Hailuo AI Review: MiniMax H3, Video Models, Pricing, Credits, Commercial Rights and Limits

Hailuo AI's most important update is MiniMax H3, released on July 31, 2026. The model changes Hailuo's position from a short visual clip generator into a more general multimodal production system. H3 can jointly understand text, images, video, and audio, then generate a new video with native stereo sound. Official MiniMax documentation supports output at 768P or 2K, with integer durations from 4 through 15 seconds.

That combination matters because much of practical video creation is reference driven. A creative team rarely starts with one sentence and no context. It may have a product photograph, a character design, a director's motion reference, a voice sample, a musical rhythm, a brand palette, and a written brief. H3 is designed to let the user explain how those materials relate. MiniMax's examples describe using one video for camera movement, one image for a character, and one audio file for performance, with the relationship expressed in natural language.

The output can include voice, effects, ambience, and music as native stereo audio. That does not turn H3 into a complete sound department, but it removes an important barrier during ideation. A filmmaker can test whether a visual and audio concept work together before rebuilding the selected idea with licensed music, controlled dialogue, sound design, and a professional mix. For social creators, a usable native soundtrack may sometimes be enough for a draft, provided every element is reviewed.

H3 also broadens duration and resolution. The official API offers 4 through 15 seconds rather than locking every job to one or two clip lengths. Supported ratios include 21:9, 16:9, 4:3, 1:1, 3:4, and 9:16, plus adaptive behavior in reference workflows. This gives creators better control over cinematic, square, and vertical delivery. Two-kilopixel output can preserve more detail for editing and reframing, although resolution alone does not guarantee correct anatomy, stable identity, readable text, or believable physics.

MiniMax presents H3 as a general-purpose model rather than a collection of isolated specialists. Earlier systems separated text-to-video, image-to-video, subject reference, motion reference, first-and-last-frame, audio generation, and editing. H3's design tries to express these relationships through language and shared context. The technical release describes a contextual representation system, a new visual tokenizer, an omni transformer, and an in-context regeneration process for 2K detail. Those are vendor claims, but the product consequence is easy to understand: one model can handle more kinds of creative instruction.

On August 3, MiniMax announced that H3 was open source. This is a meaningful step for research, compatibility, and customization, but the wording needs precision. MiniMax's open-source instructions distinguish the H3 base model from the complete official 2K workflow. A local 768P validation can use the released base. Reproducing the full official 2K result involves proprietary H3 Context-IR and regeneration services through MiniMax APIs. Buyers should not interpret open source as a promise that the entire Hailuo consumer experience or full-quality pipeline can run privately on a local workstation.

H3 arrives alongside a useful portfolio of older Hailuo models. Hailuo 2.3 supports text-to-video and image-to-video at 768P for six or ten seconds and 1080P for six seconds. Hailuo 2.3 Fast focuses on lower-cost image-to-video with the same published duration and resolution combinations. Hailuo 02 adds a 512P option and supports first-and-last-frame workflows. Director models expose explicit camera terms such as pan, tilt, push, pull, truck, pedestal, tracking, zoom, shake, and static shot. Subject-reference tools are intended to improve identity consistency.

That model choice creates a sensible production ladder. A team can use a cheaper Fast generation to explore motion, composition, and timing, then test a shortlisted shot with the full 2.3 model or H3. H3 is the stronger candidate when audio, several reference types, generalized editing, 2K delivery, or a longer clip matters. Hailuo 02 remains useful where a lower-resolution draft or first-and-last-frame control meets the brief. The best model is therefore a cost and control decision, not automatically the newest option.

The separate MiniMax API makes those trade-offs measurable. Current pay-as-you-go documentation lists H3 output at $0.08 per second for 768P and $0.13 per second for 2K. A five-second 2K output therefore has a list output cost of $0.65 before any billable reference video or extra image inputs. Audio references are listed as free, the first five image references are free, and additional images cost $0.04 each. Video reference inputs are billed by duration and output resolution. Regenerating an eligible 768P result to 2K is listed at $0.05 per output second, with input materials billed again.

Older models retain simpler per-video prices. Hailuo 2.3 Fast is listed at $0.19 for a six-second 768P generation, while full Hailuo 2.3 starts at $0.28 for the same output. A six-second 1080P generation is listed at $0.33 for Fast and $0.49 for full 2.3. Hailuo 02 starts at $0.10 for six seconds at 512P. These are API prices, not a direct conversion table for consumer Hailuo credits.

