Common Problems

Two-head problem

When generating portraits of humans using AI, it is not uncommon to encounter images where two heads appear to be connected together. This issue often arises when using image sizes that deviate from the 1:1 aspect ratio. To prevent this, you can follow these suggestions:

  • Change the style to Midjourney or 2.5D.

  • Use an image size with a 1:1 aspect ratio, such as 512×512.

Not showing full body

One common issue encountered when generating full-body images is that the keyword "full body portrait" may not always produce the desired result. In some cases, the keyword might be completely ignored by the AI. Below are a few examples where the keyword did not have the intended effect.

Describing the lower body

A more consistent approach would be to include specific keywords that depict the desired elements in the image. For instance:

  • standing

  • long dress

  • legs

  • shoes

By incorporating the keywords "standing" and "long dress," the generated images now showcase a greater portion of the body.

Now it is showing more of the body but we can do better.

Use portrait size

If you desire a full-body image, employing a portrait size would be more suitable. Here are a few examples that demonstrate the utilization of a portrait size in conjunction with the keywords "standing" and "long dress."

Messed-up finger

AI algorithms often encounter difficulties in accurately depicting fingers. It is not uncommon to observe distorted representations, an excessive or insufficient number of fingers.

To address this issue, there are two suggested fixes:

  • Incorporate keywords that specifically describe hands and fingers, such as "beautiful hands" and "detailed fingers." Including these keywords tends to prompt the AI to generate images with well-defined hands.

  • Utilize inpainting techniques. Create a mask to identify the problematic area and apply inpainting to generate multiple images. Select the image that best meets your preferences. It has been observed that reusing the original image prompt is often sufficient for achieving desirable results.

Last updated