.jpeg)
# Comprehensive Guide to Using the Illustrious Stable Diffusion Model
The Illustrious model, developed by OnomaAI Research, is a powerful Stable Diffusion XL-based tool for crafting high-quality illustrations and animations. Available on [Civitai](https://civitai.com/models/795765/illustrious-xl), it supports both natural language and tag-based prompts, making it approachable for beginners. This guide will take you step-by-step through the process of creating amazing images—from imagining your idea to generating it with the Illustrious model. Whether you're new to AI art or looking to refine your skills, you'll find everything you need here to get started.
---
## From Imagination to Prompt: How to Describe Your Vision
Before you type a single word into the Illustrious model, the first step is turning the picture in your mind into a clear, detailed description—known as a "prompt." This is the foundation of AI image generation: the better you describe your vision, the closer the output will match what you imagine. Don’t worry if this feels tricky at first—it’s a skill that improves with practice. Here’s how to do it:
### Steps to Build an Effective Prompt
1. Visualize the Scene
Take a moment to close your eyes and picture what you want to create. Is it a character, a landscape, or something abstract? Let the details come into focus in your mind.
2. Identify Key Elements
Break your mental image into its main parts:
- For a character: gender, hair color, clothing, pose, expression, etc.
- For a landscape: terrain, weather, time of day, notable features like trees or rivers.
3. Add Details
Make your description vivid with specific adjectives. Instead of "a tree," try "a tall pine tree with drooping branches covered in snow." The more detail you add, the more the AI can bring your idea to life.
4. Consider the Mood and Style
Think about the feeling you want to convey—calm, exciting, mysterious? Then decide on an artistic style, like realistic, anime, or watercolor. These choices shape the tone of your image.
5. Structure the Prompt
Put it all together in a logical order:
- Start with quality terms (e.g., "masterpiece," "best quality").
- Describe the subject.
- Add the setting, actions, and styles.
For example:
masterpiece, best quality, 1girl, long silver hair, blue eyes, wearing a flowing white dress, standing in a moonlit forest, ethereal glow, fantasy art style
6. Use Emphasis Wisely
If some parts matter more to you, highlight them with parentheses and a weight, like (ethereal glow:1.2). This tells the AI to focus on those elements, but don’t overdo it—keep it simple to start.
### Example: Building a Prompt Step by Step
Imagine a scene: "A dragon flying over a castle." Let’s turn that into a detailed prompt:
- Basic Idea: "A dragon flying over a castle."
- Add Details: "A huge green dragon with shimmering scales, soaring above a gray stone castle with pointed towers."
- Include Setting and Time: "The castle sits on a rocky hill, surrounded by rolling fields, under a golden sunset sky."
- Consider Style: "In a detailed fantasy style, with dramatic lighting."
- Final Prompt:
masterpiece, best quality, a huge green dragon with shimmering scales, soaring above a gray stone castle with pointed towers, on a rocky hill surrounded by rolling fields, golden sunset sky, detailed fantasy style, dramatic lighting
This process helps you go from a vague idea to a precise description the Illustrious model can use to create your vision. Feel free to tweak your prompt after the first try—iteration is part of the fun!
---
## How to Structure Prompts
Prompts are your instructions to the AI, so structuring them well is key to getting great results. Here’s a breakdown of the components to include:
### Quality Descriptors
Kick off your prompt with words that boost the image’s quality:
- masterpiece
- best quality
- high resolution
- very detailed
These set the standard for a polished output.
### Subject Details
Describe what’s at the heart of your image:
- For characters: 1boy, short brown hair, wearing a leather jacket, smiling.
- For objects or scenes: golden crown, old wooden bridge.
### Scene Descriptions
Add context with the setting and actions:
- Where: on a beach, in a dark cave.
- What’s happening: running, holding a lantern.
- Background: stormy clouds, twinkling stars.
### Styles and Effects
Shape the look with artistic choices:
- cartoon style
- realistic lighting
- soft pastel colors
- sharp outlines
### Emphasis and Weighting
Use parentheses to prioritize parts of your prompt:
- (smiling:1.1) gives a bit more focus to the smile.
- Keep weights between 1.0 and 1.2 for subtle emphasis—higher numbers can overdo it.
Example Prompt:
best quality, high resolution, 1boy, short brown hair, wearing a leather jacket, standing on a beach, holding a lantern, stormy clouds, realistic lighting
---
## Using Negative Prompts
Negative prompts tell the AI what not to include, helping you avoid common problems like blurry images or weird hands. Think of them as your cleanup crew.
### Common Negative Prompts
- Quality fixes: low quality, blurry, pixelated
- Anatomy fixes: bad hands, extra fingers, distorted face
- Unwanted stuff: watermark, text, logo
- Content filters: gore, nudity (if you don’t want them)
Example Negative Prompt:
Negative prompt: low quality, blurry, bad hands, extra fingers, watermark, text, logo, distorted face
Add this to every generation to keep your images clean and on track.
---
## Applying LoRAs for Customization
LoRAs (Low-Rank Adaptations) are like add-ons that let you tweak the Illustrious model with unique styles or themes—like turning your image into a gothic painting or a comic book panel. They’re optional but super fun to play with.
### How to Use LoRAs in Forge Web UI
1. Grab a LoRA: Download one from [Civitai](https://civitai.com/models/1065247/ag4jj9q9w0aigoeio?modelVersionId=1195582) (e.g., comic_style_illustriousXL).
2. Add it to Forge: Put the file in the models/lora folder.
3. Activate it: In Forge, click the purple "Extra Networks" icon under "Generate," go to the "Lora" tab, pick your LoRA, and it’ll add to your prompt.
