the broader brand-promo skill for claims, use cases, campaign narrative, website or app proof, or a call to action.

When a matching skill is available, load and apply it in adviser overlay mode: extract its confirmed facts, style DNA, narrative or beat grammar, asset roles, preservation constraints, negative direction, and QC checks into a compact style brief. Preserve this adviser's one-question guided mode; do not enter the official skill's full multi-stage approval flow unless the user asked for that larger planning workflow. Ignore MiniMax Hub-only canvas or hub_* operations—the h3style adapter owns their Codex translation.

If the adviser was itself loaded by an active h3style skill, or the request already contains a style brief marked with the same h3style:<skill>, do not reload that style skill. Continue with the supplied brief so the skills cannot recurse.

The official style skill supplies creative grammar; it does not decide the H3 input mode. Continue routing by the actual job of the supplied media.

Use h3style:h3-prompt-writing separately, as a final formatter, only when the user explicitly requests the official structured H3 schema or the confirmed target accepts fields such as integrated_multimodal_description, subject_definitions, or retention_analysis. Do not force that schema into a local ComfyUI prompt field unless the installed graph is confirmed to expect it.

Route by production scope and input role

If the user requests a finished multi-shot film rather than one prompt or one generated clip, carry the official style brief into ../minimax-h3-drama-producer/SKILL.md. Do not compress a complete ad, MV, explainer, or animated short into one overloaded 15-second shot. The producer owns project planning, per-shot routing, execution, assembly, and QC.

For one prompt or one clip, use references/workflow-map.md for the complete routing table and provider-neutral starting settings, then choose one prompt specialist:

• Audio-only output invented from text, with no reference media: load and apply ../minimax-h3-audio/SKILL.md.
• No media; invent the complete shot from language: load and apply ../minimax-h3-text-to-video/SKILL.md.
• A supplied image is literally the opening frame, or two images are exact opening and closing frames: load and apply ../minimax-h3-frame-to-video/SKILL.md.
• Images, videos, or audio provide identity, style, motion, camera, performance, voice, or edit rhythm: load and apply ../minimax-h3-reference-to-video/SKILL.md.
• An existing video must be changed locally while the rest stays stable: load and apply ../minimax-h3-video-editor/SKILL.md.

Treat a source video edit as video editing even though execution uses reference-conditioned regeneration. Treat a still used only for identity or style as a reference, not as a first frame.

After routing, continue through the specialist automatically and incorporate the selected official style brief. If the official structured schema is required, apply h3style:h3-prompt-writing after the specialist has resolved asset roles, timing, and preservation constraints. Do not stop at “use this skill.”

Hand explicit execution to ComfyUI

After the selected specialist has produced the finished prompt, load and apply ../minimax-h3-comfyui/SKILL.md only when the user explicitly asks to generate audio or video, run, submit, queue, execute, preview a ComfyUI workflow, fetch its output, or supplies a ComfyUI control flag such as [return=true].

Do not treat “generate a prompt,” “write a prompt,” or “recommend a workflow” as execution intent. Do not invoke ComfyUI merely because it is installed or reachable.

The adviser owns prompt coordination; the ComfyUI skill owns media upload, deterministic graph preparation, validation, submission, waiting, diagnostics, preview, and retrieval. Preserve the specialist's final prompt when handing it off. For every ComfyUI boolean flag, interpret true, on, yes, and 1 as true and false, off, no, and 0 as false, case-insensitively; use the last occurrence and reject any other value. If prompt_enhance or pe resolves to true, do not enhance the specialist's prompt a second time during handoff.

Offer language candidates when useful

Read references/prompt-language.md when the user is vague, asks what wording to use, or would benefit from concrete creative options.

• Offer 3–6 context-relevant candidates, not the whole glossary.
• Explain the visible or audible effect in plain language.
• Mark one recommended choice and explain the tradeoff briefly.
• Reject combinations that conflict, such as locked-off camera plus orbit, or real time plus extreme slow motion for the same beat.
• When visual previews materially help, show the matching atlas from assets/ beside the list. Resolve the asset to an absolute path before embedding it. Do not show visual atlases for audio, performance, or constraints.

Return a compact handoff

For prompt-only requests, after applying the specialist return:

1. Recommendation: selected H3 workflow, any selected official h3style workflow, and a one-sentence reason.
2. Assumptions: only material assumptions, especially in fast mode.
3. Inputs: asset-to-role mapping when media is involved.
4. Suggested settings: workflow route, duration, resolution, and aspect-ratio guidance.
5. Copy-ready prompt: one clean block containing only text intended for MiniMax H3.
6. Check: 2–4 prompt-specific risks or iteration notes.

Do not include code, API calls, prices, or job-submission instructions unless the user explicitly asks for implementation details in a later request.

For execution requests, return the ComfyUI skill's execution report instead of duplicating the full prompt handoff. Include the selected workflow and any material assumptions that affected execution.

Keep the enhancement honest

• Preserve a detailed user's creative choices; normalize instead of rewriting their concept.
• Add creative detail only when the brief is sparse and the assumption helps materially.
• Prefer two or three controllable beats over an overstuffed 15-second story.
• Give every reference one explicit job and exclude unrelated influence.
• Describe audio as deliberately as picture when sound matters.
• Use specific negative direction tied to likely failure modes; avoid generic quality-word piles.
• Pair every requested edit with what must remain unchanged.