Video model
MiniMax H3 AI video generator
MiniMax H3 is the current generation of the Hailuo video line, running here with no install and no API key. One output tier, three clip lengths, and a price you can work out before you open the form.
MiniMax H3 pricing, length by length
| 4-second clip | 160 credits at 720p. This is the base tier, and nothing is multiplied on top of it. |
|---|---|
| 8-second clip | 320 credits, exactly twice the 4-second price. |
| 12-second clip | 480 credits, exactly three times the 4-second price. Length is the only thing on the form that moves the number. |
| Text-to-video vs image-to-video | Identical. Both directions are 160 credits for a 4-second clip, so starting from a still costs nothing extra. |
| Reference image | Image-to-video takes exactly one still and sends it as the first frame. A second upload is dropped before the request leaves this server. |
| Output options | 720p · 4, 8 or 12 seconds · 16:9, 9:16 or 1:1 · start from a prompt, or from one still you upload |
The honest read on MiniMax H3
What it does well
- Settles the cost before you write anything. 160 credits at 4 seconds, 320 at 8, 480 at 12, and no third variable anywhere on the form.
- Survives a change of mind. Text-to-video and image-to-video bill the same, so moving a shot from a written description to a photo you already have does not move the budget.
- Treats vertical as a real frame. 9:16 and 1:1 are request parameters, not a crop of a 16:9 render.
- Covers a scene in one pass. A single submission can ask for up to four clips, each dispatched and charged as its own generation.
What it will not do
- Give you 1080p. H3 is pinned to 720p here and there is no control that asks for more.
- Give you a cheap draft pass. There is no 480p tier either, so the smallest possible H3 spend is 160 credits. Grok Imagine renders a 480p draft for a fraction of that.
- Take a second reference image. Exactly one still is forwarded, as frame one, and anything after it never reaches the model.
- Cut to five or six seconds. The lengths are 4, 8 and 12, with nothing in between.
- Reach MiniMax's own 2K tier. MiniMax sizes output as 768P or 2K, neither of which matches a value this site offers, so 720p is the single rate both rows are priced at.
MiniMax, Hailuo and H3 are one thing
MiniMax is the company. Hailuo is its video product line. H3 is the current generation of that line. So Hailuo 3 and MiniMax H3 are two names for one model, and whichever name you arrived with, you are already in the right place.
In the model list it appears once, as MiniMax H3, with a text-to-video row and an image-to-video row sitting behind the same name. The pill at the top of the create form is what picks between them. The model directory shows it next to everything else on the site with live credit prices, if you would rather compare before committing any.
One resolution tier, and why that is the honest number
Most video models here hand you a resolution dial. H3 does not, and that is a decision taken in this codebase rather than a gap in the model. MiniMax sizes its own output as 768P or 2K. The site offers 480p, 720p and 1080p. Those two lists do not intersect anywhere, so a 1080p button would have charged a 1080p multiplier for a request the model never receives.
720p is the rate both H3 rows are priced at, so 720p is the only tier shown. That is not a hypothetical worry: before the per-model capability table existed, four Kling ids offered a choice of 480p, 720p and 1080p, and all three selections produced byte-identical requests.
This is not unique to MiniMax. Kling 2.6 and Kling 2.5 Turbo Pro are pinned to a single tier for the same reason. If a 1080p master is not negotiable, Kling 3.0 and the Seedance family both take a real resolution parameter, and their pricing rows carry the multiplier that goes with it.
Both directions cost the same, which changes how you work
Text-to-video and image-to-video are separate rows in the pricing table and separate ids at the provider, but they carry the same base price and the same duration multipliers. A 4-second clip is 160 credits whichever way you came at it.
That reads like a footnote until a shot changes. Draft from a sentence, decide the framing has to match a photo already on your drive, upload the photo instead. The number on the Generate button does not move. On a model that charges a premium for image conditioning, the same change of mind costs you the difference every time you make it.
One still is used and it becomes the first frame, so from there the prompt's job is to describe the change rather than the picture. If the frame you want to animate needs work first, the image-to-image models can produce it, and H3 picks it up at frame one.
What length and quantity actually cost
Kie bills video per second and the duration multipliers on these rows are exactly linear, so the arithmetic is short: 160 credits at 4 seconds, 320 at 8, 480 at 12. No bulk discount for a longer take, and no surcharge either.
Quantity is the other multiplier and it is not hidden. Asking for four clips submits four independent generations and charges four times, because four independent generations is literally what happens at the provider. The button carries the total before you press it.
Credits come from a monthly plan or a one-time top-up, and a clip costs the same under either. The pricing page lists what each plan carries. A render that fails returns its credits to the balance they were taken from, with no ticket and no manual step.
Writing for a four-second take
Four seconds holds one action. Not a sequence, not a cut. One thing happening once. The prompts below follow the shape this site's own batch expander produces: shot size and camera height first, then one subject doing one countable thing, then a single slow camera move, then the key light and a palette in named colours, then the exact frame the clip should end on.
Two habits spoil more clips than anything else. The first is asking for readable text. No current video model spells, so a sharp label told to be legible comes back as confident nonsense, which is far more visible than softness. The second is writing exclusions as exclusions. These endpoints expose no negative-prompt field, and "no logos on the wall" is a reliable way to get logos. Ask for an unmarked matte wall instead.
Resolution, aspect ratio and duration are request parameters. Putting them in the prose spends words and changes nothing, and a stray ratio string occasionally renders as visible characters inside the frame. If you would rather read the comparison before spending, the 2026 video model roundup works through the lineup one model at a time.
