MiniMax H3 AI Video Generator
Make video from a written prompt or a still image, and every run comes back at 2K because that is the only resolution the model offers. Billing is per second of output, and the credit estimate updates before you commit.
There is no quality ladder to choose from — every run returns 2K.
Longer clips than most models here, priced by the second.
Start from a prompt, or hand it a still and let the model move it.
MiniMax H3 AI Video Generator
What is MiniMax H3?
It is MiniMax's video model, also sold under the Hailuo name. You give it a prompt or a photograph and it returns a moving shot at 2K. It is the successor to the Hailuo line rather than a variant of it.
On Kavel the model runs through the Poyo runtime. The thing that separates H3 from every other video model on this site is what it does not ask you: there is no resolution to pick, because the API accepts one value and returns 2K every time. That removes the most common way of accidentally overspending — choosing a tier you did not need — and replaces it with a single decision, which is how long the clip should be. A failed generation is refunded automatically, unless it broke the content policy.
Creative engine
How a run is billed
Per second of output at a single flat rate. No tier multiplier, no audio surcharge — the estimate moves only when the length does.
Resolution is not a choice
Most models here sell 720p, 1080p and 4K at different per-second rates. This one has a single output, so a draft and a final cut cost the same per second and differ only in length.
Length is the only lever
Because billing is per second and the resolution is fixed, duration is the entire cost equation. A ten-second clip costs twice a five-second one, and there is nothing else to trade against it.
Longer maximum than most
The model accepts up to fifteen seconds in a single run, where several models on this site stop at eight or ten. The generator here offers 5, 6, 8 and 10 seconds.
Two ways in
Text-to-video and image-to-video both run on the same model here. Starting from a still is the more predictable of the two, because the opening frame is already settled before the model begins.
For a tier ladder including 4K, Kling 3.0 lets you pick the output; for a much cheaper draft pass, Seedance 1.5 Pro runs an order of magnitude lower per second.
Shots made on Kavel
Clips generated on this site, shown as a guide to the kind of shot a written prompt produces. They were not all made with this particular model, and every output is AI-generated, so the same prompt will not return the same frames twice.
A product in motion
When this model is the right pick
The reason to choose it is usually that you do not want to think about output tiers at all.
You want one price to reason about
A single per-second rate means estimating a session costs you one multiplication rather than a table lookup.
You need more than ten seconds
The model runs to fifteen seconds, which several models on this site cannot reach in a single generation.
You are animating a still
Image-to-video keeps the first frame you already approved, which removes most of the variance from the result.
What the model does
What it accepts and what it returns, taken from the provider's API documentation and this site's runtime configuration.
Text-to-video
Describe subject, motion and camera in one prompt and get a moving shot back at 2K.
Image-to-video
Supply a still and the model animates it instead of inventing a new scene around it.
First and last frame
The image mode accepts a starting frame and an optional ending frame, so a run can be pointed at where it should finish.
Aspect ratios
Text runs pick from 21:9, 16:9, 4:3, 1:1, 3:4 and 9:16. Image runs follow the aspect ratio of the picture you supply.
Prompts up to 2,000 characters
Enough room to name the subject, the camera move and what changes across the clip without compressing it into one line.
Cost shown per run
The credit estimate reflects the exact length you picked, and length is the only input that changes it.
Specifications
From the provider's API documentation and this site's runtime config. Anything not listed here is not documented on our side, and this page does not guess at it.
- Provider runtime
- Poyo
- Output resolution
- 2K, fixed (no other value accepted)
- Input modes here
- Text-to-video, image-to-video
- Duration range
- 5–15s supported; 5s, 6s, 8s, 10s in the generator
- Prompt length
- Up to 2,000 characters
- Aspect ratios
- 21:9, 16:9, 4:3, 1:1, 3:4, 9:16 (text mode)
- Billing
- Per second of output, single flat rate
- Generated audio
- Not documented
- Frame rate
- Not documented
Generating with it
Four steps, and because resolution is fixed, only one of them moves the price.
Pick text or image to video
Starting from a still removes most of the variance, because the opening frame is already settled.
Choose the length
This is the whole cost decision here. Five seconds to judge an idea, longer once the shot is settled.
Set the aspect ratio
Text runs need one chosen up front. Image runs inherit it from the picture, so there is nothing to set.
Write the motion, then generate
Name the camera move and what changes across the clip. The estimate updates before you commit.
What people make with it
Four jobs where a fixed 2K output and a longer maximum are the reasons this model gets picked.
Social clips at one quality
No tier decision per post — every run comes back at the same resolution, so a batch looks consistent.
Product shots in motion
Animate an approved still so the framing and the product stay exactly as they were signed off.
Longer single takes
Fifteen seconds in one generation, rather than stitching two shorter clips and hiding the seam.
Concept passes before a shoot
A five-second run is cheap enough to explore several directions in one sitting.
Kavel pricing — pay only for what you generate
One credit pool covers Nano Banana images and Seedance and Veo video. The cost shows before every run, and failed jobs are not charged. Use a subscription for ongoing work, or a one-time pack when you just need to top up.
