MiniMax H3 (Hailuo 03) video generator

MiniMax H3, also called Hailuo 03, is MiniMax's AI video model, released July 31, 2026. It turns text, photos, video clips and audio into 4–15-second videos at up to native 2K, with stereo sound and spoken dialogue.

Try it in your browser: new accounts get 40 free credits, enough for a 10-second 768p video. No card needed.

Try MiniMax H3 freeCalculate the cost
Made with MiniMax H3 from one text prompt: 10 seconds at 2K. All three lines are spoken as written, with rain and jukebox music generated with the picture; turn the sound on to hear them.
Prompt
Late night in a small roadside diner, rain streaking down the window. A tired waitress in her fifties refills the coffee of a young man in a wet denim jacket sitting alone at the counter. She says, "You've been staring at that phone for an hour." He looks up and gives a small smile: "Waiting for a call that isn't coming." She slides a slice of pie toward him: "On the house." Warm neon light, shallow depth of field, slow push-in from a medium shot to a close-up of his face. Sound of rain on glass, a humming refrigerator, quiet jukebox music in the background. Natural lip sync, no subtitles, no text.
Try this prompt

What is MiniMax H3 good at?

Sound and dialogue, sharp 2K picture, and image to video that stays true to your photo, all from one model that reads text, images, video and audio.

Stereo sound and dialogue in every video

Speech, sound effects, ambience and music are generated with the picture, in stereo. Write spoken lines in quotes; in the diner sample above, all three lines are spoken as written.

Native 2K

2K returns 2560x1440 for 16:9, sharper than 1080p. 768p is there for quick, cheaper drafts.

Image to video that keeps your photo

Start from a photo, or a first and a last frame. The video keeps the look of your image. It ranks first for image to video on the Artificial Analysis Video Arena.

Text, images, video and audio in one model

One model reads the prompt and every reference together. On VisionMorphic: up to 5 images, 3 clips and 3 audio files per video, mentioned in the prompt as @Image1, @Image2.

Readable text in the picture

MiniMax lists readable text among H3’s strengths. Put the exact words in quotes; the "OPEN LATE" sign below came out right on the first try.

Open weights

MiniMax published the weights under its own community license. You do not need them here: VisionMorphic runs MiniMax H3 in the browser.

MiniMax H3 examples with prompts

Each video is a single MiniMax H3 generation at 2K, shown as it came out. Open any prompt in the generator with one click.

Image to video from a product photo

The start frame below was the only image. MiniMax H3 kept the bottle, the slate and the light, and added the splash, the orbit and the sound.

Start frame: an amber dropper bottle on wet black slate
Start frame. Made with an image model for this example; the video starts from it.
Prompt
Slow-motion product shot. A single drop of water falls from above and lands on the white cap of the amber dropper bottle, sending a fine crown splash into the light beam. More droplets bead and roll down the glass as the camera slowly orbits halfway around the bottle, the amber serum glowing as it catches the light. Dark studio, one soft top light, crisp detail on glass and water. Sound of water drops, a deep soft whoosh, and a low ambient tone. No text.
Try it with your product photo

Text you can read

The sign reads "OPEN LATE" exactly as written in the prompt, in one generation, with the camera tilting down to the kitchen.

Prompt
A rainy night street in a quiet city. Above a small noodle shop, a red neon sign flickers twice and then lights up steadily, clearly reading "OPEN LATE". The camera slowly tilts down from the sign to the steamy shop window, where a cook is tossing noodles in a wok. Reflections of the neon shimmer on the wet sidewalk. Sound of rain, a buzzing neon hum, and sizzling from the wok.
Try this prompt

Vertical video on the beat

A 9:16 clip for TikTok, Reels and Shorts. The beat is generated with the video, so the moves land on it.

Prompt
Vertical video. At night on an empty rooftop parking lot, a young street dancer in an oversized white hoodie performs a sharp popping and locking routine to a heavy hip-hop beat. City lights blur in the background, a light drizzle makes the concrete shine. The camera circles him at chest height, then drops low for a final freeze pose. Every hit lands on the beat. Sound: punchy hip-hop beat, sneakers squeaking on wet concrete, distant traffic. No text.
Try this prompt

How does MiniMax H3 rank?

First for image to video on the Artificial Analysis Video Arena, which ranks models from blind votes between two clips made from the same prompt. Positions as of October 20, 2026; the top entries are only a few points apart.

