Write the scene and add up to 9 reference images. You get a 4-15 second clip in native 2K at 24fps, with stereo dialogue, effects and ambience in the same pass. Every render runs on MiniMax H3, the model also known as Hailuo 03.
Every clip on this wall came out of MiniMax H3 as 2K AI video with audio. No one upscaled a frame or added a soundtrack afterwards. Hover or tap any card and you hear the stereo track it was born with.
You upscale soft renders, then spend the night rebuilding the audio bed. You download one finished MP4 — nothing to grade, nothing to dub.
One pass renders native 2K at 24fps plus stereo dialogue, effects and ambience — no tier to pick.
Your character's face drifts between shots and the series stops looking like one story. Your cast holds face, wardrobe and voice from shot one to shot ten. Drop three angles of the same illustrated courier and every new location returns the same courier.
Feed up to 9 reference images, 3 videos and 3 audio clips into one reference-to-video render.
You have the opening still and the ending still, but no way to bridge them. Your transition lands exactly where you planned it, every retry.
Image to video takes a first frame, a last frame, or both — H3 fills the motion in between.
Your track is finished, but matching motion to it by hand eats the evening. Your cuts and camera moves ride the audio you already made.
Add up to 3 wav or mp3 clips, 2-15s each, alongside an image or video reference.
You cram camera moves, dialogue and timing into a one-line prompt box. Your clip follows the script you wrote, beat by beat.
Text to video reads prompts up to 7,000 characters — beats, lines and camera notes included.
One scene, five platforms, five reframing sessions before anything goes out. Your 9:16 short and 21:9 spot come out composed, never center-cropped. Reference to video adds adaptive, which reads the frame from your references instead of forcing a ratio.
Choose from six aspect ratios plus adaptive, and any whole-second length from 4 to 15.
Publish episode after episode where the same character, wardrobe and voice survive every new scene and location. You upload nine reference images of your lead, prompt the next beat, and get a 2K episode where the face never drifts between shots.
Publish vertical dialogue clips on a daily cadence without filming, recording a voice, or dubbing anything afterward. You write the exchange, pick 9:16, and get a 2K MP4 where both characters speak in stereo with room tone already sitting under them.
Deliver a 21:9 hero cut and a 9:16 placement of the same product moment without booking a separate reframing pass. You describe the product moment once, render it at 21:9 and again at 9:16, and hand both 2K MP4s to the media buyer the same afternoon.
Turn a finished voiceover or music hook into a matching video clip without hand-timing a single cut. You pair a 15-second hook with one still, hit generate, and get a 2K clip whose cuts and motion follow the audio you already own.
You do not have to take this page's word for it. These are launch-week posts about MiniMax H3 from creators, benchmarks and the team behind the model — tap any card to read the original on X.
MiniMax (official)
@MiniMax_AI
MiniMax H3: Omni-Reference, Commercial-Grade Generation, Unbeatable Cost Efficiency, Open WeightsOfficial
MiniMax (official)
@MiniMax_AI
H3 is already live with our Day 0 partners 🚀 A huge thank you to @fal for the incredible support. Excited to see what creators build with it! #MiniMaxH3 #OpenWeightsOfficial
Hailuo AI (MiniMax)
@Hailuo_AI
Build your first game demo with H3 🧑💻🎮Official
ComfyUI
@ComfyUI
MiniMax H3 is native in ComfyUI, day zero with the weights. Model highlights: → Text-to-video, prompt only → Image-to-video → First-and-last-frame - control the opening frame, the closing frame, or both…Official
Artificial Analysis
@ArtificialAnlys
MiniMax H3 is #1 in Video Editing and ranks top 3 in both Text to Video and Image to Video on Artificial Analysis. MiniMax plans to release its weights under the MiniMax Community License, which would make it the leading open weights model by far…Benchmark
Umesh
@umesh_ai
Minimax H3 from Hailuo AI is here! One of my first tests! more coming soon..
JSFILMZ
@JSFILMZ0412
My First Impressions of the New MiniMax H3 Model -Very promising I2V -Need to test fighting scenes more -Nice progress with Omni Ref from medium close up.
Stable Diffusion Tutorials
@SD_Tutorial
MiniMax-H3 🤓 now Open sourced ! -multimodal generative model -unified architecture for text, images, video, and audio -generating audio -video with stereo -supporting up to 15 seconds at 2K resolution -can run on RTX 3060
Mark Kretschmann
@mark_k
MiniMax H3 is now open on Hugging Face! 🔥 @Hailuo_AI released the weights for H3, their general-purpose omni-modal generative model. It unifies text, images, video, and audio in one context and generates video with native stereo audio, up to 15 seconds at 2K.
Michael Guo
@Michaelzsguo
MiniMax H3 launched with plenty of fanfare, and for good reason. It puts movie production at our fingertips…
Signing up is free with no credit card, and your free plan includes starter credits. MiniMax H3 ships one 2K quality tier, so a render's cost scales with its duration alone — buy a credit pack when you are ready to render more.
You asked what MiniMax H3 really outputs, what it accepts as reference input, and what is actually free here. These are the answers, with the real limits attached.
MiniMax H3 — also called Hailuo 03 — is MiniMax's video model that renders picture and native stereo audio together in a single pass, at 2K and 24fps. This page runs it as an online hailuo 3 ai video maker: write a prompt, optionally add reference media, and download a finished MP4.
Yes. Native audio is part of every render, not a switch you turn on. Dialogue, sound effects and ambience come out with the picture in stereo. There is nothing to re-sync in an editor afterwards.
Native 2K at 24fps, and that is the only tier — there is no quality dial and nothing is upscaled afterwards. Because picture quality is fixed, a render's cost tracks its duration alone.
Yes. Image to video accepts a first frame, a last frame, or both. Bracket a move between two stills you already have, and H3 fills the motion in between.
Reference to video takes up to 9 images, 3 videos and 3 audio clips in one render. Images hold a character's face and wardrobe across shots, videos guide motion, and audio clips — wav or mp3, 2-15s each, 15s combined — set pacing and voice. Audio references travel with an image or video reference; they cannot be sent on their own.
Signing up is free and needs no credit card, and your free plan includes starter credits. Renders themselves draw on credit packs, and with one 2K tier a clip's cost scales with its duration.
You get six aspect ratios — 21:9, 16:9, 4:3, 1:1, 3:4 and 9:16. Reference to video adds an adaptive option that reads the frame from your references. Duration is any whole second from 4 to 15.
No. This is an independent hailuo 3 video generator built on MiniMax H3. The model is MiniMax's work; this site adds a browser workflow, credits and render history around it. Every capability figure on this page comes from the model's published documentation.
Run the same brief on each and judge the output. H3 gives you one native 2K tier with stereo audio in the same pass. It adds first and last frame control, plus up to 9 reference images for character consistency across shots. Sign-up here is free, so you can test it against your shortlist before committing credits.
Nothing is lost. Jobs run server-side, and your generation history restores the finished clip when you come back. A failed render shows a retry — you are not charged twice for it.
Jump straight to the generator, the example wall, pricing or the answers.
Open the hailuo 3 ai video maker, type one scene, and watch it come back with stereo sound already on. Sign-up is free with no credit card, and starter credits are included.
