README.md

metadata

title: FluxM-Guided Lightning Upscaler
emoji: ⚡
colorFrom: green
colorTo: indigo
sdk: gradio
sdk_version: 4.44.1
app_file: app_v4.py
pinned: true
license: other
tags:
  - upscaler
  - super-resolution
  - controlnet
  - flux.1-dev
  - flux.1-schnell
  - flux.1-merged
license_name: flux-1-dev-non-commercial-license
license_link: https://huggingface.co/black-forest-labs/FLUX.1-dev/blob/main/LICENSE.
short_description: Lightning fast guided upscaling with FLUX.

FLUX.1 Merged & Fused: Lightning Upscaler and Detailer

A high-performance image upscaling application built with FLUX.1 models, hosted on Hugging Face Spaces.

Core Components

• Framework: Gradio (v4.44.1)
• Main Model: FLUX.1M-8step_upscaler-cnet
• Text Encoder: T5EncoderModel from FLUX.1-merged_uncensored
• Vision Model: Moondream for image captioning

Key Features

1. Image Upscaling

   • ControlNet-based upscaling
   • Scale factor: 1-3x
   • 8-step inference for speed
   • Memory-optimized with xFormers

2. Auto-Captioning

   • Uses Moondream for image analysis
   • Generates detailed image descriptions
   • Focus area specification

3. Performance Optimizations

   • Attention slicing
   • Memory-efficient attention
   • BFloat16 precision
   • GPU acceleration

Environment Requirements

• PyTorch 2.4.0
• CUDA support
• Hugging Face token for model access
• Moondream API key

Usage

1. Upload control image
2. (Optional) Enter custom prompt or use auto-caption
3. Adjust parameters:
   • Scale (1-3x)
   • Steps (2-16)
   • ControlNet scale (0-1)
   • Guidance scale (1-30)
   • Seed (0-1000000)

License

Non-commercial license (FLUX.1-dev)