Riverflow Prompting Guide
This guide helps you decide when to prompt, when not to prompt, and how to write prompts that give Riverflow the clearest possible direction.
Choose Your Workflow
Riverflow has two different prompting workflows.
Use Photoshoots when you want Riverflow to mix a product, scene, style, and prompt together. In Photoshoots, your prompt is mainly there to steer the mixing process: what to preserve, what to change, what to borrow from the style, and what must stay accurate.
Use Images when you do not want to use a scene. In Images, your prompt works more like traditional AI image prompting: you describe the whole image from scratch, including subject, setting, composition, camera, lighting, style, people, materials, and realism.
Simple rule:
Photoshoots: prompt less, steer more.
Images: write a complete creative brief.
If You Only Remember Five Things
In Photoshoots, try generating 4 candidates without a prompt when the scene and style are already close.
For simple changes, prompt only the 1-2 things you want to change and what must stay the same.
For product accuracy, explicitly protect dimensions, label, logo, colors, text, and proportions.
For style references, say exactly what to use: full style, lighting only, texture only, mood only, or camera feel only.
In Images, write a complete creative brief: subject, setting, composition, camera, lighting, style, people, and realism. When the image is mostly right, use edit plus reference images to refine details.
Commercial Prompting Principles
Use these rules whenever the image needs to be good enough for a campaign, website, PDP, ad, email, or brand asset.
Make the product hierarchy clear. Say whether the product is the hero, a supporting prop, in-hand, foreground, background, sharp, or softly blurred.
Protect commercial details. Product shape, proportions, logo, label, flavor name, packaging text, materials, and scale should be explicit when they matter.
Write visual instructions, not abstract intentions. "Soft directional daylight from camera-left" is better than "premium lighting." "Candid table-level framing" is better than "make it cool."
Avoid conflicting cues. Do not ask for an iPhone candid photo, studio campaign lighting, disposable flash, and 85mm portrait bokeh all in the same prompt unless that combination is intentional.
Limit each generation to one clear concept. If you need many changes, generate the core image first, then use edit plus reference images for refinements.
Use references with a job description. Say whether each reference is for product detail, sizing/proportions, human appearance, lighting, texture, or mood only.
Keep avoid rules focused. Only mention failure modes that would make the image unusable: wrong label, warped product, distorted hands, plastic skin, fake text, impossible reflections, or copied reference content.
Part 1: Photoshoots — Guide The Mixing Process
In Photoshoots, some of the best results come from not prompting at all.
If your selected scene, style, and product are already close to what you want, try generating 4 candidates without a prompt first. Let Riverflow adapt the product into the scene, then choose the best result and only prompt if you need a specific correction.
Use no prompt when:
you want to explore ideas quickly
the scene already has the right composition
the style already has the right lighting or mood
you are happy for Riverflow to make creative choices
you want to compare 4 different interpretations before steering it
Use a prompt when:
a product detail must stay accurate
you want to change 1-2 specific things
you want to tell Riverflow which parts of a style reference to use
you want to override the scene
humans, hands, labels, logos, or packaging accuracy are important
Why Photoshoots Prompts Are Different
In Photoshoots, you usually have a scene, style, and product reference already selected. This means your prompt can often be short. You can say what to keep, what to change, and which parts of the style to use.
Photoshoots prompts often work best with clear steering language such as MUST PRESERVE, MUST CHANGE, and MUST NOT, because Riverflow is adapting an existing scene.
Photoshoots Prompting Routes
Choose the smallest amount of prompting that matches the job.
Route 1: No Prompt
Best for early exploration in Photoshoots.
Leave the prompt blank. Generate 4 candidates.
Use this when you trust the scene and style, and you want Riverflow to adapt the product naturally.
Route 2: Simple Change Prompt
Best when changing 1-2 elements.
Keep [what should stay the same]. Change [1-2 specific things]. Do not change [important product or scene details].
Example:
Keep the same camera angle, tabletop, lighting, and product position. Replace the orange slices with sliced pear. Keep the product label, shape, color, and dimensions exactly as shown.
Route 3: Modular Prompt
Best when you need more control but do not want to write a full brief.
Combine 2-5 blocks from the prompt assembly system below, such as:
Scene:
Keep the original camera angle, product position, surface, and overall composition.
Product accuracy:
MUST keep the product dimensions, shape, label, logo, colors, and packaging text exactly as shown in the reference.
Change:
Replace the background props with fresh citrus, ice, and condensation. Keep everything else natural and realistic.
Style:
Use only the lighting and image texture from the style reference. Do not copy its props, background, model, or composition.
Route 4: Full Prompt
Best for complex Photoshoots, major scene rewrites, humans, or campaign-style visuals where the selected scene is only a starting point.
Use the Photoshoots prompt assembly system below. If you are not using a scene at all, skip to Part 2: Images.
