Images, video, and audio references
Turn your references into 5–15 second videos with synchronized sound and output up to 2K.
Reference Media
Max 12 total: 9 images, 3 videos, 3 audio. Media 2-15s. Files 10/50/10MB.
1/12 files, 1/9 images, 0/3 videos, 0/3 audio
Available credits: 0
BIWA Character Introduction
Create a 15-second premium cinematic character reveal using the BIWA character sheet as the sole visual authority.
MiniMax H3 accepts text, image, video, and audio references. Combine only the materials your scene needs for more precise creative direction.
Prompt Guide
Name uploaded files directly in your prompt.
Use references such as @Image 1, @Video 1, or @Audio 1 so the model knows which material to use.
Material Limits
Supported Input Combinations
Note: Reference images, videos, and audio are optional. Add only the materials needed for your prompt.
Images
JPG, JPEG, PNG, WEBP, BMP, TIFF, or GIF. Up to 10 MB each.
Videos
MP4 or MOV. 2 to 15 seconds, up to 50 MB each.
Audio
MP3 or WAV. 2 to 15 seconds, up to 10 MB each.
Explore product films, music videos, short dramas, and game interfaces made with references. Open a clip to see its prompt and available input materials, then reuse its settings above.
A way to show the model what you mean, alongside your written instructions.
A reference can explain the subject, visual style, motion, or sound more clearly than words alone.
Choose assets that each contribute something specific. An image might guide a product or character, a clip might guide movement, and audio might help communicate the intended sound. Your prompt connects those materials into one shot.
For more examples of asset roles and shot instructions, read the MiniMax H3 prompt guide.
Want to create from a text prompt or start and end frames? Explore the MiniMax H3 video generator.
Guide appearance, motion, and sound with references, then choose your output settings. The clips below illustrate creative possibilities; they are not verified tests of individual reference controls.
Image references can guide character appearance, clothing, product shape, materials, and settings. Assign each image a clear role: one for the subject, another for the environment or visual style. Review faces, text, and logos in the finished clip, as fine details may vary.
Video references can guide actions, camera movement, and editing rhythm. Specify which motion to borrow while using your image references for the new subject and setting. This helps separate how the shot moves from what appears in it, without requiring a frame-for-frame copy.
Audio references give the model sound context alongside your instructions for dialogue, music, and ambience. Describe how each sound should relate to the on-screen action, then review timing and clarity in the generated clip. An audio reference does not guarantee an exact copy of a voice or soundtrack.
Choose standard H3 output at 768P or 2K, with a duration from 5 to 15 seconds. Available aspect ratios are 16:9, 9:16, 4:3, 3:4, 1:1, and 21:9. Output duration controls the length of the new video; uploaded video and audio references have separate duration limits.
Start with the materials you already have and build a focused video brief for your next project.
Turn product photos into a short reveal or feature demonstration. Reference close-up details and describe the camera move, lighting, and setting you want.
Bring together wardrobe, model, and location references for a campaign concept. Specify which details each image should guide, then describe the performance and edit rhythm.
Use character portraits and a setting reference to plan a short scene. Write the action, dialogue, and expressions, then review how well the result preserves each character.
Combine visual references with a musical direction to explore a short music-video concept. Describe the texture, typography, movement, and timing of cuts in your prompt.
Use character artwork and interface references to visualize an equipment change, menu interaction, or scene transition before committing to production.
Give a static poster or campaign image subtle movement. Describe what should animate and what should stay fixed, and choose a vertical or landscape format for your intended placement.
From brief to finished clip
Describe your shot, give each reference a clear role, then choose your output settings and review the result.
Describe the subject, action, camera, lighting, and sound. Decide which reference will guide the appearance, setting, movement, or audio so each asset has a clear purpose.
Upload your images, video clips, or audio and mention them in your prompt. Choose an aspect ratio, 768P or 2K resolution, and a 5–15 second duration.
Watch the complete clip with sound. Check how it follows your references, including subject identity, movement, composition, and audio. Refine one instruction at a time for the next version.