Starter
What's included
- 33,500 credits per year
- Works with Nano Banana images and Seedance and Veo video
- Cost preview before you generate
- Generation history
Secure checkout by Stripe. Card details never touch our servers.
Creator
What's included
- 89,700 credits per year
- Access to Seedance 2.0, Fast, and Mini
- Nano Banana 2 generation and editing
- Priority credit top-ups
Secure checkout by Stripe. Card details never touch our servers.
Studio
What's included
- 166,800 credits per year
- Built for high-volume Kavel generation
- Plenty of room for 4K rendering
- Available across the supported models
Secure checkout by Stripe. Card details never touch our servers.
Credit packs
One-time top-ups — buy extra credits any time you run low.
Credit Pack 1500
What's included
- 1,500 credits
- No ongoing subscription
- Available across the supported models
Secure checkout by Stripe. Card details never touch our servers.
Credit Pack 3600
What's included
- 3,600 credits
- No ongoing subscription
- Available across the supported models
- Use for both video and image generation
Secure checkout by Stripe. Card details never touch our servers.
Credit Pack 9000
What's included
- 9,000 credits
- No ongoing subscription
- Handy for Seedance 2.0 final renders
- Valid for 365 days
Secure checkout by Stripe. Card details never touch our servers.
Charges appear as “KAVEL AI” on your card statement. You can cancel any time from Settings → Billing; cancellation takes effect at the end of the period you already paid for.
Powered by
One credit balance for every Kavel image and video model
Use one balance across the supported image and video models (Nano Banana 2 / Nano Banana Pro / Seedance 2 / Veo 3.1), and check the credit cost before each request.
Video models
Image models
Model credit guide
What each model costs in credits
You're only charged for successful generations. The exact estimate in the generator varies by model, length, resolution, audio, and number of images.
| Type | Model | Credit cost |
|---|---|---|
| Video | Seedance 2.0 | 5s 720p image-to-video ≈ 188 credits; text-to-video ≈ 308 credits. Scales with resolution and length. |
| Video | Seedance 2 Fast | Faster and lower cost. 5s 720p text-to-video ≈ 248 credits. Mini is cheaper still. |
| Video | Veo 3.1 | Billed per video (Lite). About 34 credits at 1080p, about 23 at 720p. |
| Image | Nano Banana 2 | Generate or edit from text and images. About 8 credits per 1K image, 12 at 2K, 18 at 4K. |
| Image | Nano Banana Pro | Consistent run times across generations. About 12 credits per 1K or 2K image, 21 at 4K. |
| Image | Nano Banana 2 Lite | Faster, simpler variant. About 8 credits per image. Start here to test. |
Questions about this model
What it costs, what it accepts and where it stops.
It is billed per second at a single flat rate, so length is the only thing that moves the number. A five-second clip is about 158 credits on this site and a ten-second clip about 315. The generator shows the estimate before you start, and a failed run is refunded automatically unless it broke the content policy.
No, and that is a property of the model rather than a limit we imposed. The API accepts one resolution value and returns 2K, so there is no tier selector on this page. If you need to pick an output tier — including 4K — Kling 3.0 sells three of them.
Yes. MiniMax ships the model under both names, and the provider we run it through lists it as Hailuo 03. Anything written about Hailuo 3.0 describes the same model you are generating with here.
The model accepts whole-second durations from five to fifteen seconds. The generator on this site offers 5, 6, 8 and 10 seconds; four seconds is not available, because five is the model's floor rather than a rounding choice.
Yes. Image-to-video takes a still and animates it, and it also accepts an optional ending frame if you want to say where the shot should arrive. Aspect ratio is inherited from the image in this mode, so there is nothing to choose. Switch to the Image to video tab in the generator.
We cannot source an answer to that, so this page does not claim one either way. The provider documentation describes audio you can supply as a reference input, which is a different thing from audio the model produces, and we will not read one as the other.
Frame rate and generated audio are not confirmed on the provider side, so the specs table says so rather than filling the gap. Anything else absent from that table is absent because we could not source it.
Commercial rights follow your Kavel plan rather than this model in particular, so check the pricing page for what your tier includes. Nothing on this page grants rights the plan does not.
Try MiniMax H3
Write a prompt or upload a still, pick a length, and the credit estimate appears before you generate.
Explore Kavel Models
Kavel guides & tutorials
Phoneme Lip Sync: How Mouths Match Sound, and How AI Does It Now
What phoneme lip sync is, how phonemes collapse into 8-14 visemes, why the classic keyframe workflow still matters, and how audio-driven AI lip sync replaced it for most creators.
Chinese AI Video Generators Free: Kling, Hailuo, Seedance, Wan
The Chinese AI video generators winning the benchmarks, and the free ways to use them without a +86 phone number — access lanes compared.
How to Make a Manga with AI: Script to Panels (2026 Pipeline)
How to make a manga without drawing: script format, character reference sheets, panel prompts, lettering and assembly — the full AI pipeline.