Photoshoots: 3 Decisions Before You Prompt
Before writing anything, decide:
What should stay the same?
What should change?
What should Riverflow borrow from the scene or style?
Most poor results come from skipping one of those decisions.
Photoshoots Prompt Assembly System
Think of a prompt as a set of modules. Most users only need 2-4 modules at once.
This structure is especially useful in Photoshoots, where you are steering an existing scene. In Images, you can still borrow these modules, but the prompt should usually read more like a full image brief than a list of edit commands.
Use this order when a prompt needs more control:
Goal:
[What you want Riverflow to make or change.]
MUST PRESERVE:
[What should stay from the scene, product, or style.]
MUST CHANGE:
[The specific things Riverflow should alter.]
MUST HAVE:
[The required visual details for product, person, setting, styling, camera, materials, or color.]
STYLE / CAMERA:
[What look, lighting, texture, or camera feel to use.]
MUST NOT:
[Optional. Common failure modes to avoid.]
Use MUST PRESERVE when the source image already has something valuable. Use MUST CHANGE when only 1-2 things should be different. Use MUST HAVE when an output requirement is non-negotiable. Use MUST NOT only for likely failure modes.
The main prompt modules are:
Product fidelity: shape, label, logo, dimensions, packaging text, material, reflections.
Extra product references: alternate angles, underside views, back labels, side labels, open packaging, or close-up details.
Sizing references: hand-held images, range lineups, or scale references used only to understand relative product size and proportions.
Scene control: preserve the source scene, lightly edit it, use it for mood only, or override it.
Style scoping: use the full style, or only lighting, texture, color grade, mood, or camera feel.
Camera and composition: angle, framing, lens perspective, depth of field, crop, subject priority.
Materials and texture: condensation, frost, glass, metal, fabric, ceramic, fruit, liquid, skin.
Color and brand: exact hex colors, brand palette, background color, saturation level.
People: identity, age, pose, expression, hands, skin realism, styling, demographic specificity.
Edit refinements: small corrections using the edit function plus reference images for exact details.
Optional guardrails: what to avoid if Riverflow keeps making the same mistake.
Example Assembly
Goal:
Create a realistic lifestyle product image using the selected scene.
MUST PRESERVE:
Keep the source scene's camera angle, soft lighting, seated pose, teal background, and relaxed product-holding gesture.
MUST CHANGE:
Replace the person styling with mature urban high-fashion clothing and a short cropped bob.
MUST HAVE:
The person should look realistic and age-appropriate, with natural skin texture, subtle facial asymmetry, natural teeth, realistic hands, and fabric texture. The product label should remain readable.
STYLE / CAMERA:
Match the source scene's studio portrait perspective, framing, lens feel, and soft shadow falloff.
MUST NOT:
Do not use plain catalogue clothing, waxy skin, perfect symmetry, overly white teeth, or a front-facing passport-photo pose.
Photoshoots: Scene, Style, and Prompt
In Photoshoots, Riverflow usually works from:
Scene: the starting image, layout, setting, product placement, people, props, and composition.
Style: the visual look, such as lighting, camera feel, texture, color treatment, mood, or art direction.
Prompt: your written instruction. The prompt tells Riverflow what to keep, change, borrow, or ignore.
The scene is only a starting point. Your prompt can override the scene completely if you want it to.
Scene override example:
Use the scene only as a loose starting point. Create a new image with the product on a bathroom shelf in soft morning light. Do not preserve the original background or props.
Scene preservation example:
Keep the original camera angle, product position, tabletop, lighting direction, and overall composition. Only replace the background props with fresh citrus and ice.
Scene mood-only example:
Use the source scene only for the sunny outdoor mood, casual lifestyle feeling, natural lighting, and realistic camera texture. Do not preserve the exact people, props, product, layout, or action.
Shared Prompt Modules
Use these blocks as building pieces in Photoshoots or Images. You do not need all of them. Pick the ones that match the result you want.
Product Accuracy
MUST keep the product dimensions, shape, label, logo, colors, packaging text, and proportions exactly as shown in the reference. Do not invent new text, claims, icons, flavor names, or label details. The product must remain readable and recognizable.
Shorter version:
Use the uploaded product exactly as-is. Do not change its shape, color, label, logo, dimensions, or typography.
Label Readability
Keep the product label facing camera and readable. Preserve the logo, typography, flavor name, label layout, and visible packaging text. Do not invent, blur, warp, crop, or simplify label details.
Additional Product Reference Images
Use this when you upload more than one image of the same product, such as a different angle, back label, side profile, shoe sole, pack opening, texture close-up, or cap detail.
Use product reference 2 as an additional product-detail reference only. It shows [specific detail: side profile / back label / underside of the shoe / cap shape / material texture]. Use it to improve product accuracy for that detail. Do not copy the scene, background, lighting, hand, props, or camera angle from product reference 2.
