MiniMax H3 skills for product ads and brand promos

The MiniMax H3 repository includes two video skills that can both begin with product assets but organize the work very differently. The minimalist-product-ad-generator is a narrow physical-product workflow: lock the product, choose a restrained visual treatment, and build a 5-, 10-, or 15-second film around a few precise beats. The brand-promo-video-generator is a broader campaign workflow: verify brand truth, track asset provenance, show product evidence, and end with an approved call to action.

Choosing between these MiniMax H3 skills is therefore less about visual taste than production scope. A single product hero with controlled typography belongs in the minimalist workflow. A launch reel involving a logo, website, app screens, feature claims, narration, or several kinds of official media usually belongs in the brand-promo workflow.

Choose the skill by the real deliverable

The shortest test is: are you selling one physical object through image, motion, material, and minimal copy, or explaining a wider brand or product story with evidence and a CTA?

There is an overlap zone: both skills can direct a short film for a physical product. Use the minimalist skill when product fidelity, negative space, material detail, variant timing, and sparse typography are the whole idea. Use the brand skill when the same product must sit inside a larger campaign system with logos, claims, usage scenes, official photography, voiceover, disclaimers, or a publication-ready source trail.

Stage 1: run the right intake

The minimalist skill begins with a compact start gate. It asks for product materials, identifies the main image, confirms whether there is one variant or several, and locks duration, aspect ratio, one of four visual templates, and the copy source. Its listed templates are white-tech, dark rim light, brand color field, and light lifestyle. MiniMax H3 is the skill's default video-model strategy, while another model is considered only when the user explicitly asks or the intended structure needs a fallback.

The brand-promo intake is wider because the deliverable can carry more reputational risk. It asks for original logo and font sources, brand colors, product images or UI screenshots, official product information, claims, CTA, disclaimers, target audience, and an official URL or source package. It also resolves campaign focus, distribution channel, narration language, on-screen language, visible copy, duration, and aspect ratio before creative direction.

Do not merge the two intake forms into one giant questionnaire. For a one-product film, asking for a newsroom, media kit, and corporate narrative may slow the work without improving the image. For a brand launch, accepting one screenshot and a slogan without checking their source can leave every later shot built on an unverified claim.

Stage 2: establish truth and provenance

Both workflows need truth, but they record it at different depths.

In the minimalist workflow, inspect the supplied product material and write a product fact summary: category, main color, likely main variant, image quality, visible materials, and showable structures such as a lid, dial, fold, screen, clasp, or texture. If the source is too soft, badly lit, or too incomplete to preserve the product, the skill says to stop and give reshoot guidance. Apple-like restraint is a treatment for background, lighting, composition, and motion; it is not permission to recolor the product.

In the brand workflow, build a brand truth sheet from the strongest available sources. The documented source order favors user-provided originals, official company sites and repositories, company-controlled libraries, and licensed material. Then create a provenance manifest with a stable asset ID, role, path or source URL, source type, verification target, rights note, and authenticity status. Search thumbnails, logo aggregation sites, fan redraws, and plausible AI substitutes are not accepted as identity sources.

That distinction matters in practice. A product fact summary answers, “What must remain visually true in this object?” A provenance manifest also answers, “Where did this logo, screenshot, claim, or photograph come from, and what is its publication status?” A manifest helps review rights and authenticity; it does not grant publication rights by itself.

Stage 3: choose a product-specific narrative

The minimalist skill offers three compact spines:

1. Product Launch: reveal the main variant, show material or structure, introduce an action or variant relationship, and close on the product and copy.
2. Feature Touch: begin with stillness, trigger a visible interaction, magnify the useful detail, show the result, and settle on a closing frame.
3. Color Family: establish one main variant first, bring supporting colors in as layers, and close with an ordered set rather than a product wall.

The brand skill selects its spine by category. An AI or SaaS promo can move from user intent through planning and capabilities to a useful output and proof. A physical product can move from hero reveal through interaction and usage to result and logo. A service can show context, process, evidence, outcome, promise, and logo. The important rule is the same: actual product behavior owns the story; abstract effects cannot substitute for evidence.

In either workflow, define the motion language before generation. Give each beat one primary action, use product edges, cursor paths, UI flow, material highlights, or matched geometry to motivate transitions, and place quieter holds around copy or proof. Avoid random particles, fake interfaces, ornamental metric walls, or motion that has no relationship to the subject.

Stage 4: lock assets before motion

This is the biggest procedural difference between the two skills.

The minimalist skill calls for three independent product anchor photos, all at the chosen output ratio:

• a hero view that locks silhouette and the opening direction;
• a material or functional detail that locks finish and the key visible mechanism;
• a final-copy composition that locks product placement, type feeling, one-line layout, and the two-part color treatment.

They must be separate, complete product photographs—not a three-panel sheet, storyboard board, split screen, or collage. The skill warns that a video model may reproduce a panel layout in the final result. It also instructs the production step to use these approved anchors, rather than the raw product image, as its video references. The original remains the source for analysis and anchor creation; the anchors become the controlled visual packet.

The brand-promo skill uses a different lock: carry authenticated asset IDs and their allowed roles into the shot plan. A logo, wordmark, UI, package, mascot, person, or brand scene should come from a verified or user-authorized source, not from a generated approximation. Image generation may create abstract atmosphere, transitions, non-representational graphics, or conceptual plates, but it should not fabricate official-looking product evidence.

