How to spot AI art slop (and how to avoid making it)
A checklist for spotting AI art slop by the tells that still work, how to avoid making it, and how the AI character apps we review score on art.
Edited by Troy · Published
To spot AI art slop, check where image generators still fail: background text, hands gripping objects, repeating patterns like brick or tile, and backgrounds that melt at the frame's edges. Then ask whether a person edited the image. Slop is raw output published as final. The fix for your own work is to treat every AI output as a first draft and rework it before it goes out.
What counts as AI art slop
AI art slop is low-effort AI imagery published without meaningful human editing. Writers covering the term report that it belongs to a family of labels for mass-produced junk, alongside spam, fluff, clickbait, churnalism and kitsch. "Slop" is the one reserved for the AI-generated kind. One artist describes it as content that is low-effort, high-engagement and endlessly replicable. The Cambridge Dictionary added the word in 2025, defining it as low-quality internet content, especially that created by AI.
The word has a long history as an insult:
- In the 1700s it meant soft mud, and in the 1800s food waste.
- Later it came to mean rubbish.
- Online it has described low-effort experiences since at least 2019.
- References to AI slop date to at least 2022.
- Two years later, a poet and technologist called "deepfates" popularized it as "the term for unwanted AI generated content."
The tool does not make something slop. Intent does. Not all AI-generated content is slop, but content generated mindlessly and thrust on someone who didn't ask for it is.
How to spot AI art slop
Check the places generators still fail. Reviewers who test image generators report that by 2026 the famous tells, such as six-fingered hands, garbled text and waxy skin, have mostly been fixed, so the useful checks have moved.
- Read the background text: distant storefronts, book spines and small print. Foreground lettering has improved, but background text is still where models usually fail. The letter forms look plausible and don't combine into real words.
- Find a hand that is gripping an object at an odd angle, overlapping something or partly hidden. Those poses still produce extra or fused fingers more often than not. Older Midjourney versions (v5 and earlier) got hands wrong far more often.
- Follow a repeating pattern, such as brick, tile or fabric, from one edge of the image to the other. AI patterns drift. A brick course that runs straight on the left may curve by the right edge.
- Scan the edges of the frame. Background objects can dissolve into abstract texture instead of staying geometrically coherent.
- Zoom in on skin. Porcelain smoothness with no fine hairs, pores or blemishes is an older tell, but it still turns up.
- Judge the face as a face. AI generators build faces from the mathematical average of tens of thousands of training faces. Researchers found people rate AI faces as more symmetrical, well-proportioned and attractive than real ones, but less distinctive, memorable and expressive.
- Compare the image with others from the same account or brand. The same visual elements showing up across unrelated designs is a clear giveaway.
- On LinkedIn, report it with the "Seems like AI slop" button.
What you get at the end is a probability, not proof. Reviewers cite a 2024 meta-analysis of 56 peer-reviewed studies that put average human accuracy at spotting AI images at 55.54%, barely above chance. Visual errors like warped backgrounds and missing fingers have become far less reliable as generators improved.
Training helps. A PNAS study from Australian National University taught participants to recognize AI-generated faces, in some cases with "near-perfect accuracy." Images from Google's Gemini and Imagen models also carry SynthID, an invisible watermark embedded in every image they generate.
How to avoid making AI art slop
Use AI as a design partner and treat every output as a first draft. Design writers define slop as generic AI visuals that look polished at first but are interchangeable, because they went out without meaningful human editing.
- Take the generator's output as a starting draft, not the finished image.
- Run the checklist above on it, starting with background text, hands and patterns.
- Edit the image by hand before it goes anywhere public.
- Lay it next to your other recent designs, and change any element that repeats across unrelated work.
Expect this to cost you time, and that is the point. Making slop now costs close to nothing, and that near-zero cost is why it floods feeds.