Example:
Use product reference 2 as a reference for the sole of the shoe only. Preserve the sole shape, tread pattern, material texture, and proportions shown in product reference 2. Do not copy the background, lighting, floor, or camera angle from product reference 2.
Sizing Reference Images
Use this when you upload a reference image only to show scale and proportions, such as a product in a hand, a bottle next to a face, a pack on a shelf, or multiple products from the same range together.
How to use a sizing reference:
Go to the Products area.
Edit the product.
Upload the sizing reference as an extra product image.
In generation, select the main product image and the sizing reference image.
In the prompt, tell Riverflow which selected product reference image is for sizing/proportions only.
MUST use product reference 2 as a sizing reference ONLY. It is used to show the relative size and proportions of the product compared with [a human hand / a face / the other products in the range / the surrounding props]. Do not copy the scene, background, lighting, hand pose, model, styling, or product placement from product reference 2.
Example:
MUST use product reference 2 as a sizing reference ONLY. It shows the relative size and proportions of the can compared with a human hand. Keep the can at a believable hand-held scale in the final image. Do not copy the hand, background, lighting, or camera angle from product reference 2.
Product range example:
You MUST use product reference image 3 as a size and proportions reference only. It is used to set the relative sizing and proportions of product 1 and product 3. Do not copy the background, lighting, camera angle, or product placement from product reference image 3.
Preserve The Scene
Keep the original scene, camera angle, product position, surface, lighting direction, and overall composition. Only change the elements I explicitly mention.
Override The Scene
Do not preserve the original scene. Use the product reference, but create a new image with the setting, lighting, composition, and props described below.
Use A Style Reference
Full style:
Use the full style reference, including lighting, composition, image texture, color grade, mood, and camera feel.
Specific style:
Use only the lighting and image texture from the style reference. Do not copy its props, background, model, product, clothing, or composition.
Mood only:
Use the style reference for mood only. Keep my scene and product placement.
Exact Brand Color
Make the background #F40009.
Hex colors are especially useful for studio backgrounds, brand blocks, and campaign-style images.
Camera And Composition
Match the source scene's camera angle, lens perspective, framing, lighting direction, and depth of field. Keep the product as the clearest subject in frame.
For Images or from-scratch prompts:
Create a realistic [studio product photo / lifestyle image / editorial portrait] shot with a [camera type] and [lens type], using [lighting style], [framing], realistic exposure, and true-to-life texture.
Materials And Texture
Make the materials feel physically real: accurate reflections, surface texture, contact shadows, natural imperfections, and believable scale. Preserve the texture of [ceramic / glass / aluminum / fabric / fruit / liquid / wood / paper / plastic] without making it overly glossy or AI-smooth.
Food, Drink, And Freshness
Make the food and drink styling realistic and fresh: natural surface moisture, believable condensation, accurate liquid reflections, imperfect fruit texture, realistic ice, and true-to-life color. Avoid plastic-looking food, fake shine, impossible splashes, or over-styled props.
Realistic Human In The Background
Include one softly blurred person in the background, out of focus and indistinct, with no clear facial detail. They should feel like a natural lifestyle presence, not the main subject.
Hands Holding Product
Show one person's hand holding the product naturally. The grip should be relaxed, fingers anatomically correct, and the product label readable. Use natural skin texture and avoid overly smooth or plastic-looking hands.
Hero Person
Include one realistic person as the main subject. Use natural skin texture, subtle facial asymmetry, realistic posture, normal facial variation, and a relaxed expression. Avoid airbrushed skin, plastic-looking features, exaggerated beauty retouching, and overly perfect symmetry.
Human Realism Booster
Use this when the person is almost right but still looks too smooth, perfect, waxy, or synthetic.
Make the person feel like a real photographed subject, not an idealized AI portrait. Add about 10% more natural realism: fine skin texture, subtle pores, slight tonal variation, gentle expression lines, realistic lip texture, believable glasses or eye reflections, natural facial asymmetry, and small differences between the two sides of the face. Teeth should look naturally photographed, with slight variation in shape, spacing, brightness, and alignment. Keep the result flattering and friendly, but not airbrushed, waxy, porcelain-smooth, overly symmetrical, or beauty-retouched.
Natural Pose Asymmetry
Use this when a person looks too front-facing, centered, or posed.
Add subtle natural pose asymmetry: a gentle head tilt, relaxed uneven shoulders, a slight casual lean, and a smile that feels mid-moment rather than perfectly posed. The pose should feel candid and human, not front-facing, perfectly centered, or passport-photo symmetrical.
Demographic Specificity
Use this when age, region, or representation matters. A broad label alone is often too weak.
The person should be a realistic [region or heritage] [gender/person] around [age]. Add respectful, concrete cues: [skin tone or undertone], [hair color and texture], [eye color], [styling context], and [overall character]. Keep the result specific and realistic without stereotypes, costume, caricature, or generic stock-photo styling.
