Shot Guide: AI Character¶

For performers, models, and actors building an AI Character. This guide tells you what we need in your reference set, and why each piece matters for the quality of your finished AI Character. How you produce it is up to you — solo with a phone, with a friend, or with a full crew. We focus on the result; you decide how to get there. This is the friendly version of the Character LoRA Creation Guide.

What you're making¶

An AI Character is a digital version of you that filmmakers can cast in AI-generated films. You deliver one reference set — photos and short video clips of you, in everyday clothes. Filmmakers license your AI Character whenever they want to use you in a scene. You earn on every sale.

The quality of the AI Character comes almost entirely from what's in this reference set. Time you invest here is the difference between an AI Character that gets cast often and one that doesn't — so it's worth being deliberate about every capture below.

What we need to see¶

You, as yourself. Not in character. Not in costume. Just you.

Three things matter:

• Your face — from many angles, with many expressions, in many lights.
• Your body — your real proportions and silhouette.
• Your motion — how you walk, sit, gesture, and look around.

The AI learns you. Filmmakers add wardrobe, hairstyle, and scene later through their prompts — so we don't need any of that from you. Just the person.

Wardrobe¶

The reference set should include four to six different outfits. The point is to teach the model that clothing is variable — separate from who you are — so a filmmaker can later put your AI Character in anything. Varying the outfit across the set is what detaches the clothing from your identity. (You don't need to label or caption anything — just deliver the varied shots; our team handles the rest.)

Recommended¶

• Fitted or semi-fitted — close enough that the silhouette and body shape read clearly.
• Solid, mid-tone colors across a range of hues — heathered gray, muted blue, olive, tan, burgundy, dusty/desaturated tones. Mid-tones preserve fabric folds and body contour under varied lighting.
• Matte fabrics — cotton, wool, knit. Even, predictable surfaces.
• Simple necklines and clean cuts — crew, scoop, henley, plain collar.
• Genuine variety — different colors and silhouettes across the set, not the same shirt in five shades.
• Optionally one tighter, more body-revealing piece (fitted tank, sports bra + shorts) for the cleanest silhouette read — useful but not essential, and don't let it be your only well-lit full-body shot.

Avoid¶

• Logos, graphics, prints, stripes, checks, busy patterns — high-frequency detail the model memorizes and then smears onto outputs where it doesn't belong (and causes moiré in video).
• Pure black or pure white as most of the set — black crushes contour into shadow, white blows out; the silhouette signal is lost. Use sparingly, not as the default.
• Neon, fluorescent, or heavily saturated colors — they cast colored light onto the skin and the model can bake that tint into your face.
• Shiny, reflective, sequined, or sheer fabrics — specular highlights and see-through texture are view-dependent noise the model can't learn consistently.
• Baggy or oversized clothing — hides the body geometry that's part of your identity.
• A consistent signature accessory (same hat, glasses, necklace, watch in most shots) — it entangles with the trigger and gets baked in, exactly like a logo.

Why this matters for your AI Character. If every photo shows the same outfit, the model treats the outfit as part of you — then when a filmmaker wants you in a suit, the model fights itself and the result drifts. Varied, well-chosen clothing teaches the model to separate you from what you're wearing. The color and fabric rules above exist so that variety doesn't come at the cost of a clean silhouette or a color-cast face.

Hair, makeup, accessories¶

Unlike wardrobe and background, this section is not about variety — it's about consistency. Hair, face, and bone structure are how the model learns you. Keep them stable across the set so the model locks one clear identity; save the variety for clothing and backgrounds, which you want detached from you.

Recommended¶

• Hair: your natural, everyday style — kept consistent across the whole set. One stable look teaches the model a clean identity. (If supporting multiple hairstyles is a real goal, shoot deliberate, well-covered sets of each — don't mix them in casually.)
• Makeup: minimal and consistent — close to your real face, the same level of makeup in every shot.
• Jewelry: only small, everyday pieces — a wedding band or simple studs are fine. Tiny and edge-of-frame, so low risk.
• Face fully visible — unobstructed, well-lit, head and hairline clearly readable in every shot.

