FLUX.1 Prompt Guide
Last updated on: Aug 20, 2024
Flux.1 is an open-source AI image generation model launched by Black Forest Labs, a laboratory founded by Robin Rombach, a former core member of Stability AI. Many team members also come from the original Stable Diffusion development team. Flux.1 has set a new benchmark for AI text-to-image models with its exceptional image generation capabilities. It offers three different variant versions to meet diverse user needs. With a training parameter count of up to 12 billion, it far surpasses the 2 billion of SD3 Medium and outperforms other popular models like Midjourney and DALL-E 3 across various key metrics.
The three versions of Flux.1 are:
- FLUX.1 [pro]: A closed-source model offering top-tier performance for commercial use, accessible via the official API, and supports enterprise customization.
- FLUX.1 [dev]: An open-source model, non-commercial, distilled from FLUX.1 [pro], providing similar image quality and prompt adherence capabilities but with greater efficiency.
- FLUX.1 [schnell]: An open-source model for commercial use, tailored for local development and personal use, with the fastest generation speed and minimal memory usage.
Key Features of FLUX.1
- Exceptional Image Generation : Flux.1 excels in producing high-quality images, especially in complex scenarios like generating people in reclining poses without distortions. It outperforms models such as Stable Diffusion 3 and stands competitively against popular models like Midjourney and DALL-E 3.
- High Training Parameters: With a robust training parameter count of 12 billion, Flux.1 significantly exceeds the 2 billion parameters of SD3 Medium, resulting in superior performance in various metrics.
- Advanced Architecture: Flux.1 employs a hybrid architecture combining multimodal and parallel diffusion Transformer modules. Key innovations include flow-matching training methods, rotational positional embeddings, and parallel attention layers, all of which enhance both performance and hardware efficiency.
How to Write Better FLUX.1 Prompts?
To write effective prompts for FLUX.1, it's essential to understand how to structure your prompts, use specific keywords, and customize the desired style and outcome. FLUX.1, like other AI image generation models, relies heavily on well-crafted prompts to produce high-quality images.
In this guide, I'll walk you through the process of crafting prompts by exploring different categories of keywords, their impact on the final image, and how to optimize both positive and negative prompts for better results.
Categories of Keywords
When creating prompts for FLUX.1, you should categorize your keywords to shape the desired image. The categories typically include:
- Topic
- Material
- Style
- Artist
- Webpage Influence
- Sharpness
- Extra Details
- Shade and Color
- Lighting and Brightness
- Negative Prompts
Let's explore each of these categories in detail and see how they impact image generation in FLUX.1.
1. Topic
The topic is the core of your image and defines the main subject or scene. For example, if you want to generate an image of a futuristic warrior, the topic should specify what the warrior looks like, their pose, and their setting.
Example Prompt:
“A futuristic warrior in a cyberpunk city, holding a plasma sword, neon lights reflecting off armor.”
In this prompt, you have:
- Main Subject: Futuristic warrior
- Scene and Setting: Cyberpunk city
- Action and Pose: Holding a plasma sword
Being specific about these details will result in a more accurate and creative image.
2. Material
Material defines the medium or style in which the image is rendered. Flux AI can generate outputs resembling sketches, paintings, 3D renders, or realistic photos.
Example Keywords:
- Digital painting
- 3D render
- Pencil sketch
- Photorealistic
Example Prompt Addition:
“...digital painting, highly detailed, vibrant colors.”
3. Style
Style relates to the overall artistic direction of the image, which can be fantasy, surrealism, anime, or pop art. Combining multiple styles can create unique visual effects.
Example Keywords:
- Fantasy
- Vaporwave
- Steampunk
- Abstract
Example Prompt Addition:
“...digital painting, cyberpunk, vaporwave, futuristic neon colors.”
4. Artist
Referencing specific artists is a powerful way to influence the output’s visual style. If you mention well-known artists like Salvador Dalí, H.R. Giger, or Hayao Miyazaki, the generated image will adopt traits from their iconic styles.
Example Prompt Addition:
“...in the style of H.R. Giger and Salvador Dalí, surreal biomechanical details.”
5. Webpage Influence
Referencing art platforms such as ArtStation, DeviantArt, or Pinterest can also add a particular flair and quality to the generated images. These platforms are associated with high-quality concept art, which can improve the rendering.
Example Prompt Addition:
“...featured on ArtStation, trending on DeviantArt.”
6. Sharpness
Sharpness and clarity are crucial to make your image look polished. Words like “sharp focus” or “highly detailed” can ensure the image has clear lines and defined features.
Example Prompt Addition:
“...sharp focus, highly detailed textures.”
7. Extra Details
These are additional descriptors that enhance the image's atmosphere or make it more visually compelling. Keywords like “cinematic,” “breathtaking,” or “dystopian” can add depth.
Example Prompt Addition:
“...cinematic, atmospheric fog, dystopian ambiance.”
8. Shade and Color
Color and shading play a significant role in setting the mood. You can control the tone using color-related keywords such as “moody,” “warm tones,” or “iridescent.”
Example Prompt Addition:
“...moody lighting, iridescent blue highlights.”
9. Lighting and Brightness
Lighting impacts how the subject is perceived. Including specific lighting styles, like “backlit,” “dramatic shadows,” or “studio lighting,” can help achieve the desired effect.
Example Prompt Addition:
“...dramatic shadows, soft ambient lighting.”
10. Negative Prompts
Negative prompts help avoid unwanted elements by specifying what you don’t want in the image. They can include undesirable styles, colors, or features.
Common Negative Keywords:
- Grainy
- Blurry
- Low contrast
- Deformed anatomy
- Unwanted objects (e.g., watermark, text)
Example Prompt Addition:
“Negative: grainy, blurry, bad anatomy, extra limbs, watermark.”
Crafting an Example Prompt
Let's put everything together into a single comprehensive prompt:
Final Prompt:
“A futuristic warrior in a cyberpunk city, holding a plasma sword, neon lights reflecting off armor, digital painting, highly detailed, vaporwave aesthetic, cinematic lighting, in the style of H.R. Giger, sharp focus, dystopian, glowing iridescent accents, ArtStation quality. Negative: blurry, grainy, deformed, low contrast.”
This prompt carefully combines topic, style, material, artist influence, and extra details while excluding undesired features, resulting in a highly refined and visually appealing image.
Conclusion
Writing prompts for FLUX.1 is all about balancing clarity and creativity. By structuring your prompt into categories like topic, style, material, and adding or removing specific details, you can guide the model to generate images that meet your exact vision.
