MiniMax H3 Reference to Video LoRA: Lora Support
Generate reference-to-video with MiniMax H3 Reference to Video LoRA. Multimodal references, custom LoRAs, and native audio. Try free.
Generate reference-to-video with MiniMax H3 Reference to Video LoRA. Multimodal references, custom LoRAs, and native audio. Try free.
About this model
MiniMax H3 Open Weights Reference to Video LoRA generates coherent 480P or 768P videos from text prompts and multimodal reference inputs (up to 9 images, 3 videos, 3 audios) with native stereo audio. It supports custom LoRA fine-tunes for style consistency, character preservation, and creative control.
Cost analysis
| Provider | Cost | Notes |
|---|---|---|
| muapiapp | $0.03 - $0.06 / sec | Pay per second of video generated. No subscription. |
| Fal.ai | Not available | Custom LoRA support not available. |
| Replicate | Not available | Model variant not hosted. |
Pay per second of video generated. No subscription.
Custom LoRA support not available.
Model variant not hosted.
** Competitor pricing is estimated based on similar model architectures and usage tiers.
Configuration schema
| Parameter | Type | Description | Default |
|---|---|---|---|
| Prompt | string | Text description of the desired video. Refer to reference inputs as <Picture 1..9>, <Video 1..3>, and <Audio 1..3>. | A cinematic ocean wave at sunrise with <Picture 1> character walking, highly detailed |
| Reference Image URLs | array | Reference image URLs (up to 9). | undefined |
| Reference Video URLs | array | Reference video URLs (up to 3). Total reference video duration is budgeted to 15 seconds. | undefined |
| Reference Audio URLs | array | Standalone reference audio URLs (up to 3, trimmed to 15s each). | undefined |
| LoRAs | array | List of LoRAs to apply (maximum 3). | - |
| Aspect Ratio | Enum (7 options) | Output video aspect ratio. | 16:9 |
| Resolution | Enum (2 options) | Output video resolution. 768p is native canvas, 480p is faster. | 480p |
| Duration (seconds) | int | Output video duration in seconds. | 5 |
| Seed | int | Random seed (-1 for random) | -1 |
Text description of the desired video. Refer to reference inputs as <Picture 1..9>, <Video 1..3>, and <Audio 1..3>.
A cinematic ocean wave at sunrise with <Picture 1> character walking, highly detailedReference image URLs (up to 9).
undefinedReference video URLs (up to 3). Total reference video duration is budgeted to 15 seconds.
undefinedStandalone reference audio URLs (up to 3, trimmed to 15s each).
undefinedList of LoRAs to apply (maximum 3).
-Output video aspect ratio.
16:9Output video resolution. 768p is native canvas, 480p is faster.
480pOutput video duration in seconds.
5Random seed (-1 for random)
-1Developer documentation
Formulate Prompt & References: Write a text prompt referencing <Picture 1..9>, <Video 1..3>, or <Audio 1..3> along with your reference media URLs (images_list, videos_list, audios_list).
Configure Aspect Ratio & Resolution: Select aspect ratio (16:9, 9:16, 1:1, etc.) and output resolution (480p or 768p).
Set LoRAs & Duration: Provide up to 3 custom LoRA objects in loras and choose a video duration between 3 and 15 seconds.
Submit Request: Send a POST request to /api/v1/minimax-h3-reference-to-video-lora:
curl -X POST https://api.muapi.ai/api/v1/minimax-h3-reference-to-video-lora \
-H 'Authorization: Bearer YOUR_API_KEY' \
-H 'Content-Type: application/json' \
-d '{
"prompt": "A cinematic ocean wave at sunrise with <Picture 1> character, highly detailed",
"images_list": ["https://example.com/character.jpg"],
"resolution": "480p",
"duration": 5
}'
Frequently asked
It is a multimodal reference-to-video AI model built on MiniMax H3 open weights supporting custom LoRA adapters and native audio generation.
You can provide up to 9 reference images, 3 reference videos, and 3 reference audios in a single request.
Yes, native stereo audio is synthesized automatically alongside the generated video based on your prompt and audio references.
You can supply an array of LoRA objects under the `loras` field in your request payload.
Billing is calculated based on output video duration (in seconds) and chosen resolution (480p vs 768p).