Ellie T

This was initially meant to be my first attempt at creating an unglamorous fictional female model that could span a reasonable age variance. In the end it’s proven to be a painstaking exercise in creating a strong uniformity of prompts, particularly to control subject features and positions.

Experiments so far suggest 0.55 to 0.85 on the slider tends to work best.

Prompts trained in the following general order:-

Trigger: Ellie

Alone?: <specify if alone, or leave blank>

eg: alone

Age of woman: <20 to 65 in 5-year intervals>

eg: 30-year-old woman

Hairstyle: <parting: centre-parting, slightly-side-parting, side-parting, pulled-back, descriptive: tucked-behind-shoulders, voluminous, large, loose, tight, straightness: straight, slightly-wavy, wavy, length: very-short, short, shoulder-length, long, very-long, style: bob, bun, ponytail, half-up, updo, modifiers: with bangs, with slight-bangs, with widow's peak>

side-parting voluminous straight shoulder-length bob with widow's peak haircut

Skin colour: <fair, tanned, brown, dark, chrome, stone, plastic>

eg: fair-skinned

Hair colour: <auburn, blonde, bottle-blonde, brown, brown with highlights, brunette, grey, grey-blonde, raven, chrome, stone>

eg: raven hair

Eye colour: <blue, brown, green, hazel, almost-closed, closed, chrome, stone, no>

eg: brown eyes

Clothing: <a lot of variables here, generally colour-neckline-material-clothing, top to bottom. Note training images is a mixture of office street and science fiction>

eg: wearing black turtleneck top under black blazer, black smart miniskirt, semi-sheer tights, garnet rhombus pendant necklace, garnet drop earrings, black suede high-heeled ankle boots

eg: wearing metallic silver turtleneck catsuit with builtin gloves and builtin boots, red lipstick

eg: wearing grey scoopneck top, navy-blue smart pants

eg: wearing 1980s supergirl costume, red lipstick

Physiology: <breasts: very-small, small, normal, large, very-large, waist: very-narrow, narrow, wide, very wide, hips: very-narrow, narrow, wide, very wide, butt: small, large, optional: cameltoe>

eg: large breasts, narrow waist, wide hips, cameltoe

Body shape: <very-slim, slim, normal, hourglass, thick, very-thick>

eg: slim body

Facial expression: <open-mouth modifier: gasping, talking, expression: neutral, smiling, slightly-smiling, cheerful, concerned, worried, frightened, happy, angry, sad> faceless, no

eg: talking, neutral facial expression

Pose: <quite a lot of variance but hopefully covered it here, position: standing, sitting, walking toward camera, upright, prone, quite-prone, on-hands-and-knees, on-knees, on-haunches, laying supine, crawling, lean modifier: leaning forward, leaning backward, leaning to one side, horizontal torso-propped-slightly-upright, rotation: square-on, slightly-side-on, side-on, very-side-on, turned-away, hand/arm position: one/both arms down by side(s), slightly-outstretched, outstretched, very-outstretched, arms crossed, one/both hand(s) raised, slightly-raised, very-raised, resting on thigh(s)/knee(s)/head/surface/object, touching hair/ear/face, behind back, over stomach, grabbing belt buckle, hands together, hands on hips, one hand on hip, leg position: together, slightly-apart, apart, wide-apart, leg modifier: one/both leg(s) kicking forward, kicking backward, raised, one leg in front of the other, optional: floating>

eg: standing slightly-side-on, one arm down by side, one hand on hip, legs slightly-apart

Direction: <facing, looking just-off-, looking off-, facing sideways from, looking down off->

eg: looking just-off-camera

Setting: <a lot of variables here, easier to just give examples of training-set prompts>

eg: floating in outer space, zero gravity, planet and stars in background

eg: in cockpit of fighter plane, sitting in pilot seat

eg: plain beige room in front of white bed

eg: television news studio

eg: plain light brown studio background, sitting on black leather sofa

Distance: <full body, almost-full body, upper body, torso, quite-close-up, close-up>

eg: full-body photo

Quality: <hyperdetailed photography, normal quality image, low quality image, hyperdetailed photography, normal quality image, low quality image, "optional" background out of focus: background out of focus>

eg: hyperdetailed photography, background out of focus

So from these example prompts you might come up with something like:-

Ellie, alone, 30-year-old woman, side-parting voluminous straight shoulder-length bob with widow's peak haircut, fair-skinned, raven hair, brown eyes, wearing white transparent safety goggles, metallic silver very-highneck seamless catsuit with builtin gloves and builtin boots, red lipstick, large chunky steel hoop earrings, large breasts, narrow waist, wide hips, hourglass body, cameltoe, gasping, frightened facial expression, quite-prone side-on, both hands raised, legs slightly-apart, both legs slightly-raised, floating, looking just-off-camera, floating in outer space, zero gravity, stars in background, full body photo, hyperdetailed photography

Hopefully it’s turned out alright you can have some fun with it.

NB1: A lot of face-swapping and manual editing of the training database has gone into providing some variation in appearance, but there are some features that are dominant or at least more common (and therefore have more commonly-featured prompts in the training database): 30 and 45 ages, side-parting voluminous straight shoulder-length bob hairstyle, brunette and raven hair colours, brown eye colour, fair skin colour, catsuit and turtleneck clothing, cheerful facial expression, standing and sitting poses with both arms down by sides, and facing the camera. Most noticeable step-change in appearance is 30s to 40s.

NB2: Face blending and swapping has been used so that the database does not represent an individual, any likeness is coincidental.