The Hailuo 3 generation from MiniMax, available as H3 Turbo, H3 Max Turbo and H3 Turbo Reference. Text to video, image to video and reference to video at 480P or 768P for a flat 2 to 3 credits per clip, no watermark, on every Aura AI plan.
MiniMax H3 is the video model that MiniMax, the Shanghai company behind the Hailuo line, released on July 31, 2026. It succeeds Hailuo 2.3, and although MiniMax dropped the Hailuo name for this generation, most creators still search for it as Hailuo 3 or Hailuo H3. The design goal was an omni-modal generator: one model that understands text, images, video clips and audio tracks as references and returns a finished video with sound in a single pass.
Upstream, H3 produces 4 to 15 second clips at 24 fps up to 2K, with native stereo audio, and accepts up to 9 reference images, 3 video clips and 3 audio clips (12 files in total) in one job. It can copy camera movement from a reference video and apply instruction-based edits to an existing clip. In the Artificial Analysis rankings published on release day, H3 took first place for video editing, second for text-to-video and third for image-to-video. MiniMax also announced that open weights for the base model would follow.
On Aura AI, H3 runs through three fast Turbo endpoints. H3 Turbo handles text-to-video and image-to-video for 3 credits, H3 Max Turbo is the 2 credit option for the same modes, and H3 Turbo Reference locks characters and products from 2 to 6 reference images for 3 credits. Output is 480P or 768P, silent, delivered as an MP4 without watermark. It sits next to Kling O3, Pixverse 5 and Seedance 2.5 in the same selector.
Six things that set the H3 generation apart from Hailuo 2.3
Hailuo 2.3 on Aura AI only animated a still image. H3 Turbo and H3 Max Turbo accept a plain text prompt, or an image when you need an exact subject.
H3 Turbo Reference takes 2, 4 or 6 reference images depending on your plan and keeps the face, outfit or product consistent through the clip. Upstream the model goes further with video and audio references as well.
No per-second surcharge and no resolution multiplier. A 768P clip costs exactly the same as a 480P one, which makes H3 Max Turbo at 2 credits one of the cheapest text-to-video options in the catalog.
21:9 ultrawide, 16:9, 4:3, 1:1, 3:4 and 9:16 vertical. Hailuo 2.3 offered three. The wide and square formats matter for cinematic storyboards and product tiles that do not fit a phone frame.
Upstream, H3 can read the camera movement from a source video and reproduce it on new content. On Aura AI you steer the camera through the prompt: describe the dolly, orbit or handheld feel and the model follows.
Upstream, H3 applies a sentence-long edit to an existing clip, which is where it ranked first in the Artificial Analysis video editing leaderboard at launch. Aura AI exposes the generation modes.
Four steps from prompt to a silent 768P clip
Flat cost per clip, identical at 480P and 768P, no duration or resolution multiplier
| Variant | Modes | 480P | 768P | Reference images |
|---|---|---|---|---|
| MiniMax H3 Turbo | Text to video, image to video | 3 credits | 3 credits | None |
| MiniMax H3 Max Turbo | Text to video, image to video | 2 credits | 2 credits | None |
| MiniMax H3 Turbo Reference | Reference to video (16:9, 9:16) | 3 credits | 3 credits | Starter 2, Pro 4, Premium 6 |
Where the new MiniMax model lands among the budget video options on Aura AI
| Feature | MiniMax H3 | Hailuo 2.3 | Kling O3 | Pixverse 5 |
|---|---|---|---|---|
| Modes on Aura AI | Text, image, reference | Image to video | Text, image | Text, image |
| Max resolution | 768P (2K upstream) | 1080p (upstream) | 1080p (4K via Kling O3 4K) | 1080p |
| Duration | 5 s | 6 or 10 s (upstream) | 3 to 15 s | 5 or 8 s |
| Native audio | No | No | ✓ Yes | No |
| Aspect ratios | 6 | 3 | 3 | Multiple |
| Pricing model | Flat 2 or 3 | Flat 2 | 3 + 0.25 per extra second | 1 + 0.2 per extra second |
| Credits, 5 s clip | 2 (Max Turbo) or 3 | 2 | 3 | 1 |
| Plan | All plans | All plans | All plans | All plans |
Where a flat-priced, multi-ratio model earns its place
At 2 credits per clip, H3 Max Turbo is the model for blocking out a sequence before you spend on a 1080p render. Generate every shot, pick the framings that work, then re-run the keepers on a higher-tier model.
Feed H3 Turbo Reference two to six photos of the same person or mascot and generate multiple scenes with matching features, for episodic content where the character must stay recognizable.
Marketplace listings and catalog pages need 1:1 and 4:3 motion tiles, formats most video models skip. Start from a product photo in image-to-video mode and add a slow turn or a light sweep.
Reels and Shorts often run with a voice-over or licensed track on top, so native audio is wasted. H3 delivers clean 9:16 footage you can score in the editor, cheaply enough for dozens of variations.
Common questions about MiniMax H3 on Aura AI
Three Turbo variants, 480P or 768P, six aspect ratios, flat 2 to 3 credits per clip, no watermark. Available on every plan, worldwide.