4. Set the weight: Adjust it in the prompt, e.g., <lora:comic_style_illustriousXL:0.7>. Start with 0.7 and tweak up or down.
Check out [Civitai’s LoRA guide](https://github.com/civitai/civitai/wiki/How-to-use-models#lora) for more details.
---
## Recommended User Interfaces
The interface you use can make a big difference. Here are two great options for beginners:
- Forge Web UI: Perfect for newbies, Forge is a fast, user-friendly version of AUTOMATIC1111 Web UI. It works well even on GPUs with 6GB or 8GB of VRAM. Get started with [this installation guide](https://stable-diffusion-art.com/sd-forge-install/).
- Comfy Web UI: A bit more advanced, Comfy uses a node-based system for custom workflows. It’s great once you’re comfortable with the basics—try [this guide](https://stable-diffusion-art.com/comfyui/) to learn it.
---
## Tips for Beginners
- Start Small: Try a simple prompt first, like best quality, 1girl, standing in a park, then build from there.
- Play with LoRAs: Experiment with different styles using low weights (0.3–0.7) to see what you like.
- Always Use Negative Prompts: They’re your safety net against odd results.
- Tweak and Retry: If the image isn’t quite right, adjust a few words and generate again.
- Save Your Work: Keep a list of prompts and their results to learn what works.
- Have Fun: Don’t stress about perfection—explore and enjoy the process!
---
## Troubleshooting Common Issues
Ran into a snag? Here’s how to fix common problems:
- Blurry Images: Bump up the step count (try 30–50) or switch samplers (e.g., Euler a to DPM++ 2M).
- Weird Hands or Faces: Add details like detailed hands or use negative prompts like bad anatomy.
- Stuff You Don’t Want: List it in the negative prompt, e.g., extra arms, logo.
---
### Token Count and Quality
When crafting prompts for the Illustrious model, it's important to be mindful of the number of tokens used. Tokens are the basic units that the model processes, and there's a limit of about 75-77 tokens per prompt. Exceeding this limit means parts of your prompt might be ignored, which can affect the quality of the generated image.
Using fewer tokens can make the process more efficient, as it uses less memory. This efficiency can indirectly lead to better image quality because the model can focus on fewer, more detailed elements. For beginners, it's a good practice to keep prompts concise. You can use online tools to estimate your token count, but remember that Stable Diffusion uses a specific tokenizer, so these tools provide an approximation.
Practical Tip: Start with a simple prompt and gradually add details. If your prompt is too long, try to rephrase it or remove less important elements to stay within the token limit.
---
### VRAM and Memory Usage
VRAM (Video Random Access Memory) is a critical resource for generating images with the Illustrious model. It stores the model and the data needed during the generation process. Typically, you need at least 4-6 GB of VRAM to run Stable Diffusion smoothly, with 8 GB being ideal for higher resolutions.
If your GPU has limited VRAM, you might encounter memory errors, especially when trying to generate large images. To optimize for lower VRAM, you can use settings like --medvram in Forge Web UI. This setting splits the model into smaller parts, reducing VRAM usage but potentially slowing down the generation process.
Practical Tip: If you're using a GPU with 6 GB or less VRAM, enable the --medvram option in Forge Web UI to avoid memory issues. For best performance, consider generating smaller images and upscaling them later.
---
### Upscaling Images with SD Ultimate Upscaler
To create larger versions of your generated images, you can use the SD Ultimate Upscaler in Forge Web UI. This tool allows you to increase the resolution of your image while maintaining quality.
Steps to Use SD Ultimate Upscaler:
1. Generate your image using the txt2img tab.
2. Move the generated image to the img2img tab.
3. In the img2img tab, select the "Ultimate SD Upscale" script from the dropdown menu.
4. Set the "Scale by" option to 2 to double the image size.
5. Ensure the tile sizes (width and height) match the original image size. For example, if your image is 512x512, set the tile width and height to 512.
6. Adjust the denoising strength to around 0.35-0.5 to balance quality and detail.
7. Click "Generate" to start the upscaling process.
This method breaks the image into four parts (since you're scaling by 2x), processing each tile separately to manage memory and reduce seams.
Practical Tip: Start with a smaller image (e.g., 512x512) and upscale it to avoid memory issues. Experiment with denoising strength to find the best balance for your image.
---
### Enhancing Details with ADetailer in Forge
The ADetailer plugin in Forge Web UI is a powerful tool for improving specific parts of your image, such as faces or hands. It automates the inpainting process, allowing you to enhance details without manual editing.
Steps to Use ADetailer:
1. Install the ADetailer plugin via the Extensions tab in Forge Web UI. You can find it on [GitHub](https://github.com/Bing-su/adetailer).
2. Once installed, go to the img2img tab.
3. Upload your image and select the area you want to enhance (e.g., face).
4. Choose a detection model, such as face_yolov8n.pt for faces.
5. Adjust settings like confidence level (e.g., 0.3), mask blur (e.g., 4), and denoising strength (e.g., 0.4).
6. Click "Generate" to apply the enhancements.
ADetailer detects the specified areas and refines them, improving details and fixing common issues like distorted faces or hands.
Practical Tip: Use ADetailer after generating your image to refine specific areas. Start with default settings and adjust as needed for the best results.
## Conclusion
The Illustrious model is an incredible tool for turning your imagination into beautiful illustrations and animations. With this guide, you’ve got everything you need to start: a way to describe your ideas, tips for crafting prompts, and tools like negative prompts and LoRAs to refine your results. Pick a simple idea, try it out, and don’t be afraid to experiment—you’ll be amazed at what you can create with a little practice. Happy generating!