Avoid¶

• Changing your hair across the set (long in some, short in others, up vs. down) — to the model this reads as different people under one trigger, and identity gets mushy.
• Heavy contouring or dramatic makeup — contouring is painted-on shading that fakes bone structure; the model can't tell it from real geometry and bakes it into your actual face, where it then shows up in every output.
• Inconsistent makeup (bare-faced in some, full glam in others) — same identity-blurring problem as changing hair.
• Sunglasses, hats, scarves, masks, wigs — these hide the face and head the model is trying to learn. Worse than binding: they remove identity signal.
• Any accessory worn in most shots (signature glasses, a hat, a statement necklace) — it entangles with your trigger and gets baked in, exactly like a logo on a shirt.

Why this matters for your AI Character. Capture you as you, consistently — your real face, real structure, your everyday look. That's the foundation the model builds the identity on. Once that foundation is solid, a filmmaker can add the suit, the sunglasses, the dramatic lighting, or a new hairstyle in the scene. Get this backwards — vary your face and hide it behind accessories — and the model never learns a stable you to dress up in the first place.

What NOT to use¶

• Snapchat or Instagram filters of any kind.
• Smoothing or beautifying filters.
• Heavy makeup or character looks.
• Costumes of any kind.
• AI upscaling on any file.

Each of these distorts the signal the AI is trying to learn. A filtered photo trains a filtered AI Character, and there's no undoing it after the fact.

Photos — 60 to 80 total¶

Headshots — 20 photos¶

The most important photos in the set. This is where the AI locks in your face. Face fills about 60–70% of the frame.

• Angles: straight on, three-quarter left, three-quarter right, full profile (both sides), slight tilts up and down.
• Expressions: neutral, light smile, big smile or laugh, serious, mid-speech (mouth slightly open), eyes closed.
• Gaze: at the camera, and slightly off-camera in several directions.

The expression and gaze variety is what lets the AI hold your face together when the character speaks, laughs, or looks away from camera. A set of neutral, locked-gaze headshots produces an AI Character that breaks the moment it tries to act.

Mid-shots — 20 photos¶

Waist to head. These connect your face to your body — how you carry yourself from the chest up.

• Same angle variety as headshots.
• Poses: relaxed stance, hand on hip, arms crossed, looking over a shoulder, slight lean.

Full-body shots — 20 photos¶

Head to feet, with breathing room above and below. How you actually take up space.

• Angles: front, back, both profiles, three-quarter angles.
• Poses: standing relaxed, walking pose, sitting, mid-stride, casual lean.

Detail shots — 10 photos¶

The small things that make you specific.

• Hands — open, fist, holding something.
• Hair from the side.
• Anything distinctive — tattoos or scars, only if part of the everyday body.
• Walking gait, captured from the side.

Video clips — 10 to 15 total¶

Each clip two to five seconds. No audio needed. The clips capture how you move, not a performance — natural motion is exactly what we want. No camera jitter; the camera should stay stable.

There are nine clip types below, and all nine are required. Aim for 10–15 clips in total by shooting a second take of the most useful ones (talking, facial expressions, and the walking clips reward extra coverage).

1. Slow 360° turn — standing still, slow full rotation.
2. Walking toward the camera — natural pace.
3. Walking away from the camera — natural pace.
4. Walking across the frame — side view.
5. Sitting down and standing up — in a chair, barefoot.
6. Talking — random sentences, varied mouth shapes (A, E, I, O, U). Audio isn't used; the motion is.
7. Facial expressions — neutral → smile → laugh → serious → surprised.
8. Head turns — slow look left, right, up, down.
9. Hands moving — gesture, pick something up, wave.

Lighting¶

The set needs to include several different lighting conditions, so the AI can render your face under any light a filmmaker chooses. Aim for roughly this spread:

• Soft, even light — about a quarter of the set.
• Side light — about half. Switch the lit side partway through.
• Harder, more directional light — about a quarter.
• Warm and cool color temperature — a small number of shots in each. Optional but valuable.

One rule across all of it: keep the eyes lit. They should never disappear into shadow.

Why this matters for your AI Character. A face shot only under flat light looks flat at generation time — the AI never learned how your cheekbones, brow, and jaw behave when the light comes from somewhere interesting. The minute a filmmaker prompts hard side light or golden hour, your AI Character drifts. Lighting variety is what makes an AI Character cinematic.

Background¶

Use three to five different, uncluttered backgrounds as a floor — more is better. The goal isn't "plain," it's varied and non-distracting. Variety is what teaches the model which pixels are you and which are everything else; a single background — even a clean one — binds to your identity and gets baked in.

