Hyperflow H3 Ranks 5th in AA-Video-I2V and 3rd in AA-Video-Editing

Hyperflow H3 from Video Rebirth, an 8-step LoRA adapter for the MiniMax H3 video model, ranked 5th in AA-Video-I2V v2.1 and 3rd in AA-Video-Editing v1.1. These are two benchmarks from the analytics firm Artificial Analysis: the first evaluates video generation from a still image (image-to-video, I2V), and the second evaluates instruction-based editing of existing video. According to Artificial Analysis, Hyperflow H3 costs less than half as much as Dreamina Seedance 2.5 in both tasks.

The adapter was released in September 2026 with open weights. It now supports both image-to-video generation and video editing.

Where Hyperflow H3 is strongest

Artificial Analysis breaks its results down by category. In AA-Video-I2V v2.1, the model performed best in Live-Action Film, meaning realistic live-action scenes. Only Dreamina Seedance 2.5 ranks higher there. In Multi-Scene & Narrative, which tests whether a model can keep several scenes and a storyline coherent, only Utopai X is ahead.

For teams making ad spots, previsualization or short narrative clips, these categories are more useful than the overall leaderboard position. They show how the model handles cinematic footage and a story told across several shots.

In AA-Video-Editing v1.1, Hyperflow H3 took third place and outranked MiniMax H3 at 768p, the model the adapter is built on. The comparison is not quite like for like: the table lists the base model at 768p, while Hyperflow H3's prices are quoted for 1088p. Even so, the adapter came out ahead of its base model on this task.

Pricing and open weights

In the Video Rebirth API, a minute of video in image-to-video mode costs $9.00, and a minute in editing mode costs $18.00. Both prices are for 1088p resolution. Artificial Analysis notes that this is less than half the cost of Dreamina Seedance 2.5 on both tasks. The post does not give the competitor's exact prices, so you will need to check Seedance's price list to calculate the absolute difference.

The "8-step" name means the adapter is designed to generate video in eight steps. For diffusion video models, the number of steps affects inference speed and cost. Artificial Analysis does not report how long a render takes or what self-hosted compute costs.

Open weights offer an alternative to a paid API: you can run the adapter on your own servers. That matters to anyone who needs control over their data, fine-tuning for a particular style, or predictable costs at high volume. The source does not name the license terms, so check them separately before any commercial use.

What to consider when choosing

All the data comes from a single Artificial Analysis post. It lists rankings, categories and prices, but gives no specific scores or confidence intervals. That makes it unclear how far fifth place sits from fourth: the gap could be small or substantial.

If your task is animating frames in a live-action film style or building multi-scene narratives, add Hyperflow H3 to your tests alongside Seedance 2.5 and Utopai X. For video editing, it looks like a cost-effective option: third place at $18 per minute. Make the final decision after running it on your own footage, because a leaderboard reflects average quality, not your specific workflow.