Which apps render horns, tails, wings and scales as body rather than costume, which monster archetypes stay stable, and the prompt we start from.
A human face has an effectively unlimited reference behind it. The place where a naga's torso meets her serpent body has almost none, and neither does the join where a wing meets a shoulder blade. That is the whole problem with this category. Non-human anatomy does one of three things between generations: it drifts, so two horns become four and the curve changes; it quietly disappears, so the tail is simply gone in the next image; or it renders as costume, so the horns sit on the head like a hair band, the wings hover behind the shoulders unattached, and the scales look like body paint. Every character described on this page is an adult, human and non-human alike.
Below: the apps whose image generation and character locking hold non-human features together, our verdict on which parts of a monster body persist and which do not, and a glossary of the archetypes with what each one actually renders as.
Highest image score of the eight and it locks a character. In this category that matters more than anywhere else on the site: the moment a generation gets the horns and tail right, lock it, because you may not get them again.
Matches on image quality at 9 and leans photoreal, which is what you want if the goal is a body that reads as skin and bone rather than as an illustrated costume.
Free tier to design on. This category has the highest discard rate on the site, so somewhere to burn generations before paying is worth more here than usual.
Strong memory at $12.99. Half the work of a non-human companion is conversational, keeping the premise alive rather than letting her talk like an ordinary woman, and memory is what does that.
Same score band with a different model underneath, so it fails at different joins. Worth trying when one specific feature, usually wings, will not attach anywhere else.
Level with Lovel and Xotic on overall score. A reasonable second attempt when an archetype keeps arriving as a human in fancy dress.
Image score of 9 and a large library including anime, and monster girls are an anime convention first, so this is the most likely of the eight to already have the archetype you want.
Image score of 7, the lowest here, and romance-first by design. Complex anatomy is the worst possible ask for it. Pick it for the character, not the body.
Our editorial judgement of the category itself, not of any single app, and not measurements. They describe how non-human anatomy behaves when you build a monster companion anywhere.
| Criterion | Rating | What it means for you |
|---|---|---|
| Feature persistence across images | Ask for five images and expect the tail to be absent from at least one of them. Features you mention last are dropped first, so put the non-human details before the clothing in your prompt. | |
| Anatomy versus costume | The distinction that defines the category. Left alone the model gives you accessories: horns that sit on the hair, wings pinned to a back, scales painted on skin. You have to say the feature grows from the body and name the point it joins. | |
| Extra limbs and wing attachment | The hardest thing image models do, and this is where it is hardest. Extra arms, four wings, anything paired and non-standard, will be wrong most of the time. Frame tightly to hide the joins or accept a high discard rate. | |
| Non-human skin and scale colour | Better than you would expect but it fades. Strong reds, blues and greens desaturate toward tan over successive generations, particularly on faces. Restate the colour in every prompt rather than relying on the lock. | |
| Getting one good hero image | Generate enough and one arrives, and it is usually striking. The category rewards patience with a single image far more than it rewards a consistent series. | |
| Feels distinct | The real payoff. Almost nobody has a good monster companion, because most people give up at the costume stage, so a well-built one looks like nothing else. The opposite of the crowded human categories. |
We revise this as models improve at non-standard anatomy, which is the area moving fastest. If an archetype behaves better for you than we describe, tell us which words you used.
The rule that governs everything else here is a feature budget. Two non-human features render reasonably often, three is a stretch, four gives you a mess in which the model quietly discards whichever it finds least plausible. Pick your two, describe how they attach, and let the rest of the character be human.
| Setting | Suggested value |
|---|---|
| Archetype | Name exactly one: succubus, oni, orc, naga. Combining two, a winged vampire orc, does not give you both, it gives you an unstable average of them. |
| Feature budget | Two non-human features, no more. Horns and tail, or wings and eyes, or tusks and skin colour. The third feature costs you the reliability of the first two. |
| Attachment wording | Say where it joins the body: horns growing from the forehead, wings joined at the shoulder blades, a tail continuous with the base of the spine. This is what separates anatomy from accessory. |
| Count | State the number. "Two curved horns" holds far better than "horns", which is read as an approximate quantity and rendered as two, three or four. |
| Placement of skin detail | Name the areas: scales along the forearms, the jawline and the shoulders. Unplaced scales become an all-over texture that looks like glitter or body paint. |
| Framing | Waist-up hides the joins that break most often, which are hips, wing roots and any transition to a non-human lower body. Full-length shots are where you will spend most of your discards. |
| Human anchor | Keep the face human and describe it normally. Likeness is the one part of the character that persists reliably, and it is what makes her the same person across images that disagree about everything else. |
Prompt to copy
An adult demon woman. Two curved black horns growing from her forehead, angled back along the skull, with visible bone texture where they meet the skin. A long slender tail, continuous with the base of her spine, tapering to a point. Deep red skin, matte, with natural skin texture. Human face, dark eyes, black hair to the shoulders. Wearing a simple black dress. Waist-up portrait, low warm lighting, dark neutral background. Photorealistic. Horns and tail are part of her body, not worn.
