Want better AI images without the trial and error? The AI Image Generation Prompt Guide has 100+ prompts organized by style and category so you can find the exact look you want in seconds. Works with all major image models. 🔗 Get it here →

The biggest shift in AI image editing this year isn't the technology, it's the expectation. People aren't looking for filters anymore. They want their photos to tell a story, become a character, or land on someone's feed and make them stop scrolling. ChatGPT, powered by its Images 2.0 model, has become the tool of choice for that. And right now, some specific styles are taking off in ways that are hard to ignore.

This isn't an exhaustive list of everything AI can do. It's a focused look at what's actually going viral, why each trend works, and how to replicate it yourself, including the prompts that get you there.

Before getting into the trends, it's worth understanding why 2026 feels different from previous years.

Earlier versions of DALL-E treated prompts like keyword queries. You described something, the model pattern-matched against its training data, and you got something vaguely related. Text in images was garbled. Character details changed between generations. Complex compositions fell apart.

ChatGPT Images 2.0 applies a reasoning step before generating. It interprets what you actually want, considering spatial relationships, text placement, and visual logic before producing a single pixel. The practical result: complex prompts actually work now. "A 4x4 grid of candid iPhone photos from a couple's beach trip, grainy, amateur framing, lots of motion blur" doesn't come out as a clean stock photo. It comes out looking like someone's camera roll from 2011.

That reasoning capability is what makes the trends below possible.

There's something perversely satisfying about watching AI, a technology capable of photorealistic rendering, deliberately produce something terrible.

The MS Paint trend asks ChatGPT to redraw your photo in the most clumsy, mouse-drawn style possible: crude lines, flat color fills, slightly off proportions, the aesthetic of someone's older sibling who absolutely did not want to babysit you and drew your face in Paint to prove it.

The results get shared because they're funny. But there's something else going on. After years of hyper-polished AI images, the intentional imperfection reads as more authentic. It's also deeply nostalgic for anyone who spent time on Windows 95.

Prompt to try:

The prompt works better if you give the model explicit permission to fail. Instructions like "the worse the better" or "deliberately bad" seem to unlock a different generation mode than standard prompts.

This one requires understanding why nostalgia works on social media right now.

We're at an interesting moment where the first generation of people who grew up with smartphones are now in their late 20s and early 30s, and the visual aesthetic of early smartphone photography, camera shake, bad exposure, slightly too much saturation has become genuinely emotional for them. It's the visual equivalent of lo-fi music.

ChatGPT Images 2.0's reasoning capability handles this well because it understands the difference between "amateur photo aesthetic" and "technically bad photo." The model can simulate authentic imperfection: wrong framing, motion blur on the person you were trying to photograph, the specific color cast of early iPhone cameras.

Prompt to try:

Upload a photo of yourself if you want the subjects to resemble you. The model will adapt the characters while keeping the aesthetic consistent.

What makes this shareable: It looks like found footage. People tag their partners in it because it doesn't look AI-generated; it looks like a recovered memory.

Caricature has been an art form for centuries. The AI version is just faster and more accessible.

You upload a photo, describe the style you want (political cartoon, children's book illustration, old-school editorial caricature), and ChatGPT exaggerates the subject's most distinctive features while keeping them clearly recognizable. Good caricature isn't random distortion; it's selective emphasis. A strong jawline gets stronger. Big eyes get bigger. That's what makes the subject look more like themselves, not less.

This trend gets used for profile pictures, event programs, team pages, and birthday cards. It's also become a genuinely interesting creative exercise: seeing which features an AI decides are your "defining" ones.

Prompt to try:

One thing to watch: the model sometimes exaggerates to the point of looking mean rather than playful. Adding "warm, friendly tone" to the prompt helps. If you want something sharper, add "satirical, bold exaggeration."

The toy packaging trend combines two things people find irresistible: a personalized version of themselves and extremely precise visual formatting.

The output looks like an unopened action figure in blister packaging, your photo rendered as a plastic figurine, displayed in branded packaging with your name, a title, and a row of small accessories that define you. The variations are endless. "Remote Work Professional" with a laptop and coffee mug. "Marketing Director" with a phone and a strategy deck. "Dog Dad" with a tennis ball and a leash.

What makes this trend technically impressive is that it requires the model to handle text on packaging accurately, maintain the 3D toy aesthetic while representing a specific person, and render the plastic figurine surface convincingly. Images 2.0 does this reasonably well. The text usually comes out readable. The proportions are toy-like without being grotesque.

Prompt to try:

Personalize the title and accessories for the person you're making this for. That's what gets it shared.

This is the trend with the most staying power because it solves a real problem: most photos don't look as good as the memory of taking them.