Four prompts written for H3's shape
Each one is a single continuous take with one subject, one action and one camera move, which is what 4 to 12 seconds actually holds. None of them states a resolution, an aspect ratio or a duration: those are set on the form, and prose does not change them.
- Product, four seconds
Medium close-up at eye level, glossy studio product film. A matte-black steel water bottle stands on a brushed concrete slab, filling the central two thirds of the frame about 30cm from the lens, and one bead of condensation runs down the seam and stops halfway. Slow push-in, 100mm macro. Large softbox keyed 45 degrees camera-left, a narrow strip rim separating the edge from the background; slate grey, oxidised copper, warm bone white. The bottle stays in the focal plane, its brushed cap and seams crisply resolved, the room behind it falling off soft. Single continuous take, no cuts, ending on the bead held still at the seam.
The focus clause is the part people skip, and it is why a hero product comes back soft. Never write "shallow depth of field" into a shot whose selling point is the object itself.
- Image-to-video, one first frame
Hold the uploaded composition and let only the steam move: it lifts from the cup in a slow curl toward the upper left and thins as it goes, while the coffee surface keeps a faint ripple. Static locked-off camera, no reframing. Warm low window light from camera-right, unchanged from the still; espresso brown, cream, pale oak. Keep liquid weight and surface tension coherent. Single continuous take, no cuts, ending with the steam down to a single thread.
Your still is frame one, so this prompt's real job is to say what must not move. Only one image is forwarded; a second upload never reaches the model.
- Vertical, eight seconds
Wide shot from a low angle, handheld phone footage with a slight breathing sway and the subject locked in frame. A woman in a rust-orange coat takes four steps along wet pavement and stops at the kerb, held in the central vertical third with headroom to spare. Gentle push-in with parallax. Hard low sun from camera-left with a bounce on the camera side; slate blue, rust orange, wet asphalt black. Keep cloth inertia and hand anatomy coherent. Single continuous take, no cuts, ending as her shoulders settle.
For a 9:16 render, say it compositionally. "Held in the central vertical third" survives into the picture. The string 9:16 does not, and now and then it turns up as characters on screen.
- Twelve seconds, one move
Extreme wide at eye level, shot on 16mm with fine grain. A single freight train crosses a salt flat from left to right, entering at the left edge and clearing the frame in the final second, heat shimmer rising off the ground ahead of it. One slow half-orbit right, 50mm. Hard midday sun throwing one crisp shadow; bleached bone white, rust brown, pale sky blue. Plain slab-sided locomotive with unmarked painted panels. Single continuous take, no cuts, ending on empty flat and settling dust.
Twelve seconds is the longest take H3 offers and it is still one take. Note the exclusion written as a positive: "unmarked painted panels" rather than "no logos", which reliably summons logos.
MiniMax H3 against the rest of the video lineup
Base price for a 4-second clip at 720p, read from the same pricing table the Generate button charges against. The resolution and duration columns come from the per-model capability table, so nothing here is listed as offering a tier it cannot actually be asked for.
| 4s at 720p | Resolutions | Longest clip | |
|---|---|---|---|
| MiniMax H3 | 160 credits | 720p | 12s, 480 credits |
| Kling 3.0 Pro | 140 credits | 720p · 1080p | 12s, 420 credits |
| Veo 3.1 Fast | 150 credits | 720p · 1080p | 8s, 150 credits |
| Seedance 1.5 Pro | 35 credits | 480p · 720p · 1080p | 12s, 105 credits |
| Grok Imagine 1.5 | 45 credits | 480p · 720p | 12s, 135 credits |
MiniMax H3 questions
How much does MiniMax H3 cost?
On SynthPulse, MiniMax H3 costs 160 credits for a 4-second clip, 320 credits for 8 seconds and 480 credits for 12 seconds, all at 720p. Text-to-video and image-to-video are the same price. There is no per-seat fee and no minimum spend.
Is MiniMax H3 the same thing as Hailuo 3?
Yes. MiniMax is the company and Hailuo is its video product line, so "Hailuo 3" and "MiniMax H3" name the same generation of the same model. In the model list here it is filed under MiniMax H3.
Can MiniMax H3 animate a picture I already have?
Yes, and it costs the same as text-to-video. Switch the form to image-to-video, pick MiniMax H3 and upload one still. It is sent as the first frame, with your prompt describing what happens from there. Exactly one image is used, so there is no stack of references to curate.
What resolutions and lengths does MiniMax H3 support?
720p output at 4, 8 or 12 seconds, in 16:9, 9:16 or 1:1. No other resolution is offered, on purpose: MiniMax's own tiers are 768P and 2K, which match none of the 480p/720p/1080p values used by other models here, and 720p is the rate this model is priced at.
Can I generate more than one clip at a time?
Up to four per submission. Each clip goes to the provider as its own generation and is charged as its own generation, so four 4-second clips cost four times 160 credits. The total sits on the button before you press it.
Is there a free MiniMax H3 option?
No. Every H3 render calls MiniMax's paid model, so it always spends credits. What is free is everything up to the render: the account itself, and the AI lyrics generator. There is no subscription requirement either, a one-time top-up works, and a render that fails is returned to your balance automatically.
How do I use MiniMax H3?
Open the video studio, switch between text-to-video and image-to-video with the pill at the top of the form, pick MiniMax H3 from the model list, then describe the shot and set length and aspect ratio. In image-to-video mode you upload the still that becomes the first frame. The button shows the exact credit cost of that exact combination before you submit, and the finished clip appears on the same screen when the render lands.