Image to Video
#1 · Elo 1140
Text to Video
#5 · Elo 1158
Video Editing
#4 · Elo 1125

How much does MiniMax H3 cost?

4 credits per second at 768p and 6 at 2K. 768p is open to free accounts; 2K needs Pro or Ultra. A reference video's length is counted at the same rate. Pick a length and resolution to see what it costs:

Resolution
This video costs
40 credits
40 free credits
Covers it
Basic, 600 credits a month
15 videos a month
Pro, 3,600 credits a month
90 videos a month
Ultra, 9,000 credits a month
225 videos a month

Rates as of October 20, 2026. 10 free credits are added every day, up to 40. Free videos carry a watermark, are kept 7 days and are not for commercial use. See plans

Try MiniMax H3 free

MiniMax H3 vs Seedance and Kling: which should I use?

Use MiniMax H3 for image to video, dialogue, readable text and the cheapest 2K. Use Seedance 2.5 for videos longer than 15 seconds, Kling 4.0 for keyframe control up to 30 seconds, and Kling 3.0 Omni for multi-shot stories and 4K. All of them are in the same generator and share one credit balance.

ModelBest forLengthUp toImages / videos / audio5-second video, credits
MiniMax H3Image to video, dialogue, readable text, 2K at a low cost
Sound always on
4–15s2K5 / 3 / 320 at 768p
Seedance 2.5Longer videos and many references
Up to 30 references, Auto ratio
4–30s1080p30 / 10 / 1060 at 480p
Kling 4.0Keyframe control over clips up to 30 seconds
Up to 10 frames, sound always on
3–30s1080p10 / 5 / 365 at 720p
Kling 3.0 OmniMulti-shot stories and 4K
Storyboards, consistent subjects
3–15s4K7 / 0 / 040 at 720p
Seedance 2.0 MiniThe cheapest drafts
Auto ratio
4–15s720p9 / 3 / 320 at 480p

MiniMax H3 specs

Made by
MiniMax (the Hailuo AI team), released July 31, 2026
Other names
Hailuo 03, Hailuo 3.0
Modes
Text to Video, Image to Video (first frame, optional last frame), Multi Reference
Length
4 to 15 seconds
Resolution
768p or 2K (2560x1440 at 16:9)
Aspect ratio
16:9, 4:3, 1:1, 3:4, 9:16, 21:9; Image to Video follows the start frame
Sound
Stereo, always on
References on VisionMorphic
Up to 5 images, 3 videos and 3 audio clips (15s of video and 15s of audio in total)
Prompt length
Up to 7,000 characters

How do I use MiniMax H3?

Sign in, pick a mode, write a prompt and generate. It takes a few minutes per video.

  1. 1. Sign in with Google

    No card needed. New accounts get 40 free credits, enough for a 10-second 768p MiniMax H3 video.

  2. 2. Pick a mode

    Text to Video takes a prompt. Image to Video takes a start photo. Multi Reference takes images, clips and audio, mentioned as @Image1, @Image2.

  3. 3. Generate

    The Generate button shows the credit cost before you submit. Failed generations return their credits.

Questions

Is MiniMax H3 free?

You can try it free at 768p. New accounts get 40 free credits and MiniMax H3 uses 4 per second at 768p, so they make one 10-second video. 10 credits are added every day, up to 40. Free videos carry a watermark and are not for commercial use; 2K needs Pro or Ultra.

Is MiniMax H3 the same as Hailuo 03?

Yes. H3 is MiniMax’s name for the model; Hailuo 03 and Hailuo 3.0 come from the Hailuo AI app it powers. It follows Hailuo 02 and Hailuo 2.3.

How much does a MiniMax H3 video cost?

4 credits per second at 768p and 6 at 2K: a 10-second video is 40 or 60 credits. Basic includes 600 credits a month, Pro 3,600. See all plans.

Can I turn the sound off?

No. MiniMax H3 always generates a stereo soundtrack. To get a silent clip, mute or remove the track after downloading.

Is this the official MiniMax website?

No. VisionMorphic is an independent product. It is not affiliated with or endorsed by MiniMax.

Make your first MiniMax H3 video

40 free credits make a 10-second 768p video with sound. No card needed.

Try MiniMax H3 free