AI art slop examples
The best-known cases are viral:
- Shrimp Jesus. People who track slop describe "Shrimp Jesus" images as mass-produced for engagement farming, with no artistic intent or message. Stanford Internet Observatory studied 120 Facebook Pages whose AI images drew hundreds of millions of interactions.
- A children's book. An AI-generated Danish children's book showed hyenas with hooves, a gorilla with six fingers and impossible animal poses.
- Misleading gift ads. By January 2026, shoppers were donating, returning or reselling "AI slop gifts" bought from misleading AI-image ads.
- Italian brain rot. This AI-generated trend went viral on TikTok in 2025.
How the apps we review score on AI art slop
These scores are true at this review, and lower is better.
- Frontier, 18/100. Our review looked at the covers on Frontier's public story pages. They have clean hands, correctly spelled title lettering and a style picked for each story, from carved stone for Sky Born to a storybook sketch for Once Upon a Swamp that acts out the premise. The one generic touch is the stock glowing-magic anime look on the Hot Mage cover.
- Nomi, 48/100. Our review looked at the example selfies on Nomi's homepage. The photo-style ones pass for real photos, but the storefront sign behind one spells nothing, the anime selfie's hologram and neon signs are scribbles, and the glamour shot has the default glossy finish.
- Character.AI, 55/100. Our review looked at the avatars of five featured characters, which users upload. Monster Maker's is raw early generator output with melted limbs and floating fragments, and Clara Brooks has the glossy default AI portrait look, next to a painterly generated Loki, a manga-style panel and a photo of a crochet doll. If you make a character with an AI character generator and upload the first result, Monster Maker is the look to avoid.
- Janitor AI, 68/100. Janitor's public character pages turned us away, so our review judged its App Store screenshots. One waxy, heavy-lidded male face repeats across a dozen user-uploaded portraits under an orange cast, and the featured knight's shoulder armor turns into shapeless plates.
To judge an app by its art, look for one character who keeps the same face across screens, and for armor, robes and backgrounds that hold their shape. The same default finish repeated across unrelated characters is the tell from step 7.
Apps this guide covers
- Janitor AI review51Microwaved
Millions of community bots to chat with free; Janitor+ adds memory for $12.99 a month
- Character.AI review51Microwaved
Millions of community-made characters, free unlimited chat, and ads until you pay
- Nomi review39Microwaved
An 18+ AI companion you build yourself, with layered long-term memory
- Frontier review16Home-cooked
Human-written visual novels whose characters keep talking after the ending
Sources
The source text behind the claims above, and the day we read it.
- Receipt 1Reviewer
Reviewers report that "slop" is one of several terms (with spam, fluff, clickbait, churnalism, kitsch) for mass-produced, low-quality content, with "slop" reserved for the AI-generated variety.
"Spam, fluff, clickbait, churnalism, kitsch-slop: These are all ways to describe mass-produced, low-quality content. The latter term is reserved for the newest variety, which comes from artificial intelligence."
www.scientificamerican.com, retrieved
- Receipt 2Reviewer
Scientific American reports developer Simon Willison wrote that not all AI-generated content is slop, but mindlessly generated content thrust on someone who didn't ask for it is.
""Not all AI-generated content is slop," he wrote. "But if it's mindlessly generated and thrust upon someone who didn't ask for it, slop is the perfect term.""
www.scientificamerican.com, retrieved
- Receipt 3Reviewer
Scientific American reports AI slop references date to at least 2022, and a poet/technologist called "deepfates" popularized the term two years later as "the term for unwanted AI generated content."
"Though references to AI slop date back at least to 2022, a poet and technologist who writes under the name "deepfates" popularized it two years later as "the term for unwanted AI generated content""
www.scientificamerican.com, retrieved
- Receipt 4Reviewer
Scientific American argues many human-guided AI creations are original, surprising and affecting, and some have been displayed in museums.
"Many human-guided AI creations are original, surprising and affecting. Some have been displayed in museums."
www.scientificamerican.com, retrieved
- Receipt 5Reviewer
Scientific American says the cost of making slop has collapsed to near zero while the cost to others is high in cognitive burden, attention and environmental cost.
