Video relighting
Upload a video and a lighting reference image, then regenerate the footage with new lighting. Choose 480P up to 2K output.
Relight a videoMiniMax H3 Max tools
MiniMax H3 video tools
Relight a video, lip-sync a still image to audio, or insert a generated scene into your footage with MiniMax H3 Max, then read how the MiniMax H3 model works.
Lip sync
Upload a still image and an audio clip to create a talking video that follows the audio.
Lip-sync an imageInsert a scene
Replace a section of your video with a newly generated scene that connects to the footage around it.
Insert a sceneMiniMax H3 MaxVideo relightingLip syncInsert scene
The tools above run on MiniMax H3 Max. Standalone H3 text-to-video and image-to-video generation is not available in the Image to Layers video generator yet; the sections below describe what the MiniMax H3 model supports according to MiniMax.
Official model capabilities
What the MiniMax H3 model supports
MiniMax describes H3 as understanding visual and audio context together: a prompt, a product image, opening and ending frames, or references for identity, movement, voice, camera direction, and editing rhythm. These are the model’s capabilities, not a list of features of this site.
Multimodal creative direction
Combine instructions with images, videos, and audio references, then explain how each reference should shape the result.
Video with native sound
Generate visuals, voices, ambience, effects, and music as parts of the same audiovisual scene.
768p base, 2K by regeneration
MiniMax generates the base video at 768p and delivers 2K through its separate H3-Regenerate-2K capability, rather than as native 2K output.
4–15 second duration
Build a short product motion clip or a longer scene with dialogue, action, and planned camera changes.
First and last frame control
Provide a starting frame, an ending frame, or both to give the shot a clearer visual destination.
Reference-guided generation
Use references for a character, product, motion pattern, camera move, visual style, voice, or editing pace.
Creation modes in the MiniMax H3 model
01
Text to video
Describe the setting, subject, action, camera, lighting, timing, dialogue, and sound. H3 builds the complete scene from the instructions.
02
Image to video
Upload a product photo, illustration, poster, or character image, then describe controlled action and camera motion.
03
First and last frame
Define where the shot begins and ends for transitions, transformations, camera reveals, and planned progression.
04
Reference video
Provide media that communicates motion, camera language, performance, identity, voice, audio, or editing rhythm.
Where multimodal video models are used
Product advertising
Animate product photos with controlled camera movement, reflections, environmental effects, spoken lines, and brand-appropriate sound.
E-commerce visuals
Create demonstrations, landing-page hero videos, feature reveals, and promotional clips without organizing a full shoot.
Social media ads
Generate vertical product stories, campaign hooks, launch teasers, and short-form creative for social channels.
Animated posters
Transform static artwork into moving compositions with atmosphere, music, typography, and transitions.
Character and story scenes
Use identity references, dialogue, performance direction, and environmental sound for more directed narrative scenes.
Concept development
Visualize an advertisement, film shot, game cinematic, launch, or creative pitch before full production.
How to use the MiniMax H3 Max tools
Step 1
Pick the H3 Max tool
Choose video relighting, lip sync, or insert scene, depending on whether you want to change the light, make an image speak, or replace part of a clip.
Step 2
Upload your media
Add the video, still image, audio, or reference image that the tool asks for.
Step 3
Set the options
Select the resolution and any scene or lighting settings the tool offers, then check the credit cost before you start.
Step 4
Generate and review
Review visual continuity and audio timing, then adjust your inputs and run it again if needed.
Step 5
Download your result
Find the finished video in your dashboard and use it in advertising, social media, presentations, or further editing.
Prompt guide
Write the complete audiovisual scene
Subject + environment + action + camera + timing + dialogue + sound + preservation rules. These tips follow MiniMax’s guidance for the H3 model.
Describe timing
Use phrases such as “during the first three seconds,” “after the camera stops,” or “hold the final shot for two seconds.”
Connect sound to action
Explain when music begins, which action creates a sound effect, and whether dialogue stays in the foreground.
State what must not change
Specify logos, faces, colors, materials, clothing, typography, and proportions that must remain consistent.
Keep the scene achievable
A focused eight-second reveal is easier to control than several locations, characters, and unrelated actions.
