F.1基础算法模型-哩布在线可运行 - F.1-dev-fp8

Introduction

The F.1 Basic Algorithm Model is a versatile checkpoint designed for image generation, focusing on providing a stable and efficient base for various creative applications. This model is optimized for performance, offering different versions to suit varying hardware capabilities, from fast, stable outputs to more memory-intensive options. It is built to be adaptable across different UI environments, including standard WEBUI and ComfyUI, allowing for flexible workflow integration.

How to Use

Trigger Word: No trigger word is required to activate this model.

Recommended Settings:

• Steps: 20 (Can be increased for potentially finer detail)
• CFG Scale: 3.5 (Adjustable based on desired adherence to prompt)
• Weight: 0.8

Tips from the Original Description:

• For WEBUI Users:
  • The fp8 version is highly recommended for faster generation speeds and enhanced stability.
  • The fp16 version may load slightly slower and is more prone to VRAM exhaustion.
  • The schnell (turbo) version is generally not recommended due to suboptimal results.
• Sampling Method: Currently, the only supported sampling method is FlowMatchEuler.
• Embedding Compatibility: This model is compatible with SD1.5 embeddings. Using SDXL embeddings may lead to issues, though this is not absolute.
• For ComfyUI Users: This model offers greater combinatorial possibilities and can be extended with various streaming pipeline workflows.
• Hardware Considerations: The regular version of this model requires a minimum of 32GB of system RAM. It has been observed to fully utilize the VRAM of a 4090 GPU. For local execution, the dev-fp8 version is recommended due to its efficiency.

Example Prompts

No specific example prompts were provided for this model.

FAQ

Q: What trigger word should I use?

A: No trigger word is required.

Q: What are the recommended settings?

A: Steps: 20, CFG: 3.5, Weight: 0.8

Q: What resolution works best?

A: While not explicitly stated, common resolutions for this type of model include 1024x1024 for square images and 1024x1536 for portrait images. Be mindful of your VRAM, especially with higher resolutions.