When changing the person in an existing scene, be extra clear that the person should change while the useful parts of the scene should stay.
MUST CHANGE the person's visible facial features, age cues, hair, and styling so they clearly match the description below. MUST PRESERVE the source scene's pose, product-holding gesture, camera angle, lighting, background, and composition.
Photoshoots Recipes
Use these as starting points and swap in your own product, scene, or style details.
Product Swap
Keep the scene, lighting, camera angle, and product placement the same. Replace this product with Chaga. Replace the lion's mane mushroom with a chaga mushroom. Keep all other props and background details unchanged. Keep the new product dimensions, label, logo, and packaging text accurate to the reference.
Small Prop Change
Keep the camera angle, product placement, lighting, and background the same. Make the petals be sliced pear. Do not change the product, label, surface, or overall composition.
Brand Color Studio Background
Keep the cans exactly as shown, including shape, dimensions, labels, colors, and logo placement. Place the Coca-Cola Zero Sugar can upright and the Coca-Cola Classic can lying down. Add realistic condensation on both cans and small pieces of ice around the top edges and bases. Make the background #F40009.
Style Reference Without Copying The Scene
Use only the soft side lighting, realistic image texture, and warm color tone from the style reference. Keep my product, label, camera angle, and scene layout. Do not copy the reference props, background, person, outfit, or product.
Lifestyle Product With Background Person
Create a realistic lifestyle product image with a natural smartphone photography look.
Product accuracy:
Use the uploaded product exactly as-is. Do not change the shape, color, pattern, label, logo, dimensions, or typography.
Scene:
Place the product on a light wood dining table in a clean, bright kitchen or dining environment.
Composition:
Close-up framing at a slight 45-degree angle, casual real-life placement, not overly styled.
Lighting:
Early morning gentle sunlight from the side, soft warm daylight, soft highlights, and mild natural shadows.
People:
Include one softly blurred person sitting near the table in the background, out of focus and indistinct, with no clear facial detail.
Realism:
Natural color rendering, true-to-life textures, realistic exposure, no artificial sharpness, no studio lighting feel, no AI artifacts.
Person Holding Product
Create a candid lifestyle image of one person holding the product casually at chest height.
Product accuracy:
MUST keep the product dimensions, shape, label, logo, colors, and packaging text exactly as shown in the reference. The label must remain readable.
Human realism:
Use natural skin texture, relaxed posture, realistic hand anatomy, subtle facial asymmetry, and a calm everyday expression. Avoid airbrushed skin, plastic-looking hands, exaggerated beauty retouching, and overly perfect symmetry.
Composition:
The product should be clearly visible and in focus. The person can be slightly less sharp than the product if needed.
Lighting:
Natural window light or soft daylight, with believable shadows and realistic exposure.
Complex Lifestyle Scene Rewrite
Use this when the selected scene has the right mood, but you want a different setup.
Create a realistic outdoor lifestyle image of [number] friends [activity with product].
MUST PRESERVE:
Use the source scene only for [mood / lighting / camera texture / season / setting type]. Do not preserve the exact people, props, product, layout, or action.
MUST CHANGE:
Change the scene to [new setup]. The product should be naturally integrated into the moment.
Product accuracy:
MUST keep the product dimensions, shape, label, logo, colors, packaging text, and proportions exactly as shown in the product reference. The main foreground product must be readable.
People realism:
Make the people feel like real photographed friends: natural poses, relaxed expressions, realistic hands, casual imperfect posture, subtle motion, and normal facial variation. Avoid waxy skin, perfect symmetry, stiff posing, plastic hands, or stock-photo smiles.
Camera / lighting:
Candid lifestyle photograph, realistic natural light, believable shadows, slightly imperfect handheld framing, natural depth of field, and true-to-life color.
Tip: multiple people, hands, and readable labels are difficult in one image. Generate 4 candidates, choose the best composition, then use edit prompts to refine details.
Part 2: Images — Prompt From Scratch
Use this when there is no scene selected. In Images, Riverflow needs a fuller brief because it is creating the whole image from scratch.
Write the prompt like a clear photography or art direction brief. Use descriptive section headings. You do not need MUST in every line. Use MUST only for true non-negotiables, especially product accuracy, label readability, sizing references, or important avoid rules.
Create a [type of image] for [product / brand / use case].
Subject: [Main product, person, or object. Say what should be most visible.]
Product accuracy:
[Describe what must stay accurate: product shape, proportions, label, logo, colors, typography, packaging text, and scale.]
Setting: [Location, surface, room, outdoor environment, background, and props.]
Composition and framing:
[Close-up, wide shot, overhead, table-level, centered, off-center, cropped, foreground/background, product placement, what is sharpest in frame.]
