Firstly, I extend my gratitude to the GS developers and decision makers for taking an important step into the creative AI universe.
The AI Visualiser exhibits useful functions such as 'Iterations' and 'Prompt Strength'; nonetheless, it lacks an essential feature: 'Negative prompt'.
I am aware that the 'txt2imageCongifTemplate.json' file includes an 'hr_negative_prompt' option, but having it readily accessible in the interface would be more convenient.
Negative prompts help avoid unwanted outcomes and ensure consistency during the process.
Some examples of Negative Prompts:
- trees
- green
- water
- asphalt
- unnatural colors
- poor lighting
- deformed
- draft
- people
- overexposed
- text
The model may be able to learn from images without text, although architectural photos of shopping centres or cafes may have additional text or brand names, so it is helpful to have a negative prompt edit box for tuning.
As stated in an article, negative prompts within stable diffusion provide multiple advantages, including:
- Control: Negative prompts assist in managing the diffusion process by preventing undesirable results and guiding it in the desired direction.
- Optimization: By discouraging unwanted behaviours or reactions, negative prompts contribute to a more efficient and optimized diffusion process, which can produce better results.
- Stability: By using negative prompts, you can sustain stability during the diffusion process, making it consistent and predictable, which is crucial to succeed.
- Lowered risks: Through identifying and avoiding potential pitfalls, negative prompts can significantly reduce the risks linked to the diffusion process.
For further information, see the article on the importance of negative prompts for good outcomes at minimaxir.com/2022/11/stable-diffusion-negative-prompt/.
Please, add an option to tune negative prompts in next version of AI Visualizer. Thank you in advance!
- George
George
