The AI Smells Guide
Image Smells (And How to Fix Them)
12 visual patterns that feel generic, and the prompts that make images real.
Image Smells (And How to Fix Them)
12 visual patterns that feel generic, and the prompts that make images feel real.
Your brain categorizes an image as "stock" or "real" in under a second. That snap decision determines whether someone reads what comes next or keeps scrolling. AI image generators inherited every stock photography cliche from their training data, and they reproduce them by default.
An image that could be any workplace says "we didn't think about this." An image with specific light, real mess, and human expressions says "this is us." That is the same principle behind the writing smells in Module 1 and the formatting smells in Module 2: the defaults are generic, and generic reads as careless.
Every one of these smells has a specific, promptable fix. Here are all 12.
Expression and posture smells
1The Universal Grin
Everyone in the frame smiling at the same intensity, at the same time, directly at camera. Real workplace moments have asymmetric expressions. One person talks, another listens, a third glances at their screen. Synchronized beaming is the single fastest way to make an image feel staged.
Fix: Prompt for mixed expressions and candid moments. "Mid-conversation," "one person listening while another explains," "natural resting expressions." Specify that not everyone should face the camera.
2The Exaggerated Reaction
Eyes bugging, mouth agape in shock, fist-pump celebrations, head-thrown-back laughter. Real workplace emotions are subtle. A genuine smile doesn't show every tooth. A real moment of satisfaction is a quiet exhale and a lean back in the chair.
Fix: Prompt for understated emotion. "Quiet satisfaction," "subtle smile," "focused concentration," "mid-thought expression." The less dramatic the expression, the more believable the image.
3Everyone Facing Camera in a "Candid" Scene
The fourth wall break that destroys every supposedly spontaneous image. If the scene is a meeting, people should be looking at each other, at a screen, at their notes. Camera-awareness in a candid composition is the visual equivalent of an actor looking at the lens during a movie.
Fix: Direct gaze away from camera. "Looking at each other," "focused on a shared screen," "glancing down at notes." If one person happens to look toward camera, make it incidental.
4The Sustained Handshake
A handshake is a two-second transitional gesture. In stock photography, it's a lingering, centered, perfectly lit event. Same for high-fives, fist bumps, and group celebrations. When a momentary gesture becomes the entire composition, it looks frozen and fake.
Fix: Capture the moment around the gesture, not the gesture itself. "Just after shaking hands, both turning to sit down," "walking away from the whiteboard after presenting."
Environment and lighting smells
5Flat, Shadowless Lighting
Even illumination across every surface. No direction, no falloff, no mood. This is the lighting equivalent of a blank white wall. Stock photographers optimized for compositing flexibility, not realism. AI models default to the same flatness.
Fix: Specify a light source and direction. "Morning window light from the left," "overhead office fluorescents with visible shadows," "backlit by a laptop screen in a dim room." Shadows are what make a photo look like it happened somewhere.
6The Pristine Environment
Desks with nothing on them. Conference rooms that look like furniture showrooms. Not a single coffee cup, cable, sticky note, or personal item in sight. Real workplaces are messy. The absence of clutter is the absence of life.
Fix: Prompt for environmental specificity. "Half-empty coffee mug on the desk," "sticky notes on the monitor," "jacket draped over a chair," "visible cables." A lived-in space tells the viewer real people actually work here.
7The White Void Background
Subject isolated on pure white. Zero environmental context. This is the calling card of stock photography because it makes the image droppable onto any layout. It also makes the image feel like it belongs nowhere.
Fix: Always place subjects in a specific environment. "Open-plan office with exposed brick," "small meeting room with a window," "hallway near the kitchen." Even a blurred background with identifiable elements beats infinite white.
8Oversaturated, Over-Processed Color
Colors pushed 20-40% past natural. Greens glow neon. Skin tones drift toward spray-tan orange. The teal-and-orange color grade appears on every third business image because AI models absorbed a decade of that preset.
Fix: Specify restrained, naturalistic color. "Muted tones," "natural skin tones," "film-like color palette." If you want warmth, ask for warm light, not warm grading.
Casting and concept smells
9The Calculated Rainbow Coalition
A group shot that looks cast by a demographic spreadsheet. One person of each ethnicity, everyone the same age and attractiveness level, all in matching business casual. Real teams cluster by shared context, not by quota. Performative diversity reads as performative.