Four directions the same category can go in. Copy one, paste it into the character builder, and change the parts you disagree with.
The best represented archetype and the one to start with. Its default styling is Halloween costume, so the whole prompt is about pulling it back toward anatomy.
An adult succubus. Two curved horns growing from the forehead with bone texture at the base where they meet the skin, angled back. Slender tail continuous with the base of the spine. Pale skin with a faint warm undertone, natural skin texture, no body paint. Human face, dark hair, amber eyes. Plain black dress. Waist-up, low warm lighting, dark background. Photorealistic. The horns and tail are anatomy, not accessories or costume.
The smallest deviation from human and therefore the most reliable in the category. Worth building first if you want something you can actually keep across a series.
An adult vampire woman. Very pale skin with visible fine veins at the temples, natural skin texture. Pale grey irises with a dark ring. Slightly parted lips showing two sharp upper canines. Dark hair, long and straight. High-collared black coat. Cold blue-toned key light from one side, deep shadow, dark background. Photorealistic, sharp detail on the face, waist-up portrait.
Coloured skin plus tusks. The skin is the unstable part and desaturates toward human tones as you regenerate, so restate the colour every time.
An adult oni woman. Deep red skin, matte and evenly coloured, natural skin texture, not painted. Two short thick horns growing from the upper forehead, ivory coloured, with skin folding slightly where they emerge. Two small lower tusks visible at the corners of the mouth. Black hair pulled up. Dark simple robe. Waist-up, warm lantern light, dark background. Photorealistic, red skin, strong colour saturation.
The hardest preset here. Frame it waist-up on purpose, because the human-to-serpent transition is where the category breaks most reliably.
An adult naga woman, waist-up portrait only. Human upper body with green-gold scales along the forearms, the outer shoulders and the jawline, blending into skin, matte and reptilian rather than glittering. Slit pupils, amber eyes. Dark hair. No clothing above the waist beyond simple gold bands on the upper arms. Warm side lighting, dark background. Photorealistic, scales are part of the skin, not applied.
Each archetype has its own characteristic failure, and knowing which one you are signing up for is most of the battle. Below is what these words actually produce, rather than what they mean. All of these characters are adults.
| Term | What you actually get |
|---|---|
| Demon or succubus | The best represented archetype by a distance, so horns and tail come back most often. The default styling is Halloween costume, so counter it explicitly with attachment wording and plain clothing. |
| Vampire | Barely non-human, and therefore the most stable option in the category. Fangs are the only marker and they vanish unless the mouth is open, so ask for slightly parted lips. |
| Oni | Horns plus strong red or blue skin. The horns hold, the colour does not: it desaturates toward tan over successive generations, especially on the face. Restate the colour every prompt. |
| Orc or half-orc | Green skin and lower tusks. Tusks shrink until they disappear and you are left with a green-skinned human. Naming them as lower canines protruding at the corners of the mouth holds better than "tusks". |
| Naga or lamia | The human-to-serpent transition is the failure point, and it commonly renders as a wrapped skirt or a coil of fabric rather than as a body. Waist-up framing solves it by avoiding it. If you want a hybrid body that does hold across a series, a mermaid is the one that does, because that transition is heavily represented. |
| Harpy | Wing-arms confuse the model badly. You get arms and wings both, or wings that hang off the back with no join. Decide which you want and say the other does not exist. |
| Kitsune or werewolf | Ears and tail, with fur coverage as the axis you must state. Watch for duplicate ears, human ears and animal ears on the same head, which is the signature error and easy to miss at small sizes. |
| Dragonkin or scalekin | Scales become glitter or body paint unless you place them, so name the areas: forearms, outer shoulders, along the jaw. Ask for matte and reptilian to stop the metallic sheen. |
| Slime or goo | Translucency is genuinely unstable and effectively cannot be locked. Treat every image as a one-off and do not plan a series around it. |
| Arachne or multi-armed | The worst case in the whole category. Limb counts are where image models fail hardest, and extra arms compound the existing hand problem. Expect to discard most generations. |
Features you mention late in a prompt are dropped first when the model runs out of room, and a tail is easy to omit without making the image look wrong. Put the non-human features before the clothing in your prompt, keep to two of them, and lock the character as soon as one generation gets them right.
Describe the join. "Two curved horns growing from the forehead, with bone texture where they meet the skin" produces anatomy; "horns" produces a hair band. The same trick works for wings, joined at the shoulder blades, and tails, continuous with the base of the spine.
Vampires, because they deviate least from a human body, then demons and succubi, because they are the best represented. Naga, harpies and anything with extra arms are where you will spend the most attempts for the least consistency.
Yes on the adult-oriented platforms listed here, provided every character is an adult, which they are throughout this page. Non-human characters are a well-established part of these apps, and the anime-leaning libraries carry the widest selection of them.