AI Girlfriend Image Prompts: A Practical Guide

Why your first image disappointed you

Most people’s first attempt at an AI girlfriend image reads like a mood board rather than a picture brief. “Make her dreamy and romantic” tells the model how you want the feeling to land, but not what to draw. The generator needs decisions: what the subject is doing, where she is, what she’s wearing, the time of day and the camera angle. Without those specifics the system fills gaps with defaults you didn’t intend, and the result feels vague or mismatched.

A second common mistake is skipping setting. “In a cosy room” is still nebulous unless you add details: is it a small studio with exposed brick, a sunlit bedroom with sheer curtains, or a rainy café booth? The background anchors the subject and informs wardrobe choices, props and lighting. If you want an atmospheric neon scene instead, that’s a different set of instructions — and a different art style might suit it better. If you’re unsure which style fits your idea, the short guide on choosing an art style helps match tone to technique.

Lighting is the single attribute that most reliably transforms an ambiguous image into a clear one. “Soft light” could mean overcast daylight, candlelight, or a softbox in a studio; each produces a distinct mood and shadow pattern. Naming the direction (backlit, side-lit), the source (golden sunset, fluorescent) and the hardness (soft, rim-lit, hard shadows) lets the model place highlights and shading correctly. Without that, faces lose definition or end up lit in ways that conflict with the background, which makes the result look “off” even when composition and costume are fine.

Framing and camera decisions are the other frequent omission. Saying “close-up” versus “three-quarter shot” versus “full body” changes pose, expression and clothing detail. Portraits need face-centred framing and attention to eye contact; a full-body pose needs a defined stance and context. If you leave framing vague the system may choose something generic, which wastes a generation and the credits you used for it. If you’re experimenting, try practising with a low-cost prompt and a few dedicated partners from the roster — the companions to practice with page makes it simple to choose someone to iterate with.

Finally, understand that an image prompt must translate feelings into visual instructions. Don’t expect the generator to infer your private shorthand for “romantic” or “mysterious.” Be concrete about clothing, setting, light and lens, and include one or two reference art directions if you want a specific finish (for example, “soft film grain, warm palette”). If you’re trying very different looks — realistic textures or stylised brushwork — check out the image generator features and available all eight styles so you aren’t surprised by how the model renders hair, fabric or skin. When an iteration fails, review which of the four decisions above you skipped before spending more on another generation; the guide to what generations cost shows how quickly credits add up when you’re revising without narrower direction.

The six-part structure of a prompt that works

A reliable image prompt is a string of six clear choices: subject, wardrobe, setting, light, mood and framing. Together they tell the generator not just who to draw but how to present them. If one of those parts is vague, the result can be generic or just plain wrong — and that wastes credits. You can try variations without spending, but when you do buy extra generations, see what generations cost so you know the trade-offs before experimenting.

Start with the subject: the companion’s age (over 18), body type, face, hair and any accessories that identify her. This is the “who” of the image; if you skip it, the model will invent features. One-line example: “young woman, olive skin, shoulder-length dark hair, small crescent tattoo behind left ear.” Practise with a few friendly models from the companions to practice with list before committing credits to a final version.

Wardrobe defines style and personality. Be specific about garments, fit and fabrics — “silk slip dress, navy, ankle-length, delicate straps” gives the generator something concrete; “casual outfit” does not. Clothing choices affect composition and mood, and changing a single word (linen versus leather) can produce a very different image. One-line example: “fitted leather jacket, white tee, high-waist jeans.”

Setting tells where she is: a city rooftop, a small kitchen, a midnight alley. The setting anchors lighting and props and helps the model choose background details instead of filling the frame with unrelated objects. If you want motion or a sequence, remember that still frames are what the platform produces and they can be animated into a short clip via the video scenes workflow — it’s not a photoreal film studio, it’s animated stills. One-line example: “rooftop at dusk with city skyline behind.”

Light shapes texture, colour and mood. Describe time of day, light direction, hardness and colour temperature: “warm late-afternoon rim light from the left, soft shadows” is better than “nice lighting.” Incorrect light descriptions produce flat, muddy results that often lead users to regenerate repeatedly. One-line example: “soft golden hour side light, long shadows, warm highlights.”