"the cost for those making slop has collapsed to near zero, while the cost for others is high in terms of cognitive burden-of doubting what we see and experiencing the demands on our attention"
www.scientificamerican.com, retrieved
- Receipt 6Reviewer
Ocula says the Cambridge Dictionary added "slop" in 2025, defining it as low-quality internet content, especially that created by AI.
"The Cambridge Dictionary added the term this year, and defines it as low-quality content on the internet, especially that which is created by artificial intelligence."
ocula.com, retrieved
- Receipt 7Reviewer
Ocula quotes artist Brad Troemel describing slop as content that's low-effort, high-engagement, and endlessly replicable.
"then slop is its cultural byproduct-content that's low-effort, high-engagement, and endlessly replicable."
ocula.com, retrieved
- Receipt 8Reviewer
Reviewers report that garbled text in AI images remains a reliable tell: signs, books, badges and storefront lettering often have plausible letter forms that don't combine into real words.
"signs, books, name badges, and storefront lettering often contain letters that look almost-but-not-quite right, plausible letter forms that do not combine into real words"
www.eyesift.com, retrieved
- Receipt 9Reviewer
Reviewers report older Midjourney versions (v5 and earlier) frequently produced six-fingered hands, fused knuckles and impossible finger proportions.
"Midjourney v5 and earlier frequently produced six-fingered hands, fused knuckles, and anatomically impossible finger proportions."
www.eyesift.com, retrieved
- Receipt 10Reviewer
Reviewers report AI generators render skin with a porcelain-like over-smoothness lacking fine hairs, pores and blemishes.
"AI generators render skin with a characteristic over-smoothness at certain zoom levels, a porcelain-like quality that lacks the fine hairs, pores, capillary detail, and minor blemishes of real skin"
www.eyesift.com, retrieved
- Receipt 11Reviewer
Reviewers cite a 2024 meta-analysis of 56 peer-reviewed studies finding average human AI image detection accuracy of 55.54%, barely above chance.
"A 2024 meta-analysis of 56 peer-reviewed studies (ScienceDirect) found average human AI image detection accuracy at 55.54%, barely above random chance"
www.eyesift.com, retrieved
- Receipt 12Reviewer
Reviewers report background objects in AI images may dissolve into abstract texture at the edges rather than staying geometrically coherent.
"Background objects may dissolve into abstract texture at the edges rather than remaining geometrically coherent."
www.eyesift.com, retrieved
- Receipt 13Reviewer
Reviewers report that by 2026 the obvious tells (six-fingered hands, garbled text, waxy skin) have mostly been fixed, so useful checks have shifted.
"By 2026, image generators produce results clean enough that the obvious tells from two years ago (six-fingered hands, garbled text, waxy skin) have mostly been fixed."
www.opticflux.com, retrieved
- Receipt 14Reviewer
Reviewers report hands gripping objects at unusual angles, overlapping or partly obscured still produce extra or fused fingers more often than not.
"Hands gripping an object at an unusual angle, hands overlapping other hands, or hands partially obscured by hair or clothing still produce extra or fused fingers more often than not."
www.opticflux.com, retrieved
- Receipt 15Reviewer
Reviewers report background text (distant storefronts, book spines, small print) is still where models usually fail, even as foreground text improved.
"text sitting in the background, on a distant storefront, a license plate, a book spine on a shelf, small print on packaging, is usually still where models fail."
www.opticflux.com, retrieved
- Receipt 16Reviewer
Reviewers report AI-generated repeating patterns (brick, tile, fabric) frequently drift, e.g. a brick course that's straight on the left curves by the right edge.
"AI-generated versions of these patterns frequently drift, a brick course that's straight on the left of the image curves by the right edge"
www.opticflux.com, retrieved
- Receipt 17Reviewer
Reviewers report Google's Gemini and Imagen models embed SynthID, an invisible watermark, into every image they generate.