Reference settings
The best setup depends on what you need to control. Start simple, then add more context only when it improves the result. These are suggested starting points to test with your own references.
| Goal | Best reference | Prompt instruction | Suggested start |
|---|---|---|---|
| Character identity | Clear portrait or full-body image | Keep the face, hair, outfit, and body type from the image. | 10s · 16:9 or 9:16 |
| Product accuracy | Product photo from clean angles | Preserve the shape, material, logo area, and color. | 6–10s · 2K |
| Camera movement | Short motion video | Borrow only the camera path and pacing from the video. | 5–8s · 768P draft |
| Audio mood | Music, ambience, or voice reference | Follow the tone and rhythm without changing the scene brief. | 10–15s · pair with an image or video reference |
Inputs, output settings, and what to expect from your references.
MiniMax H3 Reference to Video combines a written prompt with uploaded images, video clips, or audio to guide a new video. You can use different assets for the subject, setting, movement, or sound, then explain how they should work together. References guide the result rather than guaranteeing an exact copy of the source material.
In the image-to-video workflow on this site, you provide a first frame and can add a last frame to guide the shot. Reference to video lets you combine assets with different roles, such as a character portrait, a location image, and a camera-movement clip. Choose it when you want to describe how several references should influence a new scene.
Upload clear images that show the details you want to preserve, such as clothing, a product silhouette, materials, or colors. In your prompt, identify which image supplies the subject and which supplies the setting or style. Start with a few useful views, and check faces, labels, and logos in the output because small details can change.
Yes. Add a short clip and explain whether it should guide the action, camera path, or editing rhythm. For example, ask for the camera movement from the clip while using an uploaded product photo for the subject. Avoid asking the model to borrow everything at once when the reference contains people or scenery you do not want in the result.
An audio reference can provide context for the sound you describe, such as the mood of music, the rhythm of a scene, or an ambient sound. Explain how it should relate to the visible action and include any dialogue instructions in the prompt. Listen to the completed video to check timing and clarity; a reference does not guarantee an identical voice or soundtrack.
This generator allows up to nine images, three videos, and three audio files, with a combined maximum of twelve assets per request. Each image or audio file can be up to 10 MB, and each video can be up to 50 MB. These are upload limits, not a recommended asset count: begin with the references that have a clear role and add more only when needed.
Each video or audio reference must be between 2 and 15 seconds long. Uploaded video clips have a combined duration limit of 15 seconds, and uploaded audio files have their own combined limit of 15 seconds. For example, two 8-second video references exceed the video limit. Trim inputs to the relevant moment; these limits are separate from the generated video duration.
Use the asset mentions in the prompt field to assign each upload a role, then describe the action, camera, lighting, and sound. For example: use the first image for the jacket and the second for the street; have the character walk toward the camera in soft evening light, with footsteps and quiet traffic. Keep instructions consistent with the references and make one focused change when refining a result.
Standard H3 in this generator offers 768P and 2K output. You can start at 768P to review composition and motion, then choose 2K for a new generation when you need a higher-resolution result. Resolution does not guarantee more accurate faces, text, or logos, so review those details separately. Available settings can differ when you select another model.
For standard H3 here, you can select an output duration from 5 to 15 seconds and one of six aspect ratios: 16:9, 9:16, 4:3, 3:4, 1:1, or 21:9. Choose a format for the intended placement, such as 9:16 for a vertical clip. The output duration sets the length of the new video rather than copying the length of an uploaded reference.
Yes. Open an example and select Create with this example to load its prompt, available input images, aspect ratio, resolution, and duration into the generator above. Review the loaded materials and adjust the instructions before generating. Reusing an example gives you a starting setup; it does not guarantee the same video or automatically include reference files that are not provided with that example.
Use clear, compatible references and state which details matter most, such as the face, outfit, product shape, or material. Remove images that contradict those details and keep the requested action manageable within a short clip. After generation, check identity, proportions, text, logos, and continuity. If something drifts, simplify the references or revise the relevant instruction before trying another version.
Bring your images, video, or audio references together with a clear prompt. Choose your output settings and start creating.
Try Reference to Video