Mood and framing complete the picture. Mood is the companion’s expression, posture and energy — playful, contemplative, coy — and framing is lens distance and angle: close-up, three-quarter, full body, eye-level, Dutch tilt. These choices control emotional impact and how the image will be cropped in your gallery; the platform saves images to where your images are kept. One-line examples: “gentle, teasing smile” and “three-quarter view, slightly above eye level.” If you’re unsure which art approach suits the idea, look through all eight styles and the posts on choosing an art style; try direct styles like realistic style, soft options like watercolour style or moodier choices such as cyberpunk style. Clear, precise prompts save credits and produce images that match the companion in your head.

Wardrobe: the detail that changes everything

Clothing is the single detail that signals era, social role and mood at a glance. A sundress and denim jacket say casual afternoon; a silk slip and opera gloves read as evening formality. When you write an ai girlfriend image prompt, make wardrobe your shorthand for story — name the garment, the fit, the fabric and a small note about wear or condition. “Wool coat, slightly worn at the cuffs” implies a lived-in character. “High-shine leather jacket, pristine” implies someone more polished. These specifics guide the image generator away from bland, generic outputs and towards a consistent character you recognise.

Think in layers: silhouette, material, and small props. For casual looks, specify silhouette and texture — “oversized knit sweater, soft cotton, relaxed fit” — and add a prop only if it matters, such as “holding a coffee cup.” Eveningwear needs notes about cut and embellishment: “floor-length satin dress with a thigh slit, low back, subtle beadwork.” For period clothing, name the decade and give structural details — “1940s tea dress, nipped waist, cap sleeves, floral print” — and consider including hair and accessory cues so the generator places the outfit in the right era. Fantasy wardrobes require rules: choose whether fabrics are grounded (‘suede, hand-stitched seams’) or fantastical (‘luminescent scale armour’), because vague fantasy prompts often produce inconsistent results.

Failure modes that waste credits are predictable. Vague clothing prompts produce multiple regenerations: “dress” can mean anything. Contradictory notes confuse the model — “Victorian corset, modern denim jeans” without a reason for the mismatch will generate odd hybrids. Overloading a single prompt with too many outfits causes inconsistent framing and costume errors. If you’re experimenting, practice with a small set of variations and use the free messages or check what generations cost to understand credit usage before buying large packs. Save versions you like to where your images are kept so you can compare and refine rather than repeating the same mistakes.

Examples work best when concrete. Casual: “light-wash high-waisted jeans, tucked-in striped tee, thin leather belt, white trainers, wind-blown hair” — this sets movement, formality and era. Evening: “deep emerald velvet off-shoulder gown, elbow-length gloves, sculpted bun, low warm key light” — that tells the generator about fabric, accessories and lighting. Period: “Regency gown, high empire waist, delicate muslin, reticulated hair accessory” signals an early 19th-century look. Fantasy: “leather corset with brass filigree, hooded cloak with embroidered constellations, faint bioluminescent trim” gives both style and internal logic.

Once you have a clear wardrobe, pair it with an art style and framing decision. If you want a lifelike portrait, choose the realistic style; for soft, painterly results try watercolour style. For neon-lit, futuristic garments use cyberpunk style. The platform lists all eight styles and a short guide on choosing an art style to help match clothing to look. If you plan motion, design the outfit with video scenes in mind: avoid elaborate flowing fabrics that will animate poorly unless you want the deliberately stylised motion the tool produces. Lastly, practise with a few of the companions to practice with to see how different system prompts influence how clothing reads on-screen.

Setting and light: writing a room in one sentence

The easiest way to visualise a scene is to treat the setting and the light as a single sentence: "a small sunlit kitchen at breakfast" is already a full stage direction, whereas "kitchen, morning" is not. Start by deciding interior or exterior. Indoors you must specify practical light sources — a window, pendant lamp, bedside lamp — because those anchors tell the generator where shadows fall and where highlights belong. Outdoors, name the time of day and the weather: "late-afternoon golden hour on a cobbled street" signals warm, directional light and long shadows; "overcast park" signals soft, diffused light and low contrast.

Time of day changes the colour temperature and the length of shadows. Morning and late afternoon produce warm, low-angle light; noon is high and harsh; twilight pulls blues and softens contrast. When you say "sunset" the generator will shift skin tones and background hues toward oranges and purples. If you want a cooler result, say "blue hour" or "misty dawn". If you omit time, the model often defaults to a neutral, flat daylight that makes faces look washed out — a common failure mode that wastes credits because you have to regenerate or refine the prompt.

