Black Forest Labs Ships FLUX Upscale: 4K Video Enhancement with Dual Modes
Black Forest Labs has released FLUX Upscale as a dedicated API endpoint and tool, enabling creators and production studios to regenerate AI video up to native 4K resolution with distinct Precise and Creative fidelity options.
What’s New
- FLUX Upscale takes video inputs from 480p upward and regenerates them at up to 4K resolution.
- Supports 1.5x, 2x, and 3x scaling factors, targeting standard 1080p, 2K, and 4K outputs.
- Natively understands FLUX diffusion output to correct smudged facial features and grid artifacts on complex textures.
- Precise mode runs in 4 steps at $0.07 per megapixel per second to maximize subject consistency.
- Creative mode runs in 8 steps at $0.10 per megapixel per second for advanced texture synthesis and detail recovery.
Why It Matters
FLUX Upscale bridges the gap between raw AI video prototypes and broadcast-ready delivery. Its dual-mode pricing lets studios trade speed against micro-detail without leaving the FLUX developer ecosystem.
Black Forest Labs has released FLUX Upscale, a standalone enhancement model and API endpoint designed to elevate generated video clips to broadcast, commercial campaign, and cinema specifications. The tool accepts standard definition or high definition footage from 480p upward, reconstructing frames at scale factors of 1.5x, 2x, and 3x to deliver crisp 1080p, 2K, and native 4K video.
Generating video directly at ultra-high resolutions has remained computationally prohibitive for frontier diffusion architectures, forcing studios to rely on post-generation upscalers. However, general-purpose interpolation algorithms often struggle with generative artifacts, blurring delicate facial features or introducing repetitive grid noise across organic surfaces like water, foliage, and fabric.
Native diffusion repair and scaling mechanics
Because FLUX Upscale was trained directly on FLUX foundation models, it understands the latent characteristics and typical error modes of synthetic video. Instead of simply multiplying pixels and applying edge sharpening filters, the upscaler performs a controlled regeneration pass that repairs smudged human features and stabilizes temporal consistency across sequential frames.
The model accepts variable input resolutions and applies scaling multipliers to match standard production delivery targets. A standard 720p draft can be scaled by 1.5x to reach full 1080p for web distribution, or by 3x to generate clean 4K footage suitable for large-format commercial displays.
Two distinct precision and cost modes
To accommodate different production requirements, Black Forest Labs structured FLUX Upscale with two operating modes:
- Precise Mode: Operating in 4 denoising steps at a cost of $0.07 per megapixel per second, Precise mode prioritizes computational speed and strict adherence to the input video. It preserves actor identity, branding elements, and reference objects without introducing creative alterations.
- Creative Mode: Operating in 8 denoising steps at $0.10 per megapixel per second, Creative mode allocates additional compute to synthesize missing high-frequency detail. While it achieves greater texture depth and edge sharpness in complex backgrounds, the higher generative latitude can subtly alter subtle identity nuances.
FLUX Upscale is accessible immediately through the official Black Forest Labs API and interactive playground, offering production teams a predictable pathway from rapid creative exploration to finished 4K deliverables.
