Overview

FluxesCore-Dev series models are fine-tuned to the Flux.1-Dev models.

Since the original Flux.1 model did not have good Asian facial features, we are training it to have better Asian facial features.

Settings

About this version (fp16)

Recommended Settings

• WebUI Platform : ComfyUI
• ModelSamplingFluxMAX_SHIFT : 1.0 - 1.1 (Recommended : 1.05)BASE_SHIFT : 0.0 - 0.1 (Recommended : 0.05)
• FluxGuidance : 2.5 - 3.3 (Recommended : 3.0)
• Steps : 20
• CFG Scale : 1.0
• Sampler : euler
• Scheduler : normal
• Denoising : 0.8 - 1.0 (Recommended : 1.0)
• Resolution : 1024 x 1536

NOTE

• Due to the characteristics of Flux.1, it is not a Danbooru tag-based study, so we recommend specifying in English.
• When outputting photorealistic or illustrative styles, the following prefixes are recommended.

e.g.)
・How to outputting photorealistic style [ Photo of a～ ]
Photo of a japanese woman wearing beige sweater and skinny jeans pants.

・How to outputting illustrative style [ Anime of a～ / Anime style of a～]
Anime style of a japanese woman wearing beige sweater and skinny jeans pants.