Practical light sources are the most reliable way to avoid flat images. Naming a lamp, window or neon sign gives the model a concrete cause for highlights and shadows. For instance, "soft window light from the left, sheer curtains diffusing sunlight" is much more effective than "soft light." It tells the generator where the light is coming from and what quality it has. Similarly, "warm bedside lamp casting a cone of light" creates a localised pool of warmth and natural fall-off on the subject. When you want mood, name the fixture: fairy lights, a candle, a streetlamp, a computer monitor — each implies a different spread, intensity and colour.

Failure modes that waste credits usually come from mixed or missing cues: saying "nighttime kitchen with bright sunlight" is contradictory and often produces muddled, low-quality results. Vague modifiers like "cozy lighting" are subjective; pair them with a practical source: "cozy lighting from a small Edison bulb." Also watch for conflicting sources that cancel each other out — if you list multiple lights, clarify which is dominant. If you plan to make a short clip rather than a single image, the same rules apply for video scenes: consistency of light across frames matters because the animation is generated from stills.

Finally, the art style you choose interacts with the lighting. A request for realistic style benefits from precise, physical light sources; a washier watercolour style can tolerate softer, less defined lighting; a neon-heavy cyberpunk style calls for coloured rim lights and reflective surfaces. If you’re still experimenting, try a few companions to practice with and save the best outputs to where your images are kept so you can compare how different lighting phrases affect the same pose. Clear, concrete light directions are the fastest path from a flat image to one that reads as a real place.

Framing and distance: portrait, half-body, full scene

Framing decides what the image is for before you write a single adjective. A tight portrait is for expression: eyes, skin texture, hair, and the small details of costume and make-up matter. Say “close-up portrait, eye-level, shallow depth of field” when you want the companion’s face and expression to be the subject. That makes shorter prompts because the generator focuses on facial features and lighting rather than furnishing or background action. If you want specific hair strands, freckles, or a visible lip gloss catch, call them out. Portraits also save credits compared with asking for a wide scene, because there’s less to generate and less room for contradictory details.

Half-body or waist-up framing brings outfit and gesture into play. It’s a good compromise when you want to see hands, an accessory, or how a jacket sits without describing an entire environment. Use measurements like “waist-up, natural posture, hands visible, holding a cup” and pair that with lighting and mood. If you’re practising prompts, pick one of the companions to practise with so your phrasing is consistent; the list of available companions to practice with shows what personalities and baseline looks you can aim for. Remember: the image generator will place attention where you place it. If you neglect to say “hands visible,” they may be cropped out or ambiguous.

A full scene requires a wider description of place, activity and secondary characters — and that equals more credits when you go beyond the daily allowance, so be specific to avoid wasted generations. For a full scene, describe the location, time of day, and any objects that define the story: “full-scene, coastal café at golden hour, table with two steaming cups, wind-blown hair, seagulls distant.” A wide frame needs consistent spatial cues (flooring, windows, furniture scale) so the companion sits believably in the space. If you plan to turn a scene into a short clip later, think ahead: video scenes are generated by animating still frames, so a well-composed full-scene prompt translates better when you export the frames for motion.

Different art styles change how much detail you must supply. In realistic style, textures and accurate anatomy matter, so call out materials and light behaviour explicitly; if you prefer a softer, illustrative result, a watercolour style or cyberpunk style will interpret details differently, and you can lean on the style’s conventions rather than micromanage every fabric fold. The page on choosing an art style compares the eight approaches and helps you decide whether to let the style carry mood or to specify it yourself; you can also browse all eight styles to see what fits. If you’re unsure how many credits a wider scene will use, consult what generations cost so you don’t exceed your allowance unexpectedly.

Finally, framing affects where your images end up and how you reuse them. Close-ups and portraits are easier to crop for avatars; full scenes are better for storyboards or downloading to your gallery for later editing. You’ll find your images kept in your personal gallery, which makes it simple to review previous attempts and iterate rather than spend credits repeating the same mistakes.

Art style: the setting you choose once

The art style you pick is the single choice that changes everything else in a prompt. Saying “watercolour” or “cyberpunk” isn’t a tiny tweak to lighting — it reshapes texture, line work, colour palette, level of detail, and even which wardrobe choices look credible. For example, a request that works well in realistic style — close fabric detail, skin texture, natural fall of light — will produce images that look wrong if you paste the same words into a watercolour style prompt, where soft edges, pigment blooms and simplified colour fields are the expectation. Likewise, cyberpunk style invites neon rim-lighting, reflective surfaces and urban clutter; if your subject line still asks for “soft, intimate bedroom lighting,” the generator will try to reconcile two incompatible directions and the result will feel indecisive.