Camera:
[Camera feel, lens, angle, distance, and capture style: iPhone wide lens, full-frame camera, 35mm film, macro, handheld, direct flash, studio campaign, etc.]
Lighting:
[Light source, direction, softness, time of day, shadows, contrast, color temperature, reflections.]
Style:
[Clean e-commerce, candid lifestyle, editorial, disposable camera, premium studio, natural social content, etc.]
People:
[Who appears, how visible they are, pose, expression, styling, hands, and whether faces should be sharp, soft, cropped, or blurred.]
Materials and realism: [Textures, reflections, condensation, fabric, skin, food freshness, shadows, scale, imperfections, believable anatomy.]
Avoid:
[Only the most important failure modes to avoid.]
Example:
Create a photorealistic outdoor lifestyle image for a canned sparkling drink.
Subject:
A chilled can of the uploaded product is the main subject, placed on a small outdoor cafe table with two other cans nearby.
Product accuracy:
MUST use the uploaded product exactly as-is. Preserve the can shape, dimensions, logo, label colors, typography, flavor name, and visible packaging text.
Setting:
Sunny outdoor cafe table in early afternoon, casual city street background, glasses with ice, citrus slices, napkins, and light summer styling.
Composition and framing:
Close-up table-level framing at a slight 45-degree angle. The main can is foreground center-left and sharply in focus. Background is softly blurred but still feels lively.
Camera:
Natural iPhone-style lifestyle photography, wide lens feel, handheld framing, realistic exposure, no studio polish.
Lighting:
Bright natural daylight from the side, crisp but believable shadows, realistic highlights on the aluminum, slight condensation catching the sun.
Style:
Fresh, social, casual, summery, e-commerce-friendly but not overly staged.
People:
Two friends are partially visible in the background, cropped at hands and torso, casually reaching for drinks. Keep faces soft or out of focus.
Materials and realism: Realistic condensation, aluminum reflections, glass reflections, natural table texture, believable product scale, imperfect casual placement.
Avoid:
Do not redesign the product, invent label text, make the scene look like a studio render, or make the people overly posed.
Images: Commercial-Grade Prompting Checklist
For commercial imagery, a strong Images prompt should make the image easy to art direct. Include the details that affect whether the result feels usable for a campaign, PDP, paid social ad, email, or brand asset.
1. Define The Job Of The Image
Say what the image is for. This helps Riverflow choose the right level of polish.
Create a clean e-commerce lifestyle image for a product detail page.
Create a premium campaign image for a paid social ad.
Create a candid UGC-style image that feels natural and unstaged.
2. Set Visual Priority
Commercial images need a clear hero. Say what should be sharpest, largest, or most readable.
The product is the hero subject. Keep the main product sharpest in frame, with the label readable and facing camera. People and props should support the product, not compete with it.
3. Control The Camera
Camera language helps the image feel photographed rather than generated. Use camera cues that match the intended image.
Natural iPhone-style lifestyle photography, wide lens feel, handheld framing, realistic exposure.
Premium studio product photography, full-frame camera feel, 70mm lens perspective, clean focus, controlled reflections.
Editorial portrait photography, 85mm lens feel, shallow depth of field, subject in sharp focus, background softly blurred.
4. Control The Light
Lighting is often what separates commercial-grade results from generic AI images.
Soft directional daylight from camera-left, realistic highlights, gentle shadow falloff, natural ambient bounce light.
Clean studio lighting with controlled reflections, crisp product edges, soft contact shadows, and no artificial glow.
5. Add Physical Detail
Ask for material truth, not generic "high quality".
Show true-to-life material texture: aluminum reflections, natural condensation, paper label texture, fabric weave, realistic skin texture, contact shadows, and believable product scale.
6. Avoid Common AI Failure Modes
Use avoid rules sparingly. Focus on the failures that would make the image unusable.
Avoid fake label text, warped packaging, plastic-looking skin, distorted hands, impossible reflections, over-smoothed textures, over-sharpened HDR, and stock-photo posing.
Using An LLM To Write Prompts
If you want help writing a prompt, paste this guide into an LLM and ask it to write a Riverflow prompt for your specific use case.
Give the LLM:
whether you are using Photoshoots or Images
the product and what must stay accurate
any scene, style, product, human, or sizing references
the desired setting, camera, lighting, mood, and level of realism
the most common failure modes you want to avoid
Ask it to keep the prompt practical, visual, and image-focused. For Photoshoots, ask it to keep the prompt short and use MUST PRESERVE, MUST CHANGE, and MUST NOT only where useful. For Images, ask it to write a complete creative brief and use MUST only for true non-negotiables.
Shared Advanced Tools
These techniques apply to both Photoshoots and Images. Use them when the product must stay accurate, the human subject matters, or the image is close but needs refinement.