"Google's Gemini and Imagen models embed SynthID, an invisible statistical watermark, into every image they generate."
www.opticflux.com, retrieved
- Receipt 18Reviewer
Venngage defines AI slop in design as low-effort, generic visuals made with AI and published without meaningful human editing, polished at first but interchangeable.
"AI slop in design means low-effort, generic visuals created with AI and published without meaningful human editing. It looks polished at first but feels interchangeable."
venngage.com, retrieved
- Receipt 19Reviewer
Venngage advises fixing AI slop by using AI as a design partner and treating every AI output as a first draft.
"To fix AI slop, use AI as a design partner. Treat every AI output as a first draft."
venngage.com, retrieved
- Receipt 20Reviewer
Venngage reports LinkedIn added a 'Seems like AI slop' button to report low-effort AI-generated content.
"LinkedIn recently added a 'Seems like AI slop' button to report low-effort AI-generated content."
venngage.com, retrieved
- Receipt 21Reviewer
Venngage says a clear giveaway of AI slop is the same visual elements showing up across unrelated designs.
"A clear giveaway of AI slop is when the same visual elements keep showing up across unrelated designs."
venngage.com, retrieved
- Receipt 22Reviewer
Venngage reports that by January 2026 shoppers were donating, returning or reselling "AI slop gifts" bought from misleading AI-image ads.
"By January 2026, disgruntled shoppers started donating, returning or reselling these "AI slop gifts" after feeling duped by the misleading ads."
venngage.com, retrieved
- Receipt 23Reviewer
Merriam-Webster chose "slop" as its 2025 Word of the Year, defining it as digital content of low quality produced usually in quantity by means of AI.
"Merriam-Webster's human editors have chosen slop as the 2025 Word of the Year. We define slop as "digital content of low quality that is produced usually in quantity by means of artificial intelligence.""
www.merriam-webster.com, retrieved
- Receipt 24Reviewer
Merriam-Webster notes slop originally meant "soft mud" in the 1700s, then "food waste" in the 1800s, then "rubbish" or a product of little value.
"The original sense of the word, in the 1700s, was "soft mud." In the 1800s it came to mean "food waste" (as in "pig slop"), and then more generally, "rubbish" or "a product of little or no value.""
www.merriam-webster.com, retrieved
- Receipt 25Reviewer
Merriam-Webster lists the 2025 slop flood as including absurd videos, off-kilter advertising images, fake news, junky AI-written books and lots of talking cats.
"The flood of slop in 2025 included absurd videos, off-kilter advertising images, cheesy propaganda, fake news that looks pretty real, junky AI-written books"
www.merriam-webster.com, retrieved
- Receipt 26Reviewer
Gizmodo reports a PNAS study from Australian National University trained participants to recognize AI-generated faces, in some cases with "near-perfect accuracy."
"researchers from Australian National University claim they used a special method to successfully train a group of participants to recognize AI-generated faces, in some cases with "near-perfect accuracy.""
gizmodo.com, retrieved
- Receipt 27Reviewer
The PNAS researchers say people rate AI faces as more symmetrical, well-proportioned and attractive than human faces, but less distinctive, memorable and expressive.
""People automatically detect these differences, rating AI faces as more symmetrical, well-proportioned, and attractive than human faces, but less distinctive, memorable, and expressive,""
gizmodo.com, retrieved
- Receipt 28Reviewer
The study says AI image generators build faces from the mathematical average of tens of thousands of training faces, making them look more typical than real faces.
"this reliance on the mathematical average makes the face seem "more typical in appearance than real human faces.""
gizmodo.com, retrieved
- Receipt 29Reviewer
Gizmodo reports older detection relied on visual errors like warped backgrounds, anatomical glitches and missing fingers, but those cues became far less reliable as generators improved.
