Gemini AI: Ethics of the Photo Trend

Lisa Ernst · 12.11.2025 · Technology · 7 min

The Google Gemini AI platform, accessible via Nano Banana, enables the creation and editing of images. This ranges from light retouching to completely new scenes. While this opens up creative possibilities, it also carries risks, especially with regard to deepfakes. Deepfakes are AI-generated media that look convincingly real, but are artificial and often misused to depict real people in intimate or demeaning portrayals. K-Pop idols and other celebrities are frequently victims of such deepfakes, leading to legal actions by agencies. This text examines the boundary between harmless fan art and problematic deepfakes and provides concrete formulations for responsible use of Gemini.

Source: Own depiction

Introduction & Basics

Gemini is Google's AI platform for generating and editing images, accessible in everyday life via Nano Banana. Users can upload photos and instruct the AI with text commands to add to, modify, or create entirely new scenes. The range extends from light retouching to new scenes that look like real memories ( gemini.google).

Fan art here refers to images that are visibly stylized, distorted, or clearly recognizable as artistic interpretation. An