FLUX 3 Max

fal, in collaboration with Black Forest Labs, has unveiled an accelerated version of the FLUX 3 Max video model. This is a fine-tuned and radically optimized variant of the original video generation model.

The new model's performance significantly surpasses initial benchmarks. Tests demonstrate a substantial reduction in rendering time while maintaining high content quality.

Key Technical Specifications and Comparative Data

• Generating a 10-second video takes just 10.95 seconds, which is 11.5 times faster than the original (126.44 s).
• The model operates approximately 22 times faster than Seedance 2.5 at comparable output quality.
• It holds an Elo rating of 1139 on Artificial Analysis, matching the level of market leaders.

This acceleration was achieved through a comprehensive development approach. The process includes additional model training using reinforcement learning, computational optimization, and enhanced server infrastructure.

This solution improved object motion quality, image detail, and adherence to text prompts without sacrificing inference speed.

Technical Capabilities and Working Formats

• Support for Text-to-Video and Image-to-Video modes.
• Generated clips can be up to 20 seconds long.
• Resolution of 720p, with upscaling capability to 1080p.
• Automatic sound generation feature.

Access to the model is exclusively available via the fal platform API. The source weights are closed-source, and request processing is handled on the provider's servers.

Usage costs $0.0425 per second of video at 720p resolution. A 10-second clip will cost $0.425. A temporary 75% discount is available until November 7.

The launch of FLUX 3 Max demonstrates a new trend in the AI market. The platform is shifting from simple model aggregation to creating proprietary products by fine-tuning base architectures.

This strategy allows for significantly higher inference speeds and offers users more efficient tools for media content creation.