Overview

This model is developed based on the architecture of FLUX.1-dev, incorporating several structural modifications. It is trained on a custom dataset containing both high-quality anime and realistic images sourced from the internet. The dataset is initially captioned by ChatGPT o3-mini and then manually rechecked to ensure accuracy. Throughout training, the model utilizes FLUX.1-dev as a teacher for guidance.

Important

• As this is a self-reconstructed model, I highly value any feedback you can provide. Your insights will not only motivate me but also help me better understand its strengths and weaknesses for future improvements.

• This is an AIO model (All-In-One) that includes all essential components: VAE, CLIP, and T5Encoder. I recommend using the default settings. Additionally, the model utilizes the bf16 data type, ensuring optimal quality in image generation.

Currently, the model is not available for use via Civitai Generation. You can visit the following website to use it:

• tensor.art

Recommended Settings:

• Guidance Scale: 2.0 - 5.0

• Steps: 20 - 30

• Sampler: Euler a / DPM++ 2M

• Schedule: Simple

Notes

• No post-processing or LoRA was applied to enhance the example images. They were generated solely using the recommended settings and a custom prompt with my base model.

• The example images were created using prompts from various sources and authors to provide an independent evaluation of the model's performance.

Acknowledgments

• Thanks to Black Forest Labs for providing an amazing model!

If you'd like to support my work, you can do so through Ko-fi!