Common Issues of Image Generation and How to Solve Them

If you are on a natural-language model

On SK-Alpha and the ZIT models there is no weight syntax, no scheduler and no CLIP skip. The fixes below are all in the sentence itself, or in one toggle.

• Wrong style — anime when you wanted photoreal, or the reverse. Do not argue with it in adjectives. Set Force Style to Realistic or Anime; it pins the render regardless of what the prompt says. This is the single highest-yield fix on this family.
• Extra people in the frame. Say it in words: "a single person, alone in the frame". (solo, alone)+++ is not a spell here — it is just three words and some punctuation.
• Something you asked for is missing. The sentence probably buried it. Move it earlier, give it its own clause, and cut anything the model is being asked to hold at the same time. Long prompts do not fail because they are long; they fail because they contradict themselves.
• The character does not look like themselves across images. Use the face reference — that is precisely what it is for. Describing a face in words will not get you consistency across a set.
• Muddy or incoherent output. Shorten. One clear scene beats a paragraph of qualifiers, and these models have a longer prompt budget than the SD ones, which makes it easy to over-fill.
• You want img2img. The ZIT models take a Soulkyn image link as a base; SK-Alpha does not. If the option is missing, that is why.

And the general one: change one thing per attempt, and regenerate before you rewrite. Two runs of the same prompt already differ.

Blurred Image

Sometimes when generating images they come out blurry it's something everybody has to deal with from time to time. This can be caused by many different things such as:

Scheduler Having an Issue

This can be solved simply by changing the Scheduler. Please remember that different Schedulers have different uses and give different final effects.

Inadequate Prompt Strength

Because prompt strength gives image generation more freedom to experiment on lower values it can also lead to muddy, unclear pictures due to not having clear enough picture in the prompt itself.

Overloaded Prompts

Having too much of information on something sometimes can also be a bad thing! Prompts with too much defined can also lead to wrong priorities when prompting images leading to blurry pictures. To solve this try to condense the prompt.

LoRA Issues

Some LoRAs just hate each other and will give you issues when used alongside one another.

Depth of Field Misinterpretation

Sometimes it's also an issue of interpreting the words too literally which adds unnecessary blur. Cinematic lighting, blurred background and others can lead to having blurred images. To solve this you can add prompts like:

sharp focus, character focus, detailed subject, clear contours, depth separation.

Missing Character Features

Whenever you make a character, make a clothing or location preset from time to time image generation doesn't show every aspect of a thing we want to generate. This most often is because AI might not know the exact thing you're prompting or you might need to add things adjacent to the meaning of it to reinforce it.

Extra People in Picture

Conflicting physical prompts can cause extra people to appear in your pictures. Very often this can be solved by adding (solo, alone)+++ to the image prompt. This tries to force image generation to generate picture only with one person.

Extra Objects

This can be because of many different things. Most often it is because AI misunderstands the thing you're trying to add. For example: Hourglass figure. This can be either understood as literal hourglass or how body is shaped. To solve this try to prompt around it.

Carpet Looking Pictures

When picture looks like a carpet under microscope change the weights in the image prompt or LoRAs.

Visual Artifacts

Due to the nature of smaller details and AI having issues generating them, like piercings, small beards and others AI can give you artifacts. To solve that you either need to add a detailing LoRA or if you're able you can add Detailed_beard :2, or Detailed_piercing :2 in this case to physical prompt.

Be mindful that it might not always solve your issues with small details.

Incorrect Anatomy

It's pretty common that image generation will mess up limbs, fingers etc. This can be alleviated by using things like proper fingers, perfect fingers and other associated limbs.

Too Many Fingers, Hands

This can be solved by specifying the amount of fingers you want to be pointing, or being shown. Same goes for hands.

(Pointing finger, one finger)+ will make character point one finger.

(Reaching out, One arm, left arm, one hand, left hand)+ will make character reach out only with one arm.

Changing Eye/Hair Color

Some of the models seem to have issues with certain colors. To try and solve it you can add Detailed_golden_eyes :2, or try to work around the name to try and find other name that solves the issue. Same goes for hair.

Character Showing XYZ

When you prompt certain character features it will make character to try and always show them. For example boobs can give naked breasts, you can try changing them to chest/breasts.

When making character that has a penis you can try adding either cock bulge in clothing or image prompt. Sometimes Hidden:cock works just fine as well. It can be as well used to hide nipple piercing.

Wrong Gender

Sometimes due to how prompts interact or clear bias of some image models male characters can turn female (most often anime models) you can add male:2 ; adult_male:2 in physicals or (1male)+++ in image prompt.

Surprise Text Appearance

Sometimes using certain tags will not only give you the desired effect but sometimes add text. Try lowering the weight on a prompt that gives you text.

My kyn shows an "⚠️ Outdated" badge

That badge means the kyn is still pinned to a retired image model. Three models were removed from the picker: SD - Realistic, SD - Realistic Freedom and SD - Realistic Photo.

Nothing is broken. Their old identifiers now resolve to SK-Alpha, so generation still works — but the kyn is carrying a setting that no longer corresponds to anything you can choose, and the badge follows it everywhere on the site, plus a warning on its profile.

The fix: edit the kyn and pick a current model. If the kyn is realistic-class, SK-Alpha is the direct successor and the one its old slug already aliases to; ZIT - Sk-Real and ZIT - Sk-Photo are the other realistic-leaning options. Remember that editing creates a new version and leaves the old one intact, and that existing threads stay pinned to the version they started on until you move them.

One thing worth expecting: the retired models were tag-prompt SD models, and all three replacements are natural-language models. Prompts written for the old one will not transfer as-is — see the natural-language section of the advanced guide.