FLUX.1-schnell-dev-merged-fp8
| Property | Value |
|---|---|
| License | Non-commercial |
| Language | English |
| Pipeline | Text-to-Image |
| Format | Safetensors |
What is FLUX.1-schnell-dev-merged-fp8?
FLUX.1-schnell-dev-merged-fp8 is a merged model combining the capabilities of FLUX.1-dev and FLUX.1-schnell, optimized for efficient image generation with FP8 precision. This model represents a significant advancement in fast inference text-to-image generation, requiring only 6-8 steps for quality output.
Implementation Details
The model implements a dual-pass generation system with specific recommended configurations: 6 steps (First Pass) with 2-3 Steps (Second Pass), or 8 steps (First Pass) with 2 Steps (Second Pass). It supports multiple schedulers including Normal, sgm_uniform, simple, and beta, with ddim_uniform showing particular strength in illustrated styles.
- Optimal Inference Steps: 6-8
- Recommended Guidance Scale: 3.5
- Primary Sampler: Euler
- Multiple Scheduler Support
Core Capabilities
- Fast, efficient image generation with minimal steps
- Dual-pass refinement system for noise reduction
- ComfyUI compatibility with provided workflows
- Support for various scheduling algorithms
- Optimized FP8 precision for efficient processing
Frequently Asked Questions
Q: What makes this model unique?
The model's distinctive feature is its ability to generate high-quality images in significantly fewer steps than traditional models, utilizing a dual-pass system for enhanced refinement.
Q: What are the recommended use cases?
This model is ideal for applications requiring quick image generation while maintaining quality, particularly suited for ComfyUI workflows. It's specifically optimized for general text-to-image generation tasks with particular strength in illustrated styles when using ddim_uniform scheduler.