Luxury product reveal
A matte-black perfume bottle stands in the center of a dim studio. A narrow beam of warm light travels across the bottle while the camera performs a controlled semicircular orbit. Fine golden particles move in the background without covering the product. A soft glass chime is heard as the logo becomes visible. Preserve the exact bottle geometry, label, logo, typography, and cap. Premium fragrance advertisement, realistic materials, restrained cinematic movement, 8 seconds, 9:16.
Fashion campaign
A fashion model stands beneath a covered city walkway after light rain. She turns toward the camera while her coat and hair move naturally in the breeze. The camera slowly tracks backward as reflected city lights appear on the pavement. Include subtle footsteps, distant traffic, and minimal electronic music. Preserve the model’s face, proportions, outfit, accessories, and skin tone. Editorial campaign, 10 seconds, 16:9.
Talking character
Use the provided character image as the identity reference. The character looks toward the window, pauses, and says, “I knew the signal would return.” Match lip movement to the sentence. Add quiet rain against the glass and a low atmospheric tone. Preserve facial identity, hairstyle, costume, and illustration style. Slow camera push-in, 7 seconds.
MiniMax H3 model specifications
- Model
- MiniMax-H3
- Input
- Text, image, video, and audio references
- Output resolution
- 768p base generation; 2K through H3-Regenerate-2K
- Duration
- 4–15 seconds
- Audio
- Native stereo audio
- Aspect ratios
- Landscape, portrait, square, and adaptive formats
MiniMax H3 or Hailuo 2.3?
MiniMax H3 (on MiniMax’s platform)
For native sound, longer clips, planned ending frames, and several visual or audio references. Base H3 generation is not part of this site’s video generator yet.
Hailuo 2.3 (available in the video generator)
For a simpler text-to-video or image-to-video workflow focused on visual generation.
Hailuo 2.3 Fast (available in the video generator)
For fast, lower-cost image motion tests before moving to a more detailed multimodal production pass.
What to know before generating
Complex hand interactions, crowded scenes, rapid identity changes, small typography, long spoken passages, and several overlapping actions can still produce inconsistencies. Review important commercial results frame by frame and confirm that product details, text, voices, music, and reference assets are appropriate for the intended use.
Frequently asked questions
What is MiniMax H3?
MiniMax H3 is a general-purpose multimodal generation model from MiniMax. It understands text, images, video, and audio references together and generates video with native stereo sound.
Can MiniMax H3 generate audio?
Yes. H3 can jointly generate speech, sound effects, ambience, music, and visuals as one audiovisual result.
How long can a MiniMax H3 video be?
Official MiniMax documentation supports integer durations from 4 to 15 seconds.
Does MiniMax H3 support 2K video?
MiniMax generates H3 video at 768p and reaches 2K through its separate H3-Regenerate-2K capability, so 2K is not native output for every generation. On Image to Layers, the video relighting tool offers output from 480P up to 2K.
Can I generate text-to-video or image-to-video with MiniMax H3 on Image to Layers?
Not at the moment. The video generator on this site does not include MiniMax H3 base generation yet. What you can use here are the MiniMax H3 Max tools: video relighting, lip sync, and insert scene. Text-to-video and image-to-video with H3 are available on MiniMax's own platform.
Which MiniMax H3 tools can I use here?
Video relighting, lip sync from a still image and audio, and insert scene, which replaces part of a video with a generated scene. Each one has its own page linked at the top of this page.
Can I provide both a first and last frame?
Yes. First-and-last-frame generation lets you define the starting and ending visual states of the video.
Can H3 use a reference video?
Yes. Video references can guide motion, camera behavior, performance, visual style, or editing rhythm when the prompt explains what to follow.
Is MiniMax H3 open source?
MiniMax released H3 model weights under its community license. The license includes geographic and usage conditions, so review it before treating the weights as unrestricted open source.
Do failed generations consume credits?
Failed or canceled provider generations are refunded according to the platform’s current credit policy.
Try the MiniMax H3 Max tools
Relight a video, make a still image speak, or replace part of a clip with a generated scene.
Read the practical MiniMax H3 guide