HeyGen Video: New MiniMax H3-Based Model at Low Cost

The launch of HeyGen Video marks a new stage in the AI video generation market. The company introduced its own model built on the open-source MiniMax H3 architecture.

Open-weight models are changing the approach to product creation. Instead of simply aggregating existing solutions, developers are now fine-tuning base models for their specific tasks.

HeyGen adapted the MiniMax H3 model to improve stability and control. Key improvements focus on frame consistency, prompt adherence, and audio synchronization.

The new model supports multiple input and output data formats:

• Video generation from text, images, and references
• Support for up to 9 images, 3 videos, and 3 audio files
• Output in 480p and 768p resolutions with durations from 5 to 15 seconds
• Frame formats ranging from 21:9 to 9:16

The key emphasis is not on generation speed, but on result predictability for commercial video production.

HeyGen Video's pricing strategy demonstrates an aggressive market entry approach. Special API usage tariffs are in effect until the end of October.

The cost of video generation is significantly lower than market alternatives:

1. 480p resolution — $0.01 per second
2. 768p resolution — $0.015 per second
3. Reference-to-Video (Ref2V) mode — $0.03 per second

For comparison, direct access to the MiniMax H3 API costs approximately $0.08 per second. Thus, HeyGen sells a derivative model cheaper than the original.

After the promotional campaign ends, the price will be $0.30 per second. The company also claims that their version outperformed Seedance 2.0, H3 Max, and Kling 3.0 in blind tests.

Access to HeyGen tools is available via API, OpenRouter, Runware, and ComfyUI nodes. The web interface is currently under development.