Final Refinement: Edit + Reference Images
Once an image is mostly working, use the edit function instead of starting again from scratch. This is often the best way to polish details like hair, earrings, makeup, skin finish, clothing shape, product details, or small styling corrections.
Use edit when:
the overall image is good, but 1-3 details are wrong
you want to remove or replace a specific detail
you have a reference image for a hairstyle, accessory, garment, product detail, or skin finish
you want to keep the original composition, pose, lighting, and product placement
The reference image should be treated as a detail reference, not a new scene. Tell Riverflow exactly what to use from it and what not to copy.
Edit With Reference Image
Use reference image 1 for [specific detail] only. Apply that detail to the original image. Preserve the original pose, composition, lighting, product placement, background, and overall image style. Do not copy the reference image's background, camera angle, pose, product, or full identity unless I explicitly ask for that.
Human Appearance Reference
Use this when you want the human model to have a specific appearance, demographic read, hairstyle, skin finish, or face character. Text prompts can help, but a reference image is often the strongest way to steer how the person should look.
Use reference image 1 as a human appearance reference only. Update the person in the original image so they have a similar overall demographic read, skin tone, facial character, hair color, and styling direction. Preserve the original image's pose, expression, body position, clothing silhouette, lighting, product placement, background, and composition. Do not copy the reference image's background, camera angle, pose, outfit, product, or exact identity.
Fashion Detail Refinement
Use reference image 1 for the hairstyle and earring shape only. Replace the current hairstyle with that hairstyle, and use the same earring shape but make the earrings silver. Preserve the original pose, product, pink background, lighting, clothing color, and composition.
Skin Finish Refinement
Remove the sunglasses. Make the person's skin slightly glossy and naturally photographed, similar to the skin finish in reference image 1. Keep the result realistic: natural pores, real skin texture, believable highlights, and no plastic or airbrushed look. Preserve the original pose, product position, background, and lighting.
Product Detail Refinement
Use reference image 1 for [specific product detail] only. Update that detail in the original image while preserving the original composition, lighting, hand position, product scale, and scene. Do not redesign the product or copy the reference image background.
For edits, short prompts often work best. Fix one cluster of details at a time: hairstyle and earrings, or skin finish and sunglasses, or sole detail and product scale. If you ask for too many unrelated changes in one edit, the image is more likely to drift.
Making People Look More Realistic
People are one of the hardest parts of AI image generation. The more central, sharp, and detailed a person is, the more carefully you should prompt.
Use this realism ladder:
Safest: blurred person in the background.
Usually reliable: cropped torso or hand holding product.
More difficult: full person visible in a lifestyle scene.
Hardest: sharp face, hands, readable product label, complex pose, and multiple people.
For the best results:
Use one person instead of a group.
Ask for natural skin texture, pores, subtle asymmetry, and normal facial variation.
Keep expressions simple: relaxed, neutral, soft smile, mid-conversation, looking away.
Use natural poses: leaning on a counter, sitting at a table, walking, holding the product casually.
Use real camera language: iPhone photo, 35mm film, handheld, direct flash, natural window light, slight depth of field.
Avoid words like flawless, airbrushed, doll-like, ultra-smooth, perfect face, or perfect skin.
If the product matters most, let the person be cropped, partly turned away, or softly blurred.
Full Human Realism Brief
Use this when the person is a hero subject, especially in Images or when creating a portrait from scratch. For Photoshoots, keep the camera language aligned with the source scene unless you are intentionally overriding it.
Create an ultra-realistic photograph of [subject].
Person:
[Region or heritage] [age]-year-old [gender/person], natural facial features, slight asymmetry, real skin texture with visible pores, fine lines, subtle blemishes, uneven skin tone, and no plastic smoothness.
Expression:
Natural, unposed expression with relaxed facial muscles, subtle micro-expression, and no stiff posing.
Lighting:
Soft directional light, cinematic but realistic shadows, no flat lighting, natural highlights and falloff, ambient bounce light.
Camera:
Photographed with a natural portrait perspective, 85mm lens feel, shallow depth of field, f/1.8 look, subject in sharp focus, background softly blurred with realistic bokeh.
Color and tone:
True-to-life color grading, natural contrast, slight film grain, realistic dynamic range, no oversaturation, no harsh HDR.
Environment:
[Realistic setting: office / street / home / studio / cafe / outdoor table], lived-in details, small imperfections, depth, and believable background context.
Materials:
Realistic fabric texture, individual hair strands visible, natural hair shape with no clumping, believable skin highlights, no waxy skin.
Composition:
Rule-of-thirds framing, subtle posture imperfections, natural body angle, realistic perspective, not perfectly centered or symmetrical.
Style:
Editorial photography with cinematic realism, premium but still natural and believable.
Avoid:
Over-smoothing, perfect symmetry, CGI skin, plastic skin, exaggerated features, extra limbs, distorted anatomy, overly sharp HDR, artificial glow, doll-like face, or stock-photo perfection.