Stage 5: turn the narrative into an exact storyboard

The minimalist storyboard uses seconds and usually contains 3–4 beats for 5 seconds, 5–7 beats for 10 seconds, or 6–9 beats for 15 seconds. Each row records the shot purpose, visual lead, product and camera motion, variant state, copy, text color, text effect, continuity, and rhythm intent. The final beat must be a stable full-frame product closing with a single-line copy treatment, not a grid or montage.

Its typography rules are unusually specific. Visible advertising copy is English, normally 3–5 words and no more than 32 characters, with only one single line visible at a time. In the white-tech treatment, the first part is black or dark gray and the second uses the actual product color; dark rim light may use white for the first part. Copy should enter as part of the composition, not sit in a subtitle position. If generated text proves unreliable, reduce the number of text beats before crowding every shot.

The brand storyboard is more campaign-oriented. At the skill's 30fps planning convention, a 15-second promo targets 5–8 major beats and a 30-second promo targets 8–12. Each beat records its time or frame range, visual owner, authentic asset IDs, primary action, product proof, copy hold, color state, transitions, and motion intent. Before generation, the user reviews the full pre-production package: source summary, truth sheet, story, shot plan, copy, CTA, narration, audio plan, and known rights or placeholder caveats.

Stage 6: generate only what the runtime supports

After approval, the minimalist instructions ask for one continuous full-frame product film controlled by the three anchors and the complete beat table. Every intended line of in-frame copy, its timing, its color split, and its same-line motion must appear literally in the generation brief. The anchors preserve identity, color, material, and composition; the storyboard controls shot order, action, typography, and continuity.

The brand instructions ask the runtime to prepare verified stills or motion-ready plates, generate clips with the same aspect ratio and authentic references, and assemble the approved sequence. Each dispatch should restate the model, ratio, source paths, and original request. Generated atmosphere can support the official media, but cannot redraw the logo, UI, packaging, person, or mascot.

These are workflow requirements, not universal runtime guarantees. The brand skill declares a set of expected source, image, video, audio, editing, media-inspection, and canvas tools. If the host does not expose a required generator or editor, the skill explicitly says to downgrade the deliverable to a pre-production package. A useful handoff can still include the truth sheet, manifest, approved anchors, storyboard, generation briefs, audio brief, and verification checklist—just do not label that package a finished film.

For the mechanics of assigning roles to image, video, and audio references, use the MiniMax H3 prompt guide. This article stays focused on production workflow rather than repeating H3 modes and prompt syntax.

Stage 7: design audio and the edit

Both skills favor H3 native audio for the first complete pass, but their fallback logic differs.

The minimalist workflow places an instrumental tech direction in the H3 video prompt so picture rhythm and sound can be created together. If that music is unpleasant, weak, too loud, out of sync, or the user requests a replacement, the skill calls for a longer standalone music-2.6 track. The editor then analyzes its rhythm and energy, selects a continuous 8–12 second segment, mutes the original video track, replaces it with that segment, and checks for silence or audible dropouts.

The brand workflow also defaults to video-native audio when the user asks for a finished reel without special audio requirements. Separate speech or music is reserved for controllable narration, dialogue, replaceable BGM, exact independent duration, remixing, or a video model without suitable audio. This avoids accidentally stacking native music and a second soundtrack. Subtitles are added only when the user explicitly requests them.

For a deeper native-audio workflow, see MiniMax H3 native audio: plan sound with the shot.

Stage 8: verify against the approved package

Before delivering either film, compare the render with decisions made before generation—not with a vague impression that it “looks premium.”

For the minimalist product ad, check:

• ratio, duration, language, product body color, main variant, and variant timing;
• stable product geometry and material across the hero and detail beats;
• one readable copy line at a time, with no accidental symbols or lower-third subtitle treatment;
• a stable product-and-copy final frame;
• balanced music, no long silence, and a confirmed beat-synced edit when replacement music was used.

For the brand promo, check:

• every logo and identity-bearing asset against the truth sheet and manifest;
• product names, feature wording, claims, metrics, CTA, and disclaimers against approved sources;
• ratio, duration, language, copy holds, logo clear space, and platform framing;
• motion ownership, so effects do not obscure product proof;
• final output location plus grouped source and generated assets when the runtime has a canvas.

If a brand asset fails authenticity review, replace it with an official or user-authorized source or pause for the original. Improving a convincing imitation is not a valid finishing pass.

Example brief 1: a minimalist physical-product film

Assume a seller supplies three original photos of a coral anodized pocket speaker, including a front view and a clearly visible side dial. No logo animation, performance metric, or audio-quality claim has been approved.

This brief avoids inventing speaker performance. The visible dial supplies the action, the real coral finish supplies the copy color, and the product itself supplies the transition logic.

Example brief 2: a verified SaaS launch promo

Assume a software team supplies original SVG logo exports, a brand guide, approved website copy, three current UI recordings, and a product page confirming the launch features. The campaign must show what the product does and direct viewers to start a team workspace.

This brief belongs in the brand workflow because the proof comes from several official assets and approved claims. Three product-style anchor images would not replace the need to trace the logo, UI, copy, and CTA back to their sources.

Getting started on OmniArt

Start with the decision table, then prepare the smallest truthful source package for the chosen workflow. For a minimalist ad, select one physical product, a main variant, three anchor roles, and a 5-, 10-, or 15-second narrative. For a brand promo, assemble official identity assets, approved claims, product evidence, CTA, and provenance notes before you design motion.

When the package is ready, use OmniArt's video workspace for the supported generation stages and keep image, video, and audio work together for review. The durable lesson from both MiniMax skill files is simple: approve facts and visual ownership first, approve the beat plan second, and generate only after the production contract is explicit.