Equally Stunning in Light or Heavy Makeup — A Gen-Z Portrait Model Pursuing Film Aesthetics and Ultimate Texture

Model Introduction

BEYOND REALITY Z IMAGE 1.0 is a checkpoint model fine-tuned from the community Z IMAGE model and merged with LoRAs, continuing the series' tradition of high clarity and high aesthetics.

How to Use

Recommended parameters for optimal results:

Sampler: euler+simple
Steps: 10-15
CFG Scale: 2
Resolution: 1024–2048
Clip Skip: 2
Negative Prompt: Add based on your needs; avoiding overly heavy negative prompts is recommended to preserve the film texture.

We have always believed: Beauty cannot be defined; it is felt. Just like the light, shadow, and color of film photography—wisdom accumulated over a century by countlessly dedicated imaging engineers—it cannot be quantified, nor will it resonate with everyone. But if it meets someone who understands, that is a true blessing.

Addressing Z IMAGE's shortcomings in texture detail representation, this update focuses on:

• Deep optimization of portrait skin textures and environmental details: Significantly enhances high-frequency detail performance, making skin pores, fabric textures, wall surfaces, and other details much more realistic and believable.
• Film photography aesthetic alignment: The overall style leans slightly toward film portraits, giving images richer color gradation and a breathing sense of light and shadow.
• Balance between realism and aesthetics: Elevates the aesthetic height of generated portraits while retaining the impressive photorealism of the Gen-Z series models.

Training was primarily based on synthetic high-quality portrait photography-style images, enabling the model to achieve a balanced aesthetic where it is "equally stunning in light or heavy makeup."

Note: Due to the characteristics of the training data, complex body poses and certain non-portrait textures might be affected. These will be continuously optimized in future releases.

Perhaps the best portrait photography Z IMAGE model available right now.

Please pay attention to the parameter settings at the bottom.

Z IMAGE has brought plenty of surprises to everyone with its blazing speed and ultra-realistic visuals, and community demand for edit workflows has been continuous.

BEYOND REALITY Z IMAGE 1.0 — "Equally Stunning in Light or Heavy Makeup" — is a checkpoint model created by fine-tuning and merging LoRAs trained on portrait datasets. It continues the signature characteristics of the BEYOND REALITY series: high clarity and high aesthetics. One of Z IMAGE's current issues is its poor performance with high-frequency information, which manifests as weak texture details in generated images. Therefore, I trained a LoRA to improve the model's texture quality. I originally intended to release just the LoRA, but considering I've always released checkpoints, I decided to stick with that habit.

I have always favored the light, shadow, and color of film photography. It is the crystallization of aesthetics built upon a century of color-grading experience by countless imaging engineers, serving as the best and most widely accepted medium for capturing light, shadow, and beauty through images. Beauty is not defined; it is felt. Anything that can be strictly defined or quantified is bound to descend into mediocrity. Thus, beauty cannot be quantified or measured; it can only be understood and felt. Nor can beauty resonate with everyone—to find even one person who understands you is already a rare fortune.

This update primarily optimizes skin textures and environmental details for portraits in Z IMAGE, while also elevating the overall aesthetic quality of portrait generations. Z itself generates images with extremely high realism—in some cases, making it impossible to tell whether an image is AI-generated or a real photograph. However, when it comes to aesthetics, Z still had room for improvement.

The overall style of this updated model leans slightly toward film portrait photography. It can generate beautiful portrait images as well as elegant environmental details, all while retaining much of Z's realism. Guided by prompts, it can produce incredibly realistic photos.

Note: Since this training mainly utilized synthetic high-quality portrait photography-style images, while elevating overall aesthetics and texture details, it may affect complex body poses and certain non-portrait textures. Future versions will continue to optimize this.

Feel free to try it out, and please share your generated images!

This release includes both BF16 and FP8 versions. The FP8 version runs smoothly on 8GB VRAM (tested and verified on a laptop).

Recommended parameters: euler+simple, 10–15 steps, CFG: 2

BEYOND REALITY, I'm back.

Model Overview (English)

BEYOND REALITY Z IMAGE 1.0 is a checkpoint model fine-tuned from the community Z IMAGE model, merging LoRA weights to preserve the signature high-clarity, high-aesthetic quality of the BEYOND REALITY series.

We believe: Beauty cannot be defined, only felt. Like the color science of analog photography—refined over a century by imaging engineers—true aesthetic lives in intuition, not metrics. It may not resonate with everyone, but finding one who understands is already a rare fortune.

This update addresses Z IMAGE's weakness in high-frequency detail representation:

• Enhanced texture fidelity: Skin pores, fabric weaves, and environmental surfaces now exhibit unprecedented realism
• Analog photography alignment: A subtle film-photography tonality enriches color grading and lighting atmosphere
• Balance of realism & aesthetics: Maintains Z's photorealistic power while elevating portrait beauty to new heights

Trained primarily on synthetic high-quality portrait photography, the model achieves a versatile "suited for both heavy makeup and bare-faced" elegance.

Note: Due to training data focus, complex poses and certain non-human textures may be affected. Continuous improvements planned.

Model Information

Item	Details
Base Model	Z IMAGE
Version	1.0
Training Method	LoRA Fine-tuning + Checkpoint Merge
Key Optimizations	Skin Texture, Environmental Details, Film Aesthetics
Recommended Resolution	1024–2048
Applicable Style	Portrait Photography, Film Texture, Everyday Scenes
License	Apache License 2.0

Recommended Use Cases

• High-aesthetic portrait generation and aesthetic alignment
• Fine representation of skin texture and material details
• AI creative simulation of film photography style
• Texture enhancement for social media visual content

More Examples