Krea Adds MiniMax H3 Max for 50x Faster Video Generation

Krea has added MiniMax H3 Max to its video generation toolset. This speed-tuned variant of the frontier MiniMax H3 model, which is the successor to Hailuo 2.3, can generate up to 15 seconds of video in roughly five seconds. Co-optimized with the inference stack from fal, the model can render a five-second clip in under three seconds. To achieve this 50-fold speedup compared to its peers, the model trades resolution for throughput, outputting video at 480p or 768p instead of the 2K resolution delivered by the base H3 model.

The model is post-tuned on H3 weights to improve prompt adherence and aesthetics. It supports text-to-video and image-to-video generation with an optional end frame. However, users must sacrifice the native stereo audio and multi-reference workflows of the flagship H3 model, which holds the top three positions on the Artificial Analysis Video Arena leaderboards for video editing, text-to-video, and image-to-video with audio. This makes the Max variant function essentially as a high-speed draft mode.

For practitioners, this rapid generation loop transforms the creative process. Instead of waiting minutes for a single 2K render, creators can test 10 prompt variations in under a minute to debug motion, composition, or camera angles. Once a desirable direction is established, they can render the final version using slower, high-fidelity models like the base H3, Seedance 2.5, Kling 3.0, Veo 3.1 Fast, or LTX-2.5.

MiniMax H3 Max is available on Krea, which offers a free tier with 100 daily units. Paid plans start at $9 per month for the Basic tier with 5,000 units, scaling to $35 for Pro with 20,000 units, $105 for Max with 60,000 units, and a Business tier with 80,000 units, with a 40 percent discount for annual commitments. Alternatively, developers can test the model directly on fal.ai with five free daily generations.