Lovspark/Models/MiniMax-H3

MiniMax-H3

An omni-modal audio-video generation model supporting multimodal inputs including text, images…

Reference 16:9 2K 15s

Scroll to learn more

MiniMax-H3 logo Video model

An omni-modal audio-video generation model supporting multimodal inputs including text, images, video, and audio. It can generate videos up to 2K resolution and 15 seconds with native stereo audio, while also supporting video editing and motion transfer for filmmaking, advertising, e-commerce, gaming, and creative production.

Why use MiniMax-H3

An omni-modal audio-video generation model supporting multimodal inputs including text, images, video, and audio. It can generate videos up to 2K resolution and 15 seconds with native stereo audio, while also supporting video editing and motion transfer for filmmaking, advertising, e-commerce, gaming, and creative production.

Resolution and framing

Outputs at 768p and 2K, across 6 aspect ratios from 16:9 to 3:4. Pick the frame before you generate instead of cropping afterwards.

Clip length and audio

Generates clips up to 15 seconds, with audio included. Long enough for a shot, short enough to iterate on.

Cost and speed

20 credits per run. Both are shown on the canvas before you run, so a batch never costs more than you expected.

How it works

How to generate AI videos with MiniMax-H3 on Lovspark

From first prompt to production-ready video in four steps.

MiniMax-H3 Step 1

Open the video generator

Pick MiniMax-H3 from the model selector. Credits and expected run time are shown before you commit.

+ Step 2

Add your references

Attach up to 9 reference images to lock the subject, the product or the style — instead of describing it all in words.

Slow push-in on the subject, late afternoon light, shallow depth of field Step 3

Write the prompt

Describe subject, composition, lighting and mood. The model reads long prompts, so detail pays off on Lovspark.

Step 4

Keep the take that works

Compare the variants, then send the take you keep into the next node on the canvas — no re-uploading, no re-describing.

Specifications

Credits per run20
Resolution768p · 2K
Aspect ratios16:9 9:16 1:1 21:9 4:3 3:4
Reference imagesup to 9
Duration5–15s
AudioSupported
Model keyMiniMax-H3

Frequently asked questions

What is MiniMax-H3?

An omni-modal audio-video generation model supporting multimodal inputs including text, images, video, and audio. It can generate videos up to 2K resolution and 15 seconds with native stereo audio, while also supporting video editing and motion transfer for filmmaking, advertising, e-commerce, gaming, and creative production. It runs inside Lovspark's video generator and on the node canvas, so it can be swapped for another model without rebuilding the workflow.

How much does MiniMax-H3 cost per run?

20 credits per run. The cost is shown on the node before you generate, and a batch multiplies it by the number of images requested.

What resolutions and aspect ratios does it support?

768p and 2K, in 6 aspect ratios: 16:9, 9:16, 1:1, 21:9, 4:3, 3:4.

How long can a clip be?

Up to 15 seconds per run, with audio.

Ready to try MiniMax-H3?

Open it on the canvas alongside every other model — switching costs nothing.

Made with Lovspark

Public work from the Lovspark community.

More models

Same canvas, same workflow — switch without rebuilding anything.

GPT2.5 Image logo

GPT2.5 Image

New

Bring your spatial vision into focus with higher image quality and precise editing. Explore materials, lighting, and furnishings through text prompts and reference images, while better preserving the original composition and surrounding details during local edits. From initial concepts to refined design presentations, replace furniture, adjust finishes, and explore different atmospheres—helping designers compare options, refine details, and communicate a cohesive vision.

30 credits 2K 7 refs
View model
GPT2 Image logo

GPT2 Image

New

GPT-2 Image (Note: actual model is GPT-Image-2, released by OpenAI in April 2026): First image generation model with "thinking" capability. Achieves ~99% text rendering accuracy (up from 90-95%), supports up to 4096×4096 resolution, generates up to 8 consistent images per request. Features Thinking mode with integrated reasoning and web search.

60 credits 2K 7 refs
View model
Nano Banana Pro logo

Nano Banana Pro

Nano Banana Pro is Google's Gemini 3 Pro Image model, released in November 2025. In the Lovspark AI image generator it turns clay models and site plans into photoreal renders, fits out a bare space from a style reference, and lays out boards and plans with legible text, at up to 4K.

25 credits 2K 6 refs
View model
Seedream-5.0-pro logo

Seedream-5.0-pro

New

Seedream 5.0 Pro is ByteDance’s latest AI creation powerhouse—it doesn’t just generate pretty pictures, it actually understands design. It turns raw data and messy text into polished infographics, lets you edit anything with a simple click or lasso (just like Photoshop), and delivers lighting and skin textures so real they’ll make you do a double take. It handles over a dozen languages natively, from English to Arabic, with flawless rendering. No more blind-box generation—this new model is now live on our system, ready to surprise you.

10 credits 1K 4 refs
View model
Grok-image logo

Grok-image

Grok’s image model is fast, detailed, and flexible in style, making it ideal for quickly creating high-quality visuals.

8 credits 1K 3 refs
View model
Nano Banana 2 logo

Nano Banana 2

Lightweight and fast version optimized for real-time, high-concurrency image generation tasks with low latency. Supports standard resolution outputs.

18 credits 2K 8 refs
View model