Yes, the featured image is supposed to say “Prompting”. I made it with “huge banner with text “Prompting”, cyberpunk style, rain, dramatic lighting”.
I’ll get right to it, a lot of ya are bad at doing this. But I sort of know why, because a lot of bad information has gotten spread around. Well, sort of, its like 95% bad info. I did some research to prove that prefixing things like “absurdres, masterpiece, best quality, ultra-detailed, 8K, RTX” and such at best does nothing, and at worse HURTS image quality.
Read about the very minor nuances here in the short version of my research results: (! A TL:DR IS AT THE END! )
Do Generic Image Quality Words Actually Matter in AI Image Prompts?
The world of AI image generation is rapidly evolving. As users explore the creative possibilities of these tools, they often rely on “prompt engineering” to guide the AI and achieve their desired results. This involves carefully crafting text prompts that describe the desired image, including details about the subject, style, composition, and other aspects.
One common practice is to include generic image quality words in prompts, such as “masterpiece,” “high quality,” “highres,” or “8K.” The assumption is that these words will influence the AI model to generate images with the specified quality. But does this actually work?
This article delves into the effectiveness of generic image quality words in AI image prompts, examining evidence from various sources, including online discussions, expert opinions, and research papers.
Research Findings
To investigate the effectiveness of generic image quality words, a comprehensive research process was conducted, involving analyzing online discussions, reviewing expert opinions, examining image reviews, consulting research papers and conducting image comparisons.
The findings from these steps are summarized below:
Impact of Generic Image Quality Words
• The research did not find strong evidence to support the claim that generic image quality words consistently improve or worsen the quality of generated images.
• Some online discussions suggested that these words might subtly influence the AI by biasing it towards training images with similar captions or tags. However, this effect was often unpredictable and could lead to unexpected results.
• Expert opinions emphasized the importance of providing specific details about the desired image, such as lighting, camera angles, and artistic styles, rather than relying on generic quality words.
• Research papers highlighted the sensitivity of AI models to prompt wording and the importance of careful prompt engineering.
• Some users reported that excessive use of “quality” words could lead to over-emphasis on certain features or styles, potentially hindering the desired outcome.
• Another interesting finding was that the timing of introducing modifier words in the diffusion process can significantly affect the generated image.
• While this research primarily focuses on generic image quality words, it’s worth noting that some research in text generation suggests that emotional prompting can influence the AI’s output. This raises the question of whether emotional cues in image prompts could also influence the AI’s interpretation and the resulting image. This could be an interesting area for future research and experimentation.
Conclusion
Based on the research conducted, it appears that generic image quality words in AI image prompts have a limited and often unpredictable impact on the generated images. While they might introduce subtle biases in some cases, there is no guarantee of consistent or noticeable improvement in quality.
Instead of relying on these generic terms, users are advised to focus on providing specific and detailed descriptions of their desired images. This includes information about the subject, composition, lighting, style, and other relevant aspects.
Additionally, it’s crucial to acknowledge the limitations of current AI image generation technology. There are constraints on prompt length, and the models might not always accurately interpret abstract concepts.
Finally, it is important to remember that AI image generation is an evolving field, and the effectiveness of different prompting techniques may vary. Continuous experimentation and exploration are crucial for users to understand how to best utilize these tools and achieve their creative goals.
(references and full version of this paper available upon request)
–TOO LONG, DIDN’T READ–
How to Write Effective Prompts:
Focus on what you want to see, not how “good” you want it to be. Imagine you’re describing the image to a friend – be specific about the subject, style, composition, and details.
Use descriptive adjectives and nouns. Instead of “a car,” try “a vintage red convertible speeding down a coastal highway.”
Reference artists or styles you like. For example, “a portrait of a woman in the style of Frida Kahlo.”