Consumer pricing is the weakest part of Hailuo's current public information. One Hailuo-authored article updated July 27 lists Standard at $9.99 per month, Pro at $34.99, and Master at $79.99. Another Hailuo article updated the same day lists Standard at $6.99, Pro at $23.99, and Master at $53.99, in a pricing section that also refers to annual subscriptions. The published subscription terms, last updated in July 2025, still list older monthly prices of $14.99, $54.99, and $119.99. Ultra and Max are shown at $124.99 and $199.99 in current material.

WhatAI cannot responsibly turn that conflict into one definitive price table. The likely explanation is a mix of monthly, annualized, promotional, or not-yet-synchronized content, but that is an inference. A buyer should sign in, open the actual checkout, select the intended country and billing cycle, and record the price, renewal, tax, credits, model eligibility, queue benefits, and cancellation terms shown there. Hailuo should also reconcile its public pages.

The credit rules are clearer in the subscription terms. Membership credits last one month and do not roll over. Purchased credits have a longer stated life. Free users receive watermarked downloads, while paid plans add watermark-free downloads and priority. The terms describe up to three queued tasks and one parallel generation for free users, compared with five queued tasks and two parallel generations for paid users. Failed or moderated video generations are supposed to receive an automatic credit refund. A technically successful but creatively poor result is not necessarily a failure for refund purposes.

Commercial rights require equally careful reading. Hailuo's subscription terms say the company does not claim ownership of content a user generates and downloads during a paid subscription, and that the paid user retains intellectual-property and commercial-use rights as between the user and Hailuo. The general terms also say Hailuo does not claim ownership, but they grant the company, affiliates, successors, and assigns a broad, perpetual, irrevocable, worldwide license over user contributions and generated content. The user remains responsible for copyright, trademark, publicity, privacy, consent, accuracy, and legality.

The practical lesson is that a paid plan is not legal clearance. A model may reproduce a protected logo, resemble a real person, distort a product claim, create unlicensed-looking music, or generate an image that conflicts with talent and client agreements. Teams need rights-cleared references, releases, a provenance log, final human review, and legal advice for valuable or high-risk campaigns. Political use is also prohibited by the current general terms.

Privacy deserves special attention because image and video workflows often contain identifiable people. Hailuo's December 2025 privacy policy says certain mobile features may temporarily derive facial features or landmarks from uploaded face photos solely for the requested generation. It says those facial data are not retained, used for identification, or shared, and that the uploaded photos and derived data are deleted after processing. That is a helpful specific promise, but it is framed around certain mobile features. Users should not extend it to every web, API, community, template, or reference workflow without confirmation.

A production workflow should assume that generation is one stage, not the finished product. Begin with a shot list and a clear reason for every generated clip. Use image-to-video when composition and product fidelity matter. Use H3 references only when their roles are explicit. Generate several controlled variants, changing one factor at a time. Inspect every frame and listen to every channel. Reject drift, anatomy errors, broken reflections, wrong labels, misleading motion, identity changes, and audio artifacts. Then trim, composite, grade, mix, caption, upscale, and deliver in a conventional editor.

Cost should be measured per usable final second. A $0.28 generation is not cheap if ten attempts and forty minutes of editing produce one usable five-second shot. A more expensive H3 result may be cheaper if multimodal control and native audio reduce rerolls. Conversely, generic office, city, lifestyle, or nature footage may still be faster and legally simpler to license from stock. Hailuo is most valuable when the required scene is specific enough that stock does not exist and a custom shoot would be disproportionate.

MiniMax H3 is a substantial step for Hailuo. It offers a more coherent way to direct image, motion, sound, identity, and style, with stronger output options and a path for developers and open-source researchers. The model still produces probabilities, not guaranteed shots. The teams that benefit most will treat it as a powerful virtual production instrument inside a disciplined human workflow, with clear references, controlled experiments, documented rights, realistic budgets, expert finishing, and final accountability.

WhatAI Decision Box

Hailuo AI has evolved from a short text-to-video and image-to-video tool into a broader multimodal creative studio. MiniMax H3 can accept text, image, video, and audio context, then generate video with native stereo sound at 768P or 2K for up to 15 seconds. Older Hailuo models remain useful for faster or lower-cost generation, 1080P six-second clips, ten-second 768P clips, first-and-last-frame control, subject reference, and explicit camera commands. The studio also includes templates and image-generation options, while developers can use the separate MiniMax API.

Hailuo is strongest for creating highly specific short shots that would be difficult to find in stock libraries: product concepts, stylized B-roll, motion studies, social hooks, reference-driven character shots, and advertising previsualization. H3's combined audio and visual generation can reduce the work needed to synchronize a rough soundtrack. Outputs still need human selection and finishing. Teams should expect multiple attempts, check people and objects frame by frame, verify sound, text, trademarks, identity, cultural context, and product accuracy, then edit, grade, caption, mix, upscale, and document provenance in conventional production tools.

