EverygenAIEverygenAI
Back to blog

MiniMax H3 vs Veo 3.1: Which AI Video Model for Ecommerce Ads?

MiniMax H3 vs Veo 3.1 for ecommerce ads: length, refs, resolution, audio, access, and billing shape—not a film ranking.

Nov 9, 2026Keke

This is not a film ranking.

MiniMax H3 vs Veo 3.1 is an ecommerce selection problem: clip length, how many references you can feed, output resolution, native audio (or a silent SKU), which API or app you can still call, and how each vendor bills. A cinematic bottle that invents a new cap is a failed take—not a “better model.”

This guide maps those published specs to ad jobs—not a quality scorecard. For a longer one-take Veo matchup, see Seedance 2.5 vs Veo 3.1. For the catalog-photo path, see AI product videos.

What each model is

MiniMax H3 is MiniMax’s next-generation video model after Hailuo 01 → Hailuo 02—not a Hailuo 02 rebrand. The research post landed on 31 July 2026; partial open weights followed on 3 August 2026. Hailuo 02 and Hailuo 2.3 remain listed as legacy. The API exposes MiniMax-H3 plus Highspeed, Lite, and Max. Open weights cover H3-Base (text-to-video, first/last frame, omni-reference). Context preprocessing and H3-Regenerate-2K stay API-side, so “partial open weights” is the accurate phrase.

Veo 3.1 is Google DeepMind’s video model, in Standard, Fast, and Lite variants. Official surfaces document native audio, SynthID watermarking, and C2PA Content Credentials. GA developer access sits on the Gemini Enterprise Agent Platform. The Gemini API preview IDs were shut down on 22 October 2026.

Two bets. H3 wants a flexible 4–15s generate, a wide aspect set, and a multimodal reference pack via API, app, or partial open weights. Veo wants a short, high-resolution beat from a few stills, a clear provenance mark, and a Google enterprise path.

MiniMax H3 vs Veo 3.1 at a glance

DimensionMiniMax H3Veo 3.1Duration4–15s (Highspeed/Lite too); Max 5–15s4 / 6 / 8s (1080p, 4K, and ref-image jobs often need 8s)Resolution768P default; 2K via H3 / H3-Regenerate-2K. Highspeed, Lite, Max: 480P/768P, no 2K720p / 1080p / 4K (Lite typically tops out at 1080p)AudioNative stereo, 32 kHzNative audio; Cloud also lists a silent / video-only SKUAspects21:9, 16:9, 4:3, 1:1, 3:4, 9:16 (+ adaptive)16:9 and 9:16 (both sides cover 9:16; H3’s set is wider)ReferencesOmni-ref: ≤9 images + ≤3 videos + ≤3 audio; mixed cap 12. Also first/last frame≤3 ingredient images + first/last frame; scene extension on supported tiersProvenanceNo official SynthID (or C2PA credential) found in H3 docs at writingSynthID + C2PAAccessAPI (MiniMax-H3 and variants), Hailuo/MiniMax app, partial open weightsGemini API preview IDs shut down 22 Oct 2026. Migrate to GA Veo on the Enterprise Agent Platform, or evaluate Gemini Omni Flash separately (≠ Veo)Billing shape**$/output-second** + input media (input video per second; extra images beyond a free quota)$/count (per generated video) by tier × resolution × audioBest ad fit10–15s shorts; many SKU/pack/motion refs; batch work on a per-second meter; API or open deploy≤8s polished hero; 1080p/4K; SynthID / Google enterprise; Fast for volume

No universal winner. Match the model to the shot and to the photos you can legally use.

Which model to pick

Pick Veo 3.1 when the brief is a short, polished interrupt:

  • You need a ≤8 second hook for Reels, TikTok, Shorts, or a muted PDP loop.

  • You want 1080p or 4K. H3’s documented top end is 2K via regeneration, not 4K.

  • Brand or legal wants a clear AI-content mark. SynthID plus C2PA is Veo’s official provenance path.

  • You are already on Google Cloud / Gemini Enterprise and can call the GA model IDs there.

  • Volume hooks: use Veo 3.1 Fast. Confirm whether Lite on your surface still supports reference images (official notes have listed that path as unavailable).

Pick MiniMax H3 when the brief needs time, many SKU refs, or a per-second batch meter:

  • You want a 10–15 second candidate in one pass: hook, product face, a beat of proof, room for a CTA.

  • You have more than three stills (front, 45°, label, lifestyle) and may reuse packaging motion or audio. H3’s omni-ref budget (≤9 images plus video/audio, mixed cap 12) is built for that.