If the image keeps producing fake-looking faces, reduce the demand on the face:
The person should be secondary to the product, slightly out of focus, looking away from camera, and partially cropped. Avoid clear facial detail.
If hands look wrong, simplify the hand pose:
Use one visible hand only, with a relaxed natural grip around the product. Keep fingers partially wrapped around the side, not spread out. The product label should stay readable.
Repeatable Human Prompt Structure
For meaningful human changes, use a structured prompt instead of a paragraph. This makes the important requirements harder to miss.
Create a realistic image based on the source scene, with the following requirements.
MUST HAVE person:
[Who the person should be. Include age, expression, realism, and any important demographic or styling cues.]
MUST HAVE hair:
[Only include this if hair matters. Be specific about length, cut, texture, and styling.]
MUST HAVE clothing:
[Only include this if styling matters. Describe the exact fashion direction and name the kind of garments.]
MUST PRESERVE scene:
[What should stay from the source scene: pose, hand placement, camera angle, background, lighting, composition.]
MUST HAVE realism:
[Skin texture, facial asymmetry, natural teeth, hands, fabric texture, camera feel, shadow falloff.]
MUST NOT:
[Optional. Add only the failure modes you want to prevent.]
Use MUST HAVE for required changes. Use MUST PRESERVE for what should stay from the source. Use MUST NOT for optional guardrails, such as avoiding long hair, plain clothing, waxy skin, or the wrong demographic.
Example: Human Style Transformation
Create a realistic studio portrait based on the source scene, with the following requirements.
MUST HAVE person:
The person must be a realistic Southeast Asian woman around 50 years old, with a warm confident expression, friendly expressive eyes, natural facial character, subtle smile lines, realistic skin texture, and age-appropriate facial detail. Add respectful, concrete cues such as a medium warm olive-to-golden tan complexion, dark brown eyes, and naturally dark brown or black hair. She should look around 50, not elderly.
MUST HAVE hair:
She must have a stylish cropped bob haircut. The bob must be short, modern, fashion-forward, and end around the jawline or just below the chin. It should have a clean shape, natural movement, and polished styling.
MUST HAVE clothing:
Replace the casual sweatshirt with mature urban high-fashion styling. The outfit must feel refined, expressive, editorial, and age-appropriate for a stylish 50-year-old city woman. Use a structured oversized jacket or sculptural top, modern tailored trousers, layered textures, a strong silhouette, and tasteful statement accessories.
MUST PRESERVE scene:
Keep the same seated pose, hand placement, relaxed product-holding gesture, teal studio background, soft studio lighting, camera angle, framing, and casual portrait composition from the source scene.
MUST HAVE realism:
Make the image feel like a real studio portrait photographed on a full-frame camera with a natural portrait lens. Use realistic lens perspective, natural skin texture, subtle pores, natural facial asymmetry, natural teeth, realistic hands, fabric texture, clothing creases, and soft shadow falloff.
MUST NOT:
Do not use long hair, shoulder-length hair, plain catalogue clothing, a simple t-shirt, conservative officewear, youthful streetwear, airbrushed skin, waxy skin, perfect symmetry, overly white teeth, or an elderly appearance.
Advanced Human And Camera Control
This section is optional. Use it when the image depends on a realistic person, a specific camera feel, a meaningful pose, or a larger transformation.
Camera Cues
Camera language is useful because it tells Riverflow what kind of photograph to make. But it should match the source scene.
For Photoshoots:
Match the source scene's camera angle, lens perspective, framing, lighting direction, and depth of field. Make the person feel like a real photographed subject in the same shoot.
For studio portraits:
Make the image feel like a real studio portrait photographed on a full-frame camera with a natural portrait lens, soft studio lighting, realistic lens perspective, natural skin texture, and believable shadow falloff.
For Images, where there is no scene:
Create a realistic [studio portrait / lifestyle photograph / editorial image] photographed on a [camera type] with a [lens type], [lighting style], [framing], realistic exposure, and true-to-life textures.
Avoid camera cues that fight the scene. Do not ask for an iPhone candid look if the source image is a clean studio portrait, and do not ask for a studio campaign look if the source image is a casual outdoor scene.
Composition And Pose Cues
Pose can make a person look much more realistic. Perfectly centered, front-facing faces often look more AI-generated.
Useful pose instructions:
"gentle head tilt"
"relaxed uneven shoulders"
"slight casual lean"
"mid-expression smile"
"off-center framing"
"looking slightly away from camera"
"hands relaxed, not spread out"
"product sharpest in frame"
Use these subtly. The goal is a natural photographed moment, not a dramatic pose.
Styling Changes
When changing clothing, describe the desired style positively and specifically. Avoid relying only on words like "premium", "age-appropriate", or "fashionable".