"warped backgrounds, anatomical glitches, or telltale mistakes like missing fingers. But as AI image generators have become more precise, those cues have become far less reliable."
gizmodo.com, retrieved
- Receipt 30Reviewer
SlopDetector describes "Shrimp Jesus" images and notes Stanford Internet Observatory studied 120 Facebook Pages whose AI images drew hundreds of millions of interactions.
"Stanford Internet Observatory studied 120 Facebook Pages posting AI images that generated hundreds of millions of interactions."
slopdetector.org, retrieved
- Receipt 31Reviewer
SlopDetector reports an AI-generated Danish children's book had hyenas with hooves, a gorilla with six fingers, and impossible animal poses.
"hyenas with hooves instead of paws, a gorilla with six fingers, and anatomically impossible animal poses."
slopdetector.org, retrieved
- Receipt 32Reviewer
SlopDetector reports Svedka ran a fully AI-generated Super Bowl ad in 2026 and post-game surveys found only 7% of viewers could identify the brand.
"Vodka brand Svedka ran a fully AI-generated Super Bowl ad during the 2026 game. Post-game surveys found only 7% of viewers could correctly identify the brand"
slopdetector.org, retrieved
- Receipt 33Reviewer
SlopDetector says Shrimp Jesus-style imagery is mass-produced for engagement farming with no artistic intent or message.
"Mass-produced sensational imagery designed purely for engagement farming. No artistic intent, no message - just bait for clicks and ad revenue."
slopdetector.org, retrieved
- Receipt 34Reviewer
Britannica defines AI slop as low-quality AI-generated media, often made with little human effort, in the form of images, videos, audio or text.
"AI slop is low-quality media content generated by artificial intelligence (AI), often with little human effort. It can come in the form of images, videos, audio, or text."
www.britannica.com, retrieved
- Receipt 35Reviewer
Britannica says critics worry AI slop contributes to misinformation and overwhelms audiences with low-value content.
"Critics worry that AI slop contributes to misinformation, such as incorrect or ill-advised AI -generated search answers, and overwhelms audiences with low-value content."
www.britannica.com, retrieved
- Receipt 36Reviewer
Britannica notes "slop" has been used online for low-effort experiences since at least 2019, and that "Italian brain rot" was an AI-generated brain rot trend that went viral on TikTok in 2025.
"On the Internet, "slop" has been used to describe low-effort online or in-person experiences since at least 2019."
www.britannica.com, retrieved
Questions people ask
What is considered AI slop?
AI slop is low-quality AI-generated media made with little human effort. It comes as images, videos, audio or text. Merriam-Webster's 2025 Word of the Year entry defines it as digital content of low quality produced usually in quantity by means of AI. Its examples of the 2025 flood include absurd videos, off-kilter advertising images, fake news, junky AI-written books and lots of talking cats.
Why is AI art slop?
AI art is not slop by default. It becomes slop when it is generated mindlessly and pushed on people who didn't ask for it. Writers on the topic point out that human-guided AI work has been original enough to be displayed in museums. Slop spreads because making it now costs close to nothing, while viewers pay for it in attention and cognitive burden, and the environment pays too.
How to identify AI slop images?
Start with the background text, such as distant storefronts and book spines. Then check hands that grip or overlap something, and follow any repeating pattern from edge to edge. Treat the result as a probability. A 2024 meta-analysis of 56 studies put average human accuracy at spotting AI images at 55.54%, barely above chance.
How to avoid AI design slop?
Treat every AI output as a first draft, and edit it by hand before you publish. Design writers define slop as generic AI visuals published without meaningful human editing. It looks polished at first glance but is interchangeable with everyone else's, so reusing the same visual elements across unrelated designs gives it away.
Is AI slop bad?
Critics say AI slop contributes to misinformation and overwhelms audiences with low-value content. One commercial example: Svedka ran a fully AI-generated Super Bowl ad in 2026, and post-game surveys found only 7% of viewers could identify the brand.
Slop check: passedThis page went through our slop counter before publishing. How we check our own copy