  • You want 9:16 and also square, 4:3, or ultrawide from the same model.

  • You care about API or open-weight control, not only a consumer Google app.

  • Draft on Lite or Highspeed (480P/768P), then spend on H3 + 2K (H3-Regenerate-2K) for finals.

If your stack is Gemini-API-only after 22 October 2026, do not assume Veo 3.1 preview still answers. Google shut down the veo-3.1-*-generate-preview IDs that day and pointed developers at Gemini Omni Flash or GA Veo on the Enterprise Agent Platform. Omni Flash is a different model—evaluate it separately, not as a Veo alias.

A shared ecommerce workflow

The model is the second decision; the product is the first.

  1. Collect multi-angle stills that match the live PDP: hero, 45°, label, and one honest in-use shot. Add a packaging or motion clip only if your model accepts video refs. Drop anything that fights the listing colorway.

  2. Pick the tier for the job. Draft on Lite, Highspeed, or Fast. Reserve standard H3 + 2K, or standard Veo at 1080p/4K, for keepers.

  3. Generate the smallest useful cut first. A short Veo hook or an H3 clip at the length you will actually run beats a maxed-out hero you have not inspected.

  4. QA labels and claims like compliance. Pause on logos, label text, proportions, color, and hands. Do not invent reviews, discounts, or clinical claims. Add captions and the real CTA in your edit.

  5. Export for the placement. Both models cover 9:16 for Meta Reels and TikTok. Crop or regenerate for 1:1 if the feed needs it. Leave safe space for captions and platform UI.

Pitfalls worth avoiding

  • Do not call H3 “Hailuo 02.” Hailuo 02 is a legacy model with a different duration and resolution matrix. H3 is the next generation, with its own API IDs and a partial open-weight release.

  • Do not say the Gemini API Veo 3.1 preview still works. Those preview IDs were shut down on 22 October 2026. A Gemini API key now points you at Gemini Omni Flash—or you need the enterprise platform for GA Veo.

  • Do not force per-second math across vendors. H3 bills in dollars per output second, plus input media. Veo’s published Cloud meter is dollars per generated video (a count), by tier, resolution, and audio. Do not divide one table into the other unit. Check the official pages before you budget.

  • Marketing quality claims are not independent benches. Vendor posts will say a model is ready for ads or brand rendering. Judge a take by freeze-framing it next to the PDP.

A practical next step

You do not need a winner ribbon—you need a clip that still looks like your product.

If you want a product-shaped path—catalog still or public URL, review draft scenes, generate only what you approve—open EveryGenAI. Choose the model on its official surface using the pick rule above. For a longer one-take alternative to Veo, see Seedance 2.5 vs Veo 3.1. For the photo-to-clip workflow, see AI product videos.

FAQ

Is MiniMax H3 the same as Hailuo 02?

No. H3 is the next-generation model after Hailuo 01 → Hailuo 02, announced on 31 July 2026, with partial open weights on 3 August 2026. Hailuo 02 remains a listed legacy model. Plan duration, resolution, and references against the MiniMax-H3 IDs (or Highspeed / Lite / Max), not Hailuo 02.

Can I still use Veo 3.1 on the Gemini API?

Not via the preview IDs. veo-3.1-generate-preview, veo-3.1-fast-generate-preview, and veo-3.1-lite-generate-preview were shut down on 22 October 2026. Google’s documented paths are GA Veo 3.1 on the Gemini Enterprise Agent Platform, or Gemini Omni Flash on the Gemini API. Omni Flash is not Veo—evaluate it separately.

Which model is better for multi-SKU or many references?

MiniMax H3, if “many” means more than three stills or you also want to reuse packaging motion or audio. Official omni-ref limits: up to nine images, three videos, and three audio clips (mixed cap of twelve). Veo 3.1 ingredient guidance is documented at up to about three images, plus first-and-last-frame control. More references only help when they agree with the live listing.

How should I compare price?

Compare billing shape, then read the live tables. H3 charges per output second, plus input-media charges (input video per second; images beyond a free quota). Veo 3.1 on Google Cloud is priced per generated video (count) by tier × resolution × audio (including a silent / video-only SKU). Do not convert one vendor’s unit into the other’s. Check MiniMax’s pricing overview and Google Cloud’s Agent Platform / Veo pricing.

Does either model watermark the video?

Veo 3.1: yes, officially—SynthID on generated video, plus C2PA Content Credentials on the enterprise model card. MiniMax H3: we did not find an official SynthID or C2PA credential documented for H3 at writing. Check current MiniMax docs and the identifier your legal team requires before paid media.

Sources