Weak:
Make her outfit age-appropriate and fashionable.
Stronger:
Replace the casual sweatshirt with mature urban high-fashion styling: a structured oversized jacket or sculptural top, modern tailored trousers, layered textures, a strong silhouette, and tasteful statement accessories. The outfit should feel refined, expressive, editorial, and polished, not basic or conservative.
Demographic Accuracy
If demographic accuracy matters, a human appearance reference image will often work better than text alone. Use the edit function with a reference image when you need the model's appearance to be specific.
Text can still help. Add concrete but respectful cues: region, approximate age, skin undertone, hair color or texture, eye color, and styling context. Avoid stereotypes, costumes, caricatures, or reducing the person to a single trait.
Weak:
Make her Southeast Asian.
Stronger:
The person should be a realistic Southeast Asian woman around 50 years old, with a contemporary Thai, Vietnamese, Filipino, Malaysian, or Singaporean appearance. Add respectful, concrete cues such as a medium warm olive-to-golden tan complexion, dark brown eyes, naturally dark brown or black hair, friendly expressive eyes, and modern city styling. Keep the result realistic and specific without stereotypes, costume, caricature, or generic stock-photo styling.
Troubleshooting
If the product changes too much:
MUST keep the product dimensions, shape, label, logo, colors, and packaging text exactly as shown. Do not redesign, simplify, relabel, resize, or recolor the product.
If the style reference is copied too literally:
Use only the lighting, image texture, and color treatment from the style reference. Do not copy its props, background, model, outfit, product, or composition.
If the scene did not change enough:
Override the original scene. Keep only the product reference. Create a new setting with the background, props, lighting, and composition described here.
If people look too artificial:
Make the person feel like a real candid photograph: natural skin texture, subtle facial asymmetry, normal posture, imperfect expression, realistic shadows, and no airbrushed or plastic-looking features.
If the face is too symmetrical or the teeth look too perfect:
Add subtle natural facial asymmetry: small differences in the smile, cheeks, eyelids, brows, and expression. The two sides of the face should not look perfectly mirrored. Teeth should look naturally photographed, with slight variation in shape, spacing, brightness, and alignment. Avoid a flawless influencer smile, veneer-like teeth, or perfectly symmetrical face structure.
If the person feels too posed:
Add a gentle head tilt, relaxed uneven shoulders, a slight casual lean, and a mid-expression smile. Keep it subtle and natural, like a real photographed moment.
If the person's demographic is not changing enough:
MUST CHANGE the person's visible facial features, age cues, hair, and styling so they clearly match the requested person description. Preserve the scene's pose, product-holding gesture, camera angle, lighting, background, and composition, but do not preserve the original person's ethnicity, age, hair, face shape, or styling.
If the person still does not read as the intended region or heritage:
Make the person specific to [target region or heritage], not generic. Use respectful concrete cues: [skin tone or undertone], [eye color], [hair color and texture], [age cues], [styling context], and [overall character]. Avoid generic Western features, stereotypes, costume styling, caricature, or stock-photo perfection.
If an extra product detail is missing:
Use product reference 2 as an additional product-detail reference only. It shows [specific detail]. Preserve that detail in the final product. Do not copy the scene, lighting, background, hand, props, or camera angle from product reference 2.
If the product scale is wrong:
MUST use product reference 2 as a sizing reference ONLY. It shows the relative size and proportions of the product compared with [hand / face / other products / surrounding props]. Keep the final product at that believable scale and proportion. Do not copy the scene, model, hand pose, lighting, or background from product reference 2.
If the image is mostly right but one detail is wrong:
Use the edit function. Preserve the original composition, pose, lighting, background, product placement, and overall style. Change only [specific detail]. Use reference image 1 for [specific detail] only, and do not copy its background, camera angle, pose, or unrelated styling.
If the prompt is getting too long:
Keep the product accurate. Keep the scene composition. Change only [specific thing]. Use only [specific part] from the style reference.
Quick Checklist
Before generating, check:
Could I try no prompt first?
Am I working in Photoshoots or Images?
Did I say what should stay the same?
Did I say what should change?
If this is complex, did I use the prompt assembly structure?
If I am working in Images, did I describe the full brief: subject, setting, composition, camera, lighting, style, people, and realism?
Did I specify whether the scene should be preserved or overridden?
If I only want the mood from the source scene, did I say not to preserve the exact layout, people, props, or action?
Did I specify which parts of the style reference to use?
Did I protect product shape, label, dimensions, and text?
If I uploaded extra product references, did I say exactly what each reference is for?
If I uploaded a sizing reference, did I select it in generation and say it is for sizing/proportions only?
Did I describe important materials, textures, or freshness cues?
Did I use a hex color if I need an exact background color?
If people appear, did I make their role simple and realistic?
If the image is mostly right, should I use edit plus a reference image instead of regenerating?