Fix: Prompt for natural group dynamics rather than demographic checklists. Vary ages, body types, and styles. Let some people be partially obscured or at the edges of frame.
10Mannequin Skin
Faces with zero pores, zero texture, zero imperfection. The human brain is extraordinarily tuned to facial texture. When it's missing, something registers as wrong before you consciously identify the problem.
Fix: Prompt for natural skin texture. "Visible pores," "natural skin imperfections," "realistic facial detail." Explicitly reject smoothing: "no airbrushed skin, no beauty filter effect."
11Props as Costumes
A stethoscope draped around a neck but never in use. A hard hat on someone standing in front of a white background. A headset worn by someone beaming at camera instead of looking at a screen. When props are worn but not used, they become costumes.
Fix: Show people using tools, not wearing them. "Doctor reviewing a chart on a tablet," "construction worker adjusting scaffolding," "customer support rep typing while talking."
12Conceptual Literalism
Lightbulb for ideas. Puzzle pieces for teamwork. Dart hitting a bullseye for goals. Mountain summit for success. These visual metaphors were cliches in PowerPoint in 2010. AI models reproduce them eagerly because the training data is saturated with them.
Fix: Replace symbols with specific human moments. Instead of a lightbulb for "innovation," try "engineer sketching on a napkin at a cafeteria table." Specificity beats symbolism every time.
The anti-stock prompting cheatsheet
| Instead of... | Prompt for... |
|---|---|
| "A team in an office" | "Three people mid-discussion in a cluttered open-plan office, morning light from left windows" |
| "A happy employee" | "A person leaning back in their desk chair with a quiet half-smile, coffee in hand" |
| "A leader presenting" | "Someone mid-sentence at a whiteboard with real diagrams, one colleague interrupting with a question" |
| "Diversity" | Varied ages, body types, and personal styles without forcing demographic symmetry |
| "Professional setting" | A specific, named environment with real-world mess and detail |
| "Good lighting" | A named light source with direction and visible shadows |
Try it: Free Anti-Stock Image Prompt
Write your image prompt however you normally would. Then copy the block below and paste it at the end before you generate. It tells the image model what to avoid and what to prioritize.
Free prompt included with this guide
Enter your email to unlock the ready-to-use prompt. Copy, paste into your LLM, and use it forever.
Write your prompt, paste this at the end, generate. The constraints do the work so you don't have to memorize all 12 smells every time you need an image.
Stock photography fails because it optimizes for maximum reusability. Everything generic enough to fit anywhere fits nowhere specifically. AI image generators inherited this bias from training data drenched in stock imagery. The fix is the same for images as it is for writing: specificity. A specific person, in a specific place, at a specific moment, doing a specific thing.
That covers all three dimensions of authentic AI-assisted communication: how it reads, how it looks on screen, and how it looks as imagery. Explore the full series at The AI Smells Guide, or go back to fix how your message reads (AI Writing Smells) and how it's formatted (Formatting Smells). Then fix who you write for: How to Talk to Anyone: 27 Subtype Communication Guide.
This is what Rally does automatically: communications that look and read like they came from someone who cares, optimized for employee engagement.
Sources
- Martin Fowler, "Code Smell," martinfowler.com, 2006. Origin of the term coined by Kent Beck.
- Rachael Kay Albers, forensic audit of stock photo search results across DepositPhotos, GraphicStock, iStock, and Shutterstock. Documented near-total absence of body diversity, age diversity, and LGBTQ+ representation.
- Brevity & Wit (Acacia Betancourt), analysis of tokenism, disability "inspiration porn," and cultural stereotyping in stock photography.
- Emily Shornick (Photo Art Director, Netflix), on the industry shift from glossy flash to natural light and film for authenticity.
- Fstoppers, analysis of HDR overprocessing, oversaturation, and the "deep-fried" look in stock post-processing.
- Topcontent, "Why Stock Photos Look Fake," on flat lighting and idealized environments.
- PictureCorrect, analysis of heavy vignetting and over-sharpening as stock photography tells.
- Campbell Addy / Design Indaba, on diversity beyond race: "shape, size, able-bodied, disabled, age, looks."
- Joshua Kissi / Nappy.co (Forbes profile), on Black stock photography being "outdated, lacked style, and screamed cliche."
- DepositPhotos / Envato Elements editorial teams, self-aware catalogs of their own visual cliches.
- Dreamstime editorial team, on natural color grading and context-aware camera angles for authentic stock.