- Using the simplified mode is recommended
We advise avoiding the advanced mode unless you have a specific need for it, such as gaining greater control over individual word weights or requiring additional details in images that are challenging to convey in the simplified mode.
This recommendation is based on the fact that, under normal circumstances, obtaining poor images is unlikely when using the simplified mode, unless you overload the 'other' section with excessive words or incorrect prompts. The 'other' section serves as a mini prompt injector.
Moreover, long prompts inevitably decrease the power/weight of the extensions selected. Think of prompt and extension ways of communicating to the AI what you desire. If your prompt contains too many different words and ideas, extensions will lose their power, especially when you encapsulate words into parentheses. If you are getting bad images or the extensions don't seem to work well, a long prompt is one of the likely reasons.
- Words to avoid in the prompt
Refrain from using terms such as 'legs,' 'hands,' or other body parts, as this may result in the generation of multiple limbs or anomalies.
- Use short words + commas (with most AI models)
An example of good prompt:
1girl, blonde hair, swimming pool, large breasts, smiling, spread legs.
An example of bad prompt:
a beautiful girl spreading her legs in a swimming pool. She also has large breasts and she's smiling.
The 'bad' prompt is characterized as natural language. The engine we're utilizing demonstrates a superior understanding of simple words and commas in comparison to more complex natural language structures.
The only exception to this is when using the "semireal" model, which seems to give better results with natural language (e.g. "a photo of a beautiful girl in a park").
- Face prompt
It is advisable to utilize a 'face prompt' as it results in regenerating the face at a higher resolution. However, this approach is not recommended when a girl or character is chosen, or when 'oral' extensions are selected.
- Always upscale 2X
If you take a look at the showcased pictures, most of them were upscaled 2X, and that is mostly why they look abysmally better compared to non-upscaled ones. Please note that
upscaling 3X uses a different algorithm than the 2X upscale, and is not as good as 2X upscaling, even if the resolution ends up being larger.
If you like a picture, chances are it could look even better if you upscaled it 2X. This is not always the case with generations that used characters/girls or images for face replication, however.
Upscaling 2X, on average, leads to more realistic images, better details, fixed eyes, and much better images in general.
- Hardcore content
While Base AI models can generate explicit content when the right extensions are selected, our HD AI models consistently deliver higher-quality results with significantly fewer malformations, leading to a lower failure rate.
- Don't select conflicting extensions
Selecting conflicting extensions (such as two entirely different poses), can lead to malformations and is not recommended. This is especially true for Base AI models.