Hailuo offers free queued generation with watermarked downloads and paid tiers with credits, priority, and watermark removal. Its current public pricing is inconsistent. A July 2026 Hailuo article lists Standard at $9.99, Pro at $34.99, and Master at $79.99 per month. Another current Hailuo article lists $6.99, $23.99, and $53.99 in a section referring to annual savings, while the legal subscription page still displays older prices. Ultra and Max are publicly listed at $124.99 and $199.99. Confirm checkout. Hailuo suits creators and teams making short-form visuals, not buyers expecting reliable long-form editing, guaranteed identity continuity, automatic legal clearance, or a one-click finished campaign.

Use Cases

Pricing varies by plan and region — see current pricing.

Plan features change — last updated: 2026-08-15.

Hailuo AI Community Discussions

Explore community discussions. Ask and answer questions on Hailuo AI to grow and learn together.

Hailuo AI Pros & Cons

How to Get Results with Hailuo AI: Step-by-Step Workflow

1. Define the finished shot

   Write the shot purpose, duration, aspect ratio, subject, action, camera, lighting, environment, audio, continuity, brand, legal, accessibility, and delivery requirements before choosing a model.

2. Choose the lowest-risk workflow

   Use text-to-video for open exploration, image-to-video for visual control, first-and-last-frame for transitions, subject reference for identity, and H3 references when motion, audio, style, or multiple assets must be related.

3. Prepare rights-cleared references

   Use only media you own or are licensed and permitted to upload. Record the source, consent, talent release, trademark status, music rights, model, project, and restrictions for every reference.

4. Lock composition before motion

   For important shots, create or select a strong still first. Validate subject, wardrobe, product, text, lighting, background, color, framing, and aspect ratio before spending video credits.

5. Prompt one motion and one camera idea

   Describe the starting state, one primary action, environmental response, camera movement, timing, ending state, and exclusions. Use documented camera commands where helpful and avoid conflicting movements.

6. Generate a controlled batch

   Create several variants with the same brief and reference set. Change one variable at a time, record the model, settings, credits, render time, seed where available, prompt, and result.

7. Inspect every frame and sound

   Check anatomy, identity, product geometry, logos, text, reflections, shadows, collisions, physics, background continuity, lips, voice, music, effects, synchronization, artifacts, and prohibited or misleading content.

8. Assemble in a real editor

   Trim, retime, stabilize, composite, grade, upscale, mix, caption, add licensed music or voice, and combine short outputs in Premiere Pro, DaVinci Resolve, Final Cut Pro, After Effects, or an equivalent editor.

9. Run brand and legal review

   Confirm talent and reference consent, commercial plan status, IP and trademark risks, disclosure requirements, platform rules, claims accuracy, cultural context, provenance, privacy, and client approval before publishing.

10. Measure total cost and archive evidence

    Calculate subscription, credits, API, failed attempts, editor time, upscaling, music, voice, storage, review, and revision cost. Archive prompts, references, terms, receipts, model versions, source files, approvals, and final exports.

Hailuo AI Gotchas and Limits to Know Before You Start

• Hailuo's current public pricing is internally inconsistent. Verify the live checkout rather than relying on one article or the older subscription terms.
• Membership credits expire after one month and do not roll over under the published subscription terms.
• Free downloads contain a watermark, while the current subscription terms associate watermark-free downloads with paid plans.
• Unlimited or Relax Mode wording applies only to specified legacy Hailuo models and should not be read as unlimited H3 generation.
• Consumer subscription credits and MiniMax API pay-as-you-go billing are separate products with different meters and terms.
• H3 is still a short-video model, with official API output limited to 4 through 15 seconds.
• Hailuo 2.3 and related models have duration and resolution combinations that vary by model and mode.
• Native stereo audio can still contain wrong speech, timing, accents, music, effects, or identity and requires listening review.
• Human anatomy, hands, faces, clothing, products, typography, logos, collisions, and fast motion can drift or deform across frames.
• Subject-reference tools improve consistency but do not guarantee the same person or object across every shot, angle, emotion, and lighting condition.
• Image-to-video inherits defects and ambiguity from the source still, so a weak reference can produce a weak or unstable clip.
• A successful generation is not necessarily a usable shot. Budget several attempts and editor time for each final clip.
• The paid subscription terms state commercial-use rights, but users remain responsible for copyright, trademark, publicity, privacy, consent, and other third-party rights.
• The general terms grant Hailuo and related parties a broad, perpetual, irrevocable worldwide license over user contributions and generated content.
• Public or community posting can expose user contributions. Do not assume a shared generation is private.
• The privacy policy's deletion promise for face photos and derived facial data is specifically described for certain mobile-app features, not as a universal retention rule for every upload and workflow.
• The service is intended for users aged 16 or older under the current terms and privacy policy.
• Content can be blocked, removed, or refused by moderation, and prohibited generations must not be used.
• Hailuo and MiniMax can change models, prices, queues, terms, allowances, and availability, so archive the conditions that applied to a client project.
• Open-source H3 does not mean the complete official 2K pipeline is fully local. MiniMax describes a hybrid workflow using proprietary Context-IR and regeneration APIs for official-quality 2K output.

