Welcome to this in-depth tutorial on mastering the Diffusitron, an AI image generation tool that allows you to create visually captivating and lifelike character portraits from a blank canvas. Whether you are an artist, a writer, or simply an enthusiast, this guide will help you to transform your ideas into vivid and realistic images with precision and control.
Please keep in mind that this tutorial is based on the experiences and practices of a fellow Diffusitron user, and your results may vary. However, the techniques and insights provided here will undoubtedly help you to better navigate the world of AI-generated art.
Before diving into the actual process, it is crucial to familiarize yourself with some key aspects of stable diffusion:
- It is an iterative, trial-and-error process where a technique that works for one image might not work for another.
- Some details you include may dominate the image, while others might not be visible at all.
- If you have a specific vision in mind, it’s wise to manage your expectations about the likelihood of achieving it exactly as you imagine.
Approaching AI image generation as a game, you will find that, like most games, there is a learning curve. However, with the help of this comprehensive guide, you will soon feel more in control of the images generated, rather than relying solely on chance.
Table of Contents
Unleashing the Power of the Anything-to-Deliberate Model: Crafting Strikingly Realistic Images with an Illustrative Flair.
[You will notice four unique models. For the specific purpose of this guide, we’ll be using models that blend harmoniously. When experimenting with image generation you will develop a general sense of what works best for your intended image. If you are trying to create a stylized illustrated image you will learn that Anything is well suited. If you want an image that is serious and gritty starting with Redshift will deliver. And if you want to unleash Diffusitron to manufacture the most splendidly bizarre and beautiful images, Open Journey is the model to use.]
Now, let’s dive into the Anything Model → Deliberate Model method, which produces strikingly realistic images with an illustrative touch.
Step-by-Step Guide to Crafting Character Portraits:
- Begin with a blank canvas. Do not use positive or negative prompts just yet. As you progress, you will gradually build negative prompts to address issues that arise throughout the process. For a baseline, when aiming for greater realism and addressing common mistakes, use the following negative prompt: “Cartoon, blurry, deformities, disfigurement, disproportionalities, unrealistic”. In the Advanced Settings menu, ensure that both the “Steps” and “CFG Scale” sliders are set to the maximum value.
- In the positive prompt, start with a basic description of your desired image. The order in which these traits are entered doesn’t seem to impact the final result. Consider the following categories when building your prompt:
- Subject: Establish the foundation for your image. Since this tutorial focuses on characters, relevant subjects could be “young woman”, “grizzled old man”, or “reptilian humanoid”.
- Subject Description: Use descriptive words, phrases, or even abstract concepts to enrich your character (e.g., old, beautiful, shiny, wearing specific garments, austere, covered in lobsters, profoundly irritating). The more details you provide, the more intricate and engaging your image will be. However, be mindful that Diffusitron may emphasize certain aspects (colors are notorious for this) or disregard others entirely.
- Background: Incorporate a simple description (e.g., open field, castle, pink and yellow brick wall) to set the scene. Complex background descriptions may sometimes overshadow or replace the subject. If you leave the background to Diffusitron’s discretion, it will typically generate either a backdrop based on the prompt details or a basic monochromatic background.
- Style: Specify an artistic style (e.g., realism, pop art, baroque), a genre that influences details (e.g., cyberpunk, medieval), and/or a particular artist for Diffusitron to emulate.
- After entering your desired details, choose the ‘Anything’ model and click the ‘Diffuse’ button to initiate the magic!
4. Select Use, and tap the ‘Deliberate’ model and adjust the ‘Image Strength’ Slider. I find 40-60% produces the best results.
Observe how the image is now enveloped in a rich tapestry of intricate details and vivid colors. You may wish to revisit the previous image to fine-tune the positive and negative prompts for further refinement. To accomplish this, simply swipe downward.
Presented below is an additional example of a comparable image, crafted using the same method but with more comprehensive concrete details and a negative prompt honed through an iterative trial-and-error process.
Mastering the Redshift-to-Deliberate Model: Generating Photorealistic Landscapes and Character Portraits with Unparalleled Precision
Now, let’s explore the Redshift Model → Deliberate Model approach, which is similar to the Anything → Deliberate method. By using Deliberate as a final touch, this method can generate striking landscapes and character portraits that are impressively close to photorealistic.
However, be mindful of some limitations:
- The further away a character’s face is, the less detailed and more prone to distortion it becomes.
- While this model excels at creating beautiful character portraits, it is less effective at generating complex backgrounds.
Here’s a step-by-step guide to employing the Redshift Model → Deliberate Model method:
- Start with the negative prompt you developed earlier, modified to address issues unique to this model: “Cartoon, deformities, disfigurement, disproportionalities, unrealistic, deformed face, misplaced eyes, elongated arms, elongated fingers, deformed hands, too many fingers, multiple limbs, two heads, blurry, unclear, nose bling, nose blemish, unnatural shine”.
- When inputting the positive prompt, adhere to the same general format as in step two of the previous method. It is recommended to provide greater detail using concise phrases to encourage Diffusitron to generate more intricate images. However, be aware that some details may be overshadowed by others or ignored altogether (for example, eye color might blend into other aspects of the image). Keep in mind that stable diffusion, while powerful, isn’t perfect, and you may encounter unexpected results – but these surprises can sometimes be serendipitous.
3. Select “Deliberate” to refine the image. When adjusting the “Image Strength” slider, 80-90% provides great results while maintaining the originality of the image. Adjustments in the 50-80% range will make smoother images with more realized details though the departure from the original design will be more dramatic. The final image below was diffused at 80%.
After crafting a robust prompt, you’ll find that making subtle adjustments to the details can dramatically alter the resulting image. By tweaking just a few tags, you can generate entirely new visuals. This iterative process is an excellent way to explore Diffusitron’s vast potential and gain a deeper understanding of its wide-ranging capabilities. In the final image showcased below, diffusion was set at 50%, which notably enhances the intricacy of the background elements.
Embracing the Art of the Unexpected: Unlocking the Open Journey Model’s Potential for Unconventional and Breathtaking Imagery
The final tutorial in this comprehensive guide explores the remarkable capabilities of the Open Journey model. This innovative model generates truly one-of-a-kind images that stand out from the crowd. To unleash its full potential, allow ample creative freedom by providing imaginative and surreal subject descriptions. The more unconventional your ideas, the better Open Journey will perform in delivering visually captivating and extraordinary results!
A simple negative prompt is sufficient: “Cartoon, blurry, bad art, bad proportion, dull color, boring uninspired art”. Some of these tags may be rather subjective, but it just works! The final image below was diffused at 80% “Image Strength”.
With OpenJourney’s remarkable capabilities, it can generate images of breathtaking beauty as well as those that evoke a profound sense of terror. The image showcased below demonstrates this versatility, having been diffused at a 50% strength.
We hope that this comprehensive tutorial has served as a solid foundation for you to begin crafting images that embody your distinct artistic flair. To explore a plethora of inspiring creations and to receive valuable insights from a warm and supportive online community, we invite you to join the Diffusitron Discord.