bitflow/video-super-resolution-rife-pro

Super video quality enhancement featuring fast upscaling with TensorRT and frame interpolation with RIFE.

Public
186 runs
Playground API Examples README Versions

Examples

View more

Run time and cost

This model runs on Nvidia L40S GPU hardware. We don't yet have enough runs of this model to provide performance information.

Readme

🚀 bitflow: The Ultimate 4K/60fps Cinema Engine

Replicate

Stop using “Plastic” upscalers. Most video enhancers just sharpen the noise.

bitflow is a high-performance reconstruction engine that transforms low-fidelity 15fps footage into Cinema-grade 4K 60fps masterpieces using NVIDIA L40S power and TensorRT 10.x optimization.

⚡ Why bitflow? (The “Topaz” Killer)

While industry standards focus on generic sharpening, bitflow is engineered for creators who need speed without compromising soul.

  • True Reconstruction: Using the FoolhardyRemacri, RealESRGAN, Swin2SR or AnimeSharp model to eliminate “plastic” AI artifacts and restore organic textures.
  • Insane Speed: Optimized for L40S / A100 architectures. Experience processing speeds up to 5x faster than local consumer-grade software.
  • Storytelling & Viral Content Specialist: Perfect for Story-based Shorts, AI Avatars, and Commentary videos. Our RIFE v4.7 integration provides the smoothest motion interpolation for low-framerate sources.
  • Massive Cost Savings: Get Pro-level results at a fraction of the cost of subscription-based desktop tools.

🛠️ The bitflow Pipeline

bitflow employs a sophisticated multi-stage pipeline to ensure stability and quality:

  1. AI Orientation Sense: Automatically detects and adapts to Landscape, Portrait, or Square formats.
  2. Audio-Safe Extraction: Maintains frame-accurate audio sync throughout the entire process.
  3. RIFE v4.7 Fluid Motion (Optional): When target_fps is 30 or 60, RIFE interpolates at source resolution for maximum VRAM efficiency. Use original to keep source FPS and skip interpolation.
  4. Fast Mode (Smart Scaling): For 720pm 1080p sources, we use a Smart Downscale → 4x Upscale strategy for pixel-perfect 4K output.
  5. TensorRT Reconstruction: FoolhardyRemacri, RealESRGAN, Swin2SR or AnimeSharp engines rebuild every pixel from the ground up.

🎛️ API Parameters

Parameter Type Default Pro-Tip
video File (required) Your raw source footage.
target_resolution string 4K Go for 4K. The L40S handles it with ease.
target_fps string original Choose 30/60 for smoothness, or original to keep original FPS and save time.
upscale_model string general general: Realistic content (people, animals, landscapes).
anime: 2D/cel-shaded content. Scale (2x/4x) is auto-selected.
aspect_ratio_mode string fill Use fill for edge-to-edge cinematic immersion.
max_seconds float 20.0 Duration to process (1-120s). Auto-capped to ~30s when interpolation is active.
fast_mode bool false Faster processing with slightly reduced quality. Downscales before upscaling.

💎 Which model should I choose?

  • general — Best for real-world footage: people, animals, landscapes, architecture.
  • anime — Optimized for 2D animation, cel-shaded 3D, and illustrated content. Cleans compression noise while preserving clean outlines.

💰 Value Comparison

Feature Legacy Desktop Tools bitflow (L40S Cloud)
Cost (4K/60fps) High Subscription/Per-use Pay-as-you-go (Fractional Cost)
Processing Speed Hardware Limited TensorRT Accelerated (L40S)
Interpolation Basic / Generic RIFE v4.7 (State-of-the-Art)
Animation/Shorts Support Moderate Dedicated Low-FPS Tuning
  • When converting 540p/15 FPS video to 4K/60 FPS, it offers the best value for money.

💎 Creator’s Guide

  • Viral Storytelling: Transform 15fps Commentary/Story videos into 4K/60fps. High-quality visuals trigger better social media distribution.
  • Anime Remastering: anime model doesn’t just upscale; it cleans compression noise and re-draws clean line art.
  • Efficient Testing: Use the start_time parameter to render a 5-second highlight before committing to a full-length upscale.
  • For better quality and processing speed: We guarantee 100% quality for videos under 25 seconds. For videos longer than 25 seconds, there is a trade-off between quality and processing time. If you want lossless, high-quality results, please convert a 24-second video multiple times and visit video-concat-safe-pro to merge the clips.

📜 Acknowledgments

bitflow automatically selects models optimized for processing user video input and builds upon the achievements of its predecessors.

This pipeline integrates the following:

  • RIFE interpolation by hzwer

  • Foolhardy Remacri, Anime Sharp by Kim2091

  • Real ESRGAN by Xintao Wang, Liangbin Xie, Chao Dong, Ying Shan

  • SPAN - ESR by hongyuanyu

  • TensorRT by NVIDIA

⚙️ Technical Specifications

  • Compute Target: Native optimization for NVIDIA L40S (48GB VRAM).
  • Acceleration: Leveraging TensorRT 10.x to minimize inference latency.
Model created
Model updated