Think of style as the setting you choose once for a set of related images. If you’re experimenting, try a single companion and run a few prompts in one chosen mode; companions to practice with are there so you can test voices and looks without committing credits. Keep the core elements — subject, wardrobe, setting, light, mood, framing — aligned to that style. For example, in a realistic portrait you specify fabric weave and skin undertone; in an illustrated or anime prompt you swap those details for line weight, cell shading or expressive eyes. The platform lists all eight styles so you can compare them side by side before you start rewriting prompts.

Failure modes that waste credits usually come from changing style mid-series or mixing incompatible visual cues. Common problems: the outfit flips between photoreal textures and flat colours; a single image set contains both highly detailed facial pores and cartoon-style eyes; or lighting directions contradict the chosen environment. When the model is asked to satisfy contradictory constraints it often produces muddled results — and if you’re buying extra runs, that costs money. Review what generations cost before committing to many retries, and favour iterations within one consistent style to refine rather than reset.

There are legitimate reasons to switch styles: you want a different emotional tone, you’re planning a short animated beat for video scenes using stylised frames, or you’re assembling a mixed-media gallery. If you plan a project that includes both portrait stills and an animated clip, pick a dominant style and note which elements should translate when frames are animated. Your images save to your personal space — see where your images are kept — so consistency makes an album feel cohesive. If you’re unsure which style suits a character, our primer on choosing an art style helps you narrow down options before you spend credits changing direction.

Finally, think in runs: one style per scene or series, small controlled tweaks within that style, then a deliberate switch only when you know what you want to explore next. The image generator will follow the style you specify, but it won’t decide for you. Consistency saves credits, creates recognizable imagery, and makes it easier to spot what in your prompts is actually broken rather than simply mismatched to the art style.

Keeping her face consistent across a gallery

Keeping a recognisable face across a gallery is one of the hardest parts of making a believable AI girlfriend series. The single most useful thing you can do is lock in a consistent art direction up front. Choose an art family and stick with it — whether that’s the realistic style, watercolour style or something bold like cyberpunk style — and repeat that choice in every prompt. Consistency in style reduces one major source of drift: when the model is asked to switch between visual grammars it often reinterprets facial features, proportions and texture, so the “same” face ends up different.

Next, be precise and consistent about hair and facial features. Short, repeatable phrases work better than long paragraphs that try to reinvent the description each time. Define length, parting, colour, and any signature details you expect to recur: “shoulder-length wavy chestnut hair with a centre part, light freckles across the nose, prominent left eyebrow scar” is more reliable than “pretty girl with brown hair.” Use the same wording across prompts rather than paraphrasing; small changes in language can nudge the generator toward a different look. If you keep reusing a companion’s defining traits, you’ll see a higher rate of likeness.

Wardrobe callbacks help the eye accept small differences. You can’t force an identical facial mesh across every generated image, but repeating accessories or clothing details — a silver hoop earring, a specific pendant, a knitted cardigan pattern — creates a visual anchor. Short callbacks are enough: a brief “wearing a blue knitted cardigan; small silver hoop on left ear” at the end of a prompt will guide the generator without overconstraining pose or setting. That approach also makes it easier to animate a portrait later when you put frames together into video scenes from the same gallery.

Even with consistency in style, hair and wardrobe, some failure modes persist. The model can still alter jawline, eye spacing or expression between generations. Lighting and pose are common culprits: an extreme side light or a three-quarter pose tends to hide or exaggerate features, breaking likeness. Very different camera angles are also problematic; keep framing consistent — for example, three-quarter bust portrait at eye level — if you need a series of matching faces. You’ll waste fewer credits if you avoid repeatedly asking for wildly different angles and lights when your goal is likeness; see what generations cost so you don’t burn through credits on retries.

Practical workflow tips: save successful variants to where your images are kept as soon as you get them, and use those as visual references when composing new prompts. Practice with a few companions to practice with or in a single art licence, and try switching among all eight styles only when you’re experimenting — or consult guidance on choosing an art style to match mood and technical needs. Consistency won’t guarantee pixel-perfect identity, but it will give you a recognisable, coherent gallery much more quickly.

Failure modes that waste credits

The most common way credits are wasted is a simple visual error: hands that look wrong. Fingers might be fused, too many, or misshapen. Fix this by including a short, explicit clause: "natural hands, five fingers visible, no extra digits, realistic hand poses." If you’re experimenting with a specific hand position (holding a cup, brushing hair), describe it and add a synonym or two — for instance "gently holding a mug, fingers wrapped naturally" — so the model has alternatives. If you’re unsure which phrasing works, try the same prompt with different companions to compare results; practising prompts with a couple of companions to practice with lets you iterate without spending credits blindly.

