Conch AI

What is Hailuo AI?

Conch AIIt is MiniMax’s video creation platform, capable of generating dynamic shots from text, images, and a variety of reference materials. It is suitable for advertising spots, character shorts, animated posters, and storyboard validation; specifically, MiniMax H3 integrates video generation, native audio, and command editing into a single creative workflow, allowing users to continue modifying characters, objects, scenes, and sounds based on an existing direction.

MiniMax was founded in early 2022 by Yan Junjie, and its R&D covers modalities such as text, speech, images, video, and music. The earliest publicly available web demo of Hailuo Video dates back to August 2024; Hailuo 02 was released on June 18, 2025, and support for controlling the first and last frames was added in August of the same year. H3, released on July 31, 2026, further integrated multimodal references with audio-visual generation.

Key Capabilities of H3

Text, First Frame, and First and Last Frame Generation

When there is no ready-made footage, you can use text to describe the subject, action, setting, and shot; when you already have a composition, use the image as the first frame and shift the creative focus to motion. Adding a closing frame to the "first and last frame" mode is suitable for transitions with clear starting and ending states, changes in objects, or short actions.

Keyframes define the endpoints, but it is still necessary to check whether the intermediate process is coherent. For example, if the product needs to be moved from a desk to a display stand, the intermediate path generated by the model, as well as the object’s shape and occlusion relationships, must all be visually reviewed. The prompt should specify both the action and the features that need to be preserved.

Multimodal Reference

Omni ReferenceCombinable images, videos, and audio: Images are suitable for illustrating the appearance of people or products; videos can serve as references for movement, composition, and camera work; and audio can provide directional cues. H3 supports a total of up to 12 reference files, including up to 9 images, up to 3 videos, and up to 3 audio clips; audio must be paired with either an image or a video.

Having a large number of reference materials does not automatically make the requirements clearer. It is more important to specify the purpose of each reference—for example, “This image is only for the character’s appearance,” or “This video clip is only for camera movement.” If two reference materials show conflicting elements—such as clothing, movements, or lighting—you should decide which to prioritize.

Video and audio are generated together

H3 supports videos ranging from 5 to 15 seconds in length, offering native stereo sound, common aspect ratios, and output up to 1440p. Dialogue, ambient sound, or music can be incorporated as part of the shot composition, rather than being added only after the footage is complete. When editing the final video, you should still verify that the order of dialogue, the content of the audio, and the on-screen events align.

Continue editing existing footage

H3 can add, remove, or replace characters and objects, and adjust the background, lighting, visual effects, dialogue, and sound according to instructions. When changing a detail, you should clearly specify “what to change” and “what to keep,” and then check whether the unmodified areas have also changed. MiniMax’sH3 IntroductionGeneration and editing are integrated into a single context, but these capabilities are specific to H3 and cannot be directly applied to other Haishuo models.

Which production stages is it suitable for?

• Advertisements and Product Updates:Use confirmed product images to set up close-ups, wide shots, or shots showing changes in the surroundings, and create candidate shots. Pay close attention to packaging text, logos, and product design.
• Character Shorts and Visual Storyboards:Prepare character appearance and action references, break a scene down into several distinct shots, generate them individually, and then organize them into a narrative.
• Extensions to Static Design:Add subtle movement to posters, illustrations, or game concept art while maintaining the existing visual direction, and avoid redesigning both the subject and the motion at the same time.
• Camera adjustments:Experiment with changes to the background, lighting, objects, or sound in the existing video, compare different creative versions, and then decide which ones to include in the final edit.

How to Create a Reference-Driven Shot

Suppose you want to have a pre-designed character walk into a coffee shop using the slow tracking shot from the reference clip. You can start by preparing the character art and a reference clip of the camera movement that you have permission to use.

1. Select a mode.Go to the creation page and select "MiniMax H3" from the model selector. Use "First Frame" only when you have a starting image; select "Omni Reference" when you need to use both character and motion references.
2. Organize the input.Place the reference material in the "Refs" section and clearly label each item's purpose. Describe the new scene, character actions, camera angles, and sound requirements, and avoid requesting that irrelevant characters or backgrounds from the reference clips be included.
3. Set the delivery specifications.Based on the content of the shot, select a duration of 5 to 15 seconds, as well as the aspect ratio and resolution. It’s easier to review the results when you clearly explain a single brief action than when you try to cram multiple scenes into a short clip.
4. Generate and check each item one by one.First, look at the characters' appearance and movements, then listen to the dialogue, ambient sounds, and music. When changes are needed, provide specific instructions—for example, adjusting only the lighting or a particular line of dialogue—and then check to see if other elements remain unchanged.
5. Include the complete, edited footage.After selecting the shots, standardize the editing rhythm, subtitles, and audio levels. For projects with multiple shots, check the characters’ costumes, props, and scene orientation, and add transition shots as needed.

Authorization must be obtained for all character portraits, audio, and reference videos. Web-based generation, APIs, and open-source models are different usage channels, each with varying fees, geographic scope, and licensing terms; when using the Hailuo Platform, projects must be organized according to the corresponding service plan, and the model deployment terms cannot be directly applied.

How to Choose Between Hailuo and Vidu

Both Conch and Vidu can create short clips from text, images, and reference materials. The H3 public workflow explicitly allows for the combination of images, video, and audio, and enables further editing of existing audio-visual content, making it ideal for comparing how motion, sound, and editing commands work together.

Vidu offers organizational tools such as Reference Assets, RefHub, and My References, making it easy to save characters, props, and scenes for later use in further generation; Q3 can also generate clips with dialogue, sound effects, and music. When the same subject needs to be reused across multiple shots, it’s important to compare the material management and reference features of both platforms—native audio should not be considered exclusive to just one of them.

A practical way to compare options is to prepare the same character designs and the same shot assignment, then examine how the input is organized, how easy it is to make revisions, and how the final video specifications and budget are calculated. Different models, durations, and the number of reference materials will affect costs, so it’s not advisable to base decisions about the entire production process solely on a single price quote or a promotional sample clip.