improved-flux-prompts-photoreal-portrait

## Photo Portrait Prompt Dataset for FLUX

## Overview
This dataset contains a curated collection of prompts specifically designed for generating photo portraits using FLUX.1, an advanced text-to-image model. These prompts are crafted to produce high-quality, lifelike portraits by leveraging sophisticated prompting techniques and best practices.

## Latest Version
Improved on October 3, 2024.

This version has undergone curation and improvement. What is new?

- Cleaned up the prompt dataset by removing highly repeated terms.
- Provided a more diverse range of ethnicities and styles.
- Rephrased some prompts to make them more understandable and engaging.
- Included photographic medium terms to help guide the photographic style.
- Made minor changes to some prompts to make them more consistent with the rest of the dataset.
- Added a double exposure effect to certain prompts.

We continue to curate the dataset.

## Example Results
We've put together sample output grids to give you a visual representation of what the dataset can do.









Every image in the grid features a tag that displays its corresponding prompt ID, making it easy to reference.

## Improve Your Output Using Lora.
Take note that, while FLUX.1 Dev/Schnell generates female faces using a basic key, Lora can assist in introducing variance in facial features, making your outputs even more original and diverse.
An excellent Lora that can bring more variety to female face features is [SameFaceFix](https://civitai.com/models/766608/sameface-fix-flux-lora). Another great Lora to experiment with is [Amateur Photography](https://civitai.com/models/652699/amateur-photography-flux-dev). The dataset that we're proposing here is a fantastic resource for experimenting with different generation settings!

## Dataset Creation Process
The dataset was developed using a multi-step approach:

1. **Base Examples**: We started with a selection of top-performing prompts from various sources.

2. **LLM Enhancement**: These prompts were refined and expanded using the Language Model (LLM).

3. **FLUX Optimization**: The prompts were further tailored for FLUX.1, incorporating best practices and techniques known to produce outstanding results with this model.

## Prompt Features
Each prompt in this dataset is designed to include:

- Detailed descriptions of facial features and expressions
- Specific lighting setups for portrait photography
- Guidance on pose and composition
- Skin texture and tone details
- Hair styling and color information
- Background and environment descriptions
- Photographic style references (e.g., studio, environmental, candid)

## Data Quality Note
While efforts have been made to refine and enhance the prompts for FLUX.1, users should be aware that some inconsistencies may still exist. Initial results have been promising, but users are encouraged to fine-tune prompts as needed for their specific requirements.

## System Message for LLM
The Hermes3 LLM was guided by a carefully crafted system message to ensure high-quality, FLUX-optimized prompts for photo portraits. Key points of this system message include:

- Guidelines for creating clear and comprehensive descriptions of portrait subjects
- Focus on photographic details instead of artistic interpretations.
- Recommendations for including key elements such as facial features, lighting, pose, and style
- Strategies for developing impactful prompts, including the use of photography terminology and portrait composition techniques

The complete system message can be found in the `flux_photo_portrait_system_message.txt` file in this repository.

## Dataset Structure
The dataset is organized in a JSONL (JSON Lines) format, with each line representing an individual prompt. Each entry includes:

- `id`: A unique identifier for the prompt
- `prompt`: The specific text of the prompt

## How to Use
This dataset is designed for experimental use with FLUX.1 and similar text-to-image models for creating photo portraits. Researchers and developers can use these prompts to:

1. Generate high-quality portraits
2. Explore effective prompting techniques for portrait photography
3. Evaluate the model's performance in creating lifelike human faces

## Acknowledgements
We extend our gratitude to Black Forest Labs for developing the FLUX.1 model, which makes this project possible. Their work in text-to-image has pushed the boundaries of AI-generated imagery.

We also thank the Flux Reddit community for their valuable insights and discussions on portrait generation techniques. Their shared experiences have greatly informed our approach to prompt engineering for portraits.