Crowds and unintended people in frame are another credit sink. Prompts that mention "a busy cafe" or "party" will often produce extra faces and bodies you didn’t ask for. If you want a single-subject portrait, say "solo subject, no other people in frame." For controlled background activity, specify "bokeh background, indistinct shapes, no other recognizable people." If you’re planning to turn stills into short animations, remember that Amora’s video scenes are generated by animating still frames. That process can exaggerate stray figures, so clear the frame at the prompt stage.

Text appearing in the image — signs, logos or random words — is a persistent failure mode. Optical character generation tends to be unreliable and often produces meaningless strings. If you don’t want any text, add "no text, no logos, no watermarks." If you do need legible writing (a book cover, a sign), accept that it’s likely to be imperfect and budget for multiple attempts; the page explaining what generations cost will help you estimate how many credits you might spend trying to get a crisp label.

Mirrors and reflections are practical troublemakers. They can double the subject in odd ways or produce impossible reflections. If your scene includes reflective surfaces, prompt for "accurate reflections" and describe the relative positions, or avoid mirrors entirely. For glassy or metallic details, specifying "soft reflections, consistent with camera angle" reduces contradictions. If you want an illustrated or stylised reflection, choosing a specific approach from all eight styles — for example the cleaner lines of realistic style or the painterly approach of watercolour style — helps the generator follow a consistent visual rule.

Food and motion both cause artifacts: messy textures, floating crumbs, and blur that looks like a rendering error. For food, be precise — "clear plate, single slice of cake, no crumbs" — and for motion, decide whether you want freeze-frame sharpness or motion blur. If you want a short moving clip, remember the platform animates generated frames rather than filming continuous motion; this makes large, complex movements harder to render cleanly. Choosing a stable pose and letting subtle movement carry the scene will usually yield better results than demanding dramatic action. When prompts get long and contradictory, the model can ignore earlier instructions; prune to the essential five elements — subject, wardrobe, setting, light, mood — and, if you need a certain aesthetic, point to the guidance on choosing an art style such as cyberpunk style. Store successful images in where your images are kept so you can reuse or refine them rather than re‑spending credits on repeats.

A prompt library: six worked examples

1. Prompt — soft morning portrait, casual home

"Subject: young woman, mid-20s, warm olive skin, loose shoulder-length brown hair, subtle freckles, relaxed smile. Wardrobe: oversized cream knitted sweater slipping off one shoulder, simple gold pendant. Setting: sunlit bedroom, unmade bed, potted fern on windowsill. Light: warm, morning sunlight through sheer curtains, soft shadows. Mood: intimate, candid, comfortable. Framing: waist-up, slight tilt, shallow depth of field, eyes near upper third. Style: realistic, natural skin texture, minimal retouching." Why it works and what to expect: specific age range, skin cues and clothing give the generator clear anchors. Asking for "realistic" steers toward that aesthetic — see the realistic style — and reduces ambiguity about anime or cartoon looks. Expect a polished portrait with gentle lighting and some background detail. Failure mode: vague wardrobe or "make her sexy" prompts produce inconsistent outfits and wasted credits.

1. Prompt — city-night cyberpunk rendezvous

"Subject: woman, late 20s, short silver hair, cybernetic temple implant, confident expression. Wardrobe: high-collared black coat with neon blue trim, fingerless gloves. Setting: rainy neon street, reflective puddles, holographic signage. Light: high-contrast neon highlights, cool blue and magenta rim light. Mood: mysterious, cinematic. Framing: three-quarter view, full body, slight low angle. Style: cyberpunk, gritty textures, rain streaks." Why it works and what to expect: naming neon colours, specific clothing accents and a clear environment directs the generator toward the cyberpunk style and keeps results consistent. Expect stylised, high-contrast still images with visible rain and reflections. If you want motion instead of a still, plan a short storyboard with the video scenes tool — note the platform animates generated stills rather than making a photoreal film.

1. Prompt — playful anime picnic

"Subject: young woman drawn in anime style, long black hair in twin tails, playful wink. Wardrobe: sailor-collar blouse and pleated skirt, pastel palette. Setting: sunny park blanket with picnic basket and cherry blossoms. Light: bright, gentle sunlight, slight bloom. Mood: cheerful, lighthearted. Framing: seated, knees to camera, waist-up. Style: anime, soft linework, warm colours." Why it works and what to expect: explicitly calling "anime" and listing props avoids generic or hybrid outputs. Expect clear, stylised characters with simplified anatomy and vibrant colours. If the result leans too cartoonish, switch to another art style — the guide to choosing an art style covers trade-offs between styles and realistic or watercolour options.