ChatGPT can apply cinematic color grading to uploaded photos, the specific color science of film stocks, anamorphic lens characteristics, the grain structure of 35mm, and the color bleed of overexposed slide film. The difference between this and an Instagram filter is intentionality. You can specify the film stock (Kodak Portra 400 for warm skin tones and creamy shadows, Fuji Velvia for saturated landscape color, Kodak Vision3 for the specific look of 70s Hollywood), the shooting conditions (golden hour, overcast, harsh midday), and the lab processing style (cross-processed, push-processed, standard).

The results aren't perfect. AI grain is slightly different from real grain if you know what to look for. But for most purposes, social media, personal projects, and presentation visuals, the output is genuinely good.

Prompt to try:

For a more dramatic result, add "golden hour lighting, slight overexposure, and anamorphic lens flare."

Why this outperforms filter apps: You can describe exactly what you want instead of scrolling through presets until something looks close enough.

This one started as a niche creative experiment and became a proper trend.

The Charlie and Lola aesthetic, with thin sketchy outlines, flat pastel colors, hand-drawn textures, and simplified facial features, is technically specific enough that it produces consistent results when prompted correctly. People use it to create personalized illustrations for kids, family portraits in a softer style, and custom storybook pages where their child is the protagonist.

It works because it doesn't try to be photorealistic. The simplicity is the point. A child's drawing quality feels warmer than a polished illustration, and the pastel palette is non-threatening in a way that detailed AI faces often aren't.

Prompt to try:

Knowing the trends is one thing. Getting consistent, usable results is another. These techniques apply across all the styles above.

Every competitor article misses this, and it's probably the most useful thing to know.

When ChatGPT generates or edits an image, you don't have to accept or reject the entire thing. Click into the image and find the paintbrush/edit option. This opens an in-painting tool where you can select a specific area of a shirt, a background, or an object and give targeted instructions. "Change this shirt to navy blue." "Replace this background with a sunset." "Remove this person from the background."

Without this, you regenerate the entire image for a small change, and the model frequently changes things you didn't want changed. In-painting gives you surgical control.

At the beginning of a chat session, specify your hex codes:

The model will apply them consistently for that session. If you use ChatGPT Memory (available on Plus), store this preference so it applies across all future conversations.

If you see an image, whether AI-generated or a real photograph, that has a look you want to replicate, upload it to ChatGPT and ask:

The model will break down the lighting, color treatment, composition, and aesthetic. Use that breakdown as the basis for your next prompt. This is how you move from random outputs to something that feels intentional and consistent.

One clear instruction at a time works better than a list of changes in a single prompt. "Make the lighting warmer," then "increase the contrast slightly," then "add a subtle vignette" gets you to a better result than one long prompt trying to specify all three simultaneously.

The reasoning: each edit builds on the last. The model has context for what you're trying to achieve. Single-instruction iteration also makes it easier to identify which change caused a problem if something goes wrong.

As of 2026, both free users can generate and edit images in ChatGPT. The real difference is in iteration.

Free users get limited daily generations and slower processing. If you're using these editing techniques properly, doing 5-10 iterations to get to a strong result, you'll hit the free limit quickly on anything serious. Plus, users have enough headroom to actually refine.

For one-off personal experiments, free is fine. For consistent creative work, content production, or client deliverables, the Plus plan is the realistic option.

Can ChatGPT edit my existing photos or only generate new ones? Both. You can upload a photo and give editing instructions (color grading, style transfer, object removal, background swap), or generate entirely new images from a text description. The editing capabilities have improved significantly with Images 2.0.

What's the best way to get consistent results across multiple images? Start each session by specifying your style requirements upfront. For a series of images, keep them in the same chat window so the model has context from previous generations. Upload reference images alongside your prompts to anchor the style.

Can ChatGPT remove the background from a photo? Yes. "Remove the background from this image and replace it with a white background" works reliably. For more complex replacements ("change the background to a city street at night"), results vary by image complexity.

Why does ChatGPT sometimes change parts of my image I didn't want changed? This happens when you give editing instructions without using the in-painting tool. The model interprets your instruction as a full regeneration with modifications. Use the paintbrush selection tool to highlight only the area you want changed.

Is the free version good enough for serious photo editing? For occasional experiments, yes. For anything requiring multiple rounds of refinement, which is where quality actually comes from, the free tier limits will become a bottleneck quickly.

How does ChatGPT Images 2.0 compare to dedicated tools like Adobe Firefly or Midjourney? Different strengths. ChatGPT handles multi-turn conversation-based editing better than most alternatives, which you can refine iteratively in natural language without learning a new interface. Firefly has an advantage for commercially safe content (its training data is licensed). Midjourney still produces more striking artistic outputs for pure image generation. The right tool depends on what you're making and who owns the rights to the output.

Can I use AI-edited photos commercially? This depends on the platform and your use case. OpenAI's current terms allow commercial use of ChatGPT-generated images for Plus users. Review the terms before using generated images in paid advertising, product packaging, or client work.