## License
MIT

## 适用于FLUX的人像摄影提示词数据集

## 概述
本数据集为经精选整理的提示词集合，专为使用FLUX.1（一款先进的文本转图像模型）生成人像摄影作品而设计。这些提示词通过运用成熟的提示工程技巧与行业最佳实践，旨在生成高质量、栩栩如生的人像作品。

## 最新版本
本数据集于2024年10月3日完成更新。本次更新对数据集进行了精选与优化，新增内容如下：
- 清理提示词数据集，移除大量重复术语
- 提供更多元化的种族与风格选项
- 重新措辞部分提示词，使其更易懂且更具吸引力
- 加入摄影媒介术语，辅助引导摄影风格
- 对部分提示词进行小幅调整，使其与数据集整体风格更统一
- 为部分提示词添加双重曝光效果
我们仍在持续对该数据集进行精选优化。

## 示例效果
我们整理了样本输出网格图，以直观展示该数据集可实现的生成效果。
每张网格图中的图像均带有对应提示词ID的标签，便于快速引用检索。

## 使用LoRA提升生成效果
请注意，尽管FLUX.1 Dev/Schnell可通过基础关键词生成女性面部图像，但LoRA（Low-Rank Adaptation，Lora）可辅助实现面部特征的多样化，让生成结果更具原创性与多样性。
两款效果出色的LoRA可用于丰富女性面部特征多样性：[SameFaceFix](https://civitai.com/models/766608/sameface-fix-flux-lora)。另一款值得尝试的LoRA为[Amateur Photography](https://civitai.com/models/652699/amateur-photography-flux-dev)。本数据集是尝试不同生成设置的绝佳资源！

## 数据集构建流程
本数据集采用多阶段流程开发：
1. **基础示例**：从各类来源中筛选出表现优异的提示词作为初始基础。
2. **大语言模型（Large Language Model，LLM）优化**：借助大语言模型对这些提示词进行细化与扩展。
3. **FLUX.1适配优化**：进一步针对FLUX.1模型调整提示词，融入已知可在该模型上获得出色效果的最佳实践与技巧。

## 提示词特征
本数据集的每条提示词均包含以下核心元素：
- 面部特征与表情的详细描述
- 人像摄影专用的特定布光方案
- 姿势与构图指引
- 皮肤纹理与色调细节
- 发型与发色信息
- 背景与环境描述
- 摄影风格参考（例如影棚、环境、纪实抓拍风格）

## 数据质量说明
尽管已尽力为FLUX.1模型优化与完善提示词，但仍需提醒用户，数据集可能仍存在部分不一致之处。初步生成结果已颇具前景，但我们鼓励用户根据自身需求对提示词进行个性化微调。

## 大语言模型系统提示词
我们通过精心设计的系统提示词引导Hermes3大语言模型，以生成适配FLUX的高质量人像提示词。该系统提示词的核心要点包括：
- 制定人像主体清晰且全面的描述准则
- 优先聚焦摄影细节而非艺术诠释
- 推荐纳入面部特征、布光、姿势与风格等关键元素
- 提供打造高效提示词的策略，包括运用摄影术语与人像构图技巧
完整的系统提示词可在本仓库的`flux_photo_portrait_system_message.txt`文件中查看。

## 数据集结构
本数据集采用JSONL（JSON Lines）格式组织，每一行代表一条独立的提示词。每条条目包含以下字段：
- `"id"`：提示词的唯一标识符
- `"prompt"`：提示词的具体文本内容

## 使用方法
本数据集专为使用FLUX.1及同类文本转图像模型生成人像摄影作品的实验场景设计。研究人员与开发者可利用这些提示词完成以下工作：
1. 生成高质量人像作品
2. 探索人像摄影领域的高效提示工程技巧
3. 评估模型在生成逼真人脸方面的性能表现

## 致谢
我们谨向Black Forest Labs致以诚挚谢意，感谢其开发FLUX.1模型，为本项目提供了坚实基础。该团队在文本转图像领域的工作推动了AI生成图像技术的边界拓展。
我们同时感谢Flux Reddit社区，感谢其在人像生成技巧方面的宝贵见解与讨论。社区成员的共享经验极大地启发了我们的人像提示工程方案。

## 许可证
MIT许可证

搜集汇总

背景与挑战

背景概述

该数据集专门为FLUX.1文本生成图像模型提供精心设计的肖像照片生成提示词，于2024年10月更新以提升多样性和清晰度。每个提示词包含详细的面部特征、光照设置和摄影风格描述，旨在帮助用户生成高质量、逼真的肖像图像。