1. Prompt — elegant watercolour study

"Subject: woman in her 30s, long auburn hair in loose braid, serene expression. Wardrobe: flowing teal dress, minimal jewellery. Setting: window seat with tea cup. Light: soft diffuse daylight. Mood: contemplative, gentle. Framing: three-quarter portrait. Style: watercolour, visible brush texture, muted palette." Why it works and what to expect: asking for "watercolour" signals texture and colour limits; the watercolour style emphasises washes and brush edges. Expect painterly softness rather than photographic detail. Failure mode: requesting photoreal detail together with watercolour will confuse the output and usually wastes a credit.

1. Prompt — noir evening cigarette scene

"Subject: woman in her early 30s, cropped dark hair, strong jawline. Wardrobe: vintage trench coat, silk scarf. Setting: rain-slick alley, dim streetlamp. Light: hard directional light, high contrast, heavy shadow. Mood: moody, film-noir. Framing: waist-up, cigarette in hand, smoke curling. Style: noir, monochrome, film grain." Why it works and what to expect: noir tonal directions and props (cigarette, scarf) focus the generator on a specific aesthetic. Expect dramatic black-and-white stills with strong shadow. Avoid generic "moody" alone; it’s too vague and often produces disappointing images.

1. Prompt — cosy home chef close-up

"Subject: woman mid-30s, curly hair tied back, warm expression. Wardrobe: faded denim apron over shirt. Setting: kitchen counter with chopping board and herbs. Light: warm late-afternoon window light, soft rim. Mood: inviting, domestic. Framing: close-up hands at work, shallow depth of field. Style: realistic, natural colours." Why it works and what to expect: coupling action (hands at work) with a close framing reduces face ambiguity and yields a natural candid feel. Expect detailed stills showing food textures and hands. If you plan multiple moments, consider creating several frames and animating them with the video scenes feature; remember image generations cost credits — check what generations cost — and saved images go to your gallery so you can review variants later. If you want practice, try prompts with one of the companions to practice with and iterate quickly.

From image to scene: planning before you spend credits

Before you spend credits, plan the scene the way you’d sketch a short film. A single generated image costs 5 credits and your free daily allowance includes six images, so wild experimentation can be done without paying — but those free images still run out. A video scene costs 45 credits, so attempting a scene without a clear storyboard is the quickest way to waste a large chunk of credits. Check what generations cost if you’re unsure of the arithmetic for a session, and remember that the platform keeps track of your outputs in where your images are kept, which makes reviewing earlier drafts easier than recreating them from scratch.

Start by deciding whether you need a single image or a short animated scene. Use the image generator for a one-off portrait, and reserve video scenes for moments that benefit from movement — but plan for the technical reality: video scenes are created by generating still frames and animating them, not by recording live-action footage. That process means you’ll want to storyboard each key frame so the transitions make sense. A rough four- or six-panel storyboard will often reveal problems you’d otherwise only see after spending 45 credits.

When you storyboard, break the scene into subject, wardrobe, setting, lighting, mood and framing. Write a one-line intent for each frame: what the companion is doing, how her outfit reads from the chosen angle, and what the light should highlight. If you’re experimenting with appearance, try a couple of art styles on the same storyboard to compare mood: the difference between realistic style, watercolour style or cyberpunk style can be dramatic. The site lists all eight styles and a longer guide on choosing an art style if you want a quick reference before you commit.

Use small, focused tests to validate high-risk elements. If a particular pose or expression repeatedly fails, spend one image credit on a cropped, close-up test of just the face and another on the full-body shot; that tells you whether the model is misunderstanding expression or clothing. If you’re learning, practice with the free companions — there are many characters to try — so you can iterate without draining credits. The companions to practice with page is a good place to pick a familiar face for repeated tests.

Finally, keep notes in the caption or a separate prompt template so you can reuse successful phrases for wardrobe, lighting or framing. Planning forces you to convert vague wishes into concrete prompt elements, and that precision is what saves credits. Be realistic about limits: planning reduces wasted generations but cannot guarantee a perfect result every time, and video scenes will never be a live-action cinematic recording — they are animated from generated stills. If you keep a simple storyboard habit, you’ll spend fewer credits and get images and scenes that match your vision more often.