Recommended¶

• Varied across the set — different surfaces and a few different contexts (a plain interior wall, a simple room, soft outdoor daylight). Spread matters more than any single backdrop being perfect.
• Uncluttered, low-distraction surfaces — neutral or muted walls, simple environments. Quiet, not necessarily featureless.
• Mid-tone, low-saturation colors if a backdrop is solid — soft gray, muted beige, desaturated tones that won't throw colored light onto you.
• Clear separation between you and the background — stand well in front of it, lit separately, so background color and light don't wrap onto your skin or hair.
• Outdoors: soft daylight, gently blurred background, no sharp landmarks.

Avoid¶

• Green screen / chroma backgrounds in the raw training set — the saturated green casts onto your edges and skin and gets learned as part of you, and one flat color across the set binds to your identity. Only use green screen if you key it out cleanly (proper matte + despill) and composite varied real backgrounds in before training.
• Any single solid color across the whole set — even neutral. One constant background binds, exactly like one constant outfit.
• Saturated or bright-colored walls (red, blue, green, etc.) — they cast colored light onto your face and hair.
• Text, logos, posters, signage — high-frequency detail the model memorizes and smears onto outputs.
• Other people — a character model can entangle your trigger with the wrong face. No exceptions.
• Sharp landmarks or busy, recognizable scenery — distinct background features bind to you and reappear where they shouldn't.

Why this matters for your AI Character. If the AI sees one background — or one background color — across the whole set, it treats it as part of you, and you'll see it bleed into generations later. Varied, quiet backgrounds with clean subject separation teach the model which parts of the image are you and which are everything else. That's also why green screen backfires: it looks like "no background," but to the model it's one loud, constant, color-casting background — the worst case, not the neutral one.

Camera and file requirements¶

What we can work with — and what we can't.

• Photos: minimum 1920×1080 pixels (4000×6000 is better). PNG, JPG, or HEIC. Originals only — no screenshots, no edited exports, no AI-upscaled files.
• Video: minimum 1080p (4K preferred). MP4 or MOV. Fixed frame rate at 24, 30, or 60 fps.
• Sharp focus on the eyes for stills and on the subject plane for video.
• No camera jitter on video — the camera should stay stable.
• No filters, no beauty modes, no color grading, no AI upscaling applied before delivery.

What your AI Character will and won't do¶

Honesty here matters — so you know what to expect, and so we set the right expectations with filmmakers too.

What it does well¶

• Recognizable face from any angle, in any light.
• Body proportions and silhouette.
• Distinctive features — hair color, eye color, skin tone, characteristic marks.

What it can't do yet¶

These are limits of AI video today, not problems with your reference set. Every filmmaker is told about them up front.

• Hands holding things or making complex shapes — fingers and grip are still hard for AI video. (We still ask for hand shots in your reference set — they sharpen the model overall — but scene-level grip is a current limitation.)
• Reading text inside scenes — signs, logos, written documents won't be legible.
• Lip-syncing to specific audio — needs a separate tool.
• Multiple people interacting physically — handshakes and fight choreography get messy.
• Long unbroken takes — most AI video tools work in four-to-six-second windows.
• Detailed brand logos or fine fabric patterns.

After you upload¶

1. Upload. Drop your files into the upload tool on your AI-Cast creator dashboard.
2. We build. Our team trains your AI Character — about an hour of AI work.
3. You review. We share preview clips. You approve what filmmakers see.
4. List or keep. List it on the marketplace, or keep it private.

We hold back five to eight of your photos as a "test set" — never shown to the AI during training. We use them afterward to confirm the AI Character looks like you. If it doesn't, we retrain — at our cost.

Checklist before you submit¶

• All four photo groups covered (Headshots, Mid-shots, Full-body, Detail)
• All required angles per group (front, back, sides, three-quarters, tilts)
• Expression variety (neutral, smile, laugh, serious, mid-speech, eyes closed)
• Gaze variety (at camera, off-camera left, off-camera right, up, down)
• 4–6 different outfits across the set; no single outfit dominant
• Lighting variety (soft, side, harder, warm/cool); eyes lit in every shot
• 3–5 different uncluttered backgrounds (varied, not just neutral); more is better
• All 9 clip types captured (10–15 clips total), 2–5 seconds each, stable camera
• Resolution minimums met; originals only; no filters, no upscaling, no grading
• Release form signed in your AI-Cast dashboard
• Age verified with government ID in your dashboard