Which Hailuo AI Feature Fits Your Use Case

Starter Prompts for Hailuo AI

What is Hailuo AI?

Hailuo AI is MiniMax's consumer platform for generating short videos and images from text and reference media. Its current flagship, MiniMax H3, combines text, image, video, and audio context and can generate 4 to 15 second video with native stereo sound at 768P or 2K.

Is Hailuo AI free?

Yes. Hailuo provides a free service with basic queued generation and watermarked video downloads. Its subscription terms say free users can queue up to three tasks and generate one video in parallel. New users may receive a one-time welcome credit package that expires after three days. Live allowances can change.

How much does Hailuo AI cost?

Hailuo's own current pages conflict. One July 2026 article lists monthly Standard, Pro, and Master prices of $9.99, $34.99, and $79.99. Another lists $6.99, $23.99, and $53.99 in a section referring to annual savings. The legal subscription page still shows older higher prices. Ultra and Max are listed at $124.99 and $199.99. Confirm checkout.

What is MiniMax H3?

MiniMax H3 is a general-purpose multimodal generation model launched on July 31, 2026. It accepts text, images, video, and audio as context and generates video with native stereo audio. Official API documentation supports 768P or 2K output from 4 through 15 seconds. MiniMax open-sourced H3's base model on August 3.

What is the difference between H3 and Hailuo 2.3?

H3 unifies multimodal reference and editing tasks, supports native stereo sound, 2K output, and variable 4 to 15 second duration. Hailuo 2.3 is a focused text-to-video and image-to-video model supporting 768P at six or ten seconds and 1080P at six seconds. Hailuo 2.3 Fast reduces the cost of image-to-video generation.

Can Hailuo AI generate sound with video?

Yes. MiniMax says H3 generates native stereo audio together with video and can use audio references alongside text, images, and video. Users should still check spoken words, timing, music, effects, voice identity, rights, loudness, and synchronization before publication.

How long are Hailuo AI videos?

H3 supports integer durations from 4 through 15 seconds in the MiniMax API. Hailuo 2.3 and 2.3 Fast support six or ten seconds at 768P and six seconds at 1080P. Hailuo 02 also supports six or ten seconds at 512P or 768P and six seconds at 1080P. Studio options may differ.

Can I use Hailuo AI videos commercially?

Hailuo's subscription terms say the company does not claim ownership of content generated and downloaded during a paid subscription and that paid users retain commercial-use rights. The general terms also grant Hailuo and related parties a broad, perpetual license over user contributions and generated content. Review both documents and obtain legal advice for important campaigns.

Do Hailuo credits roll over?

The published subscription terms say membership credits are valid for one month and do not roll over. Purchased credits have a longer stated validity, extending to December 31 of the second calendar year after purchase. Promotional credits can have shorter expiry, including three days for the described new-user package.

What happens when a Hailuo generation fails?

The subscription terms state that credits are automatically refunded when a video generation fails or does not pass review. This does not mean credits are refunded merely because a successful output has artifacts, poor composition, inaccurate motion, or otherwise fails the user's subjective quality standard.

How does Hailuo handle uploaded face photos?

Hailuo's privacy policy says certain mobile features may temporarily derive facial features or landmarks from uploaded face photos solely to generate the requested video. It says that face data is not used for identification, stored, retained, or shared, and that the uploaded photos and derived facial data are deleted after processing.

Does Hailuo AI have an API?

Yes. MiniMax provides asynchronous APIs for H3 and earlier Hailuo models. Current pay-as-you-go pricing lists H3 at $0.08 per output second for 768P and $0.13 per second for 2K, with additional charges for some reference materials. Consumer Hailuo subscriptions and API billing are separate.

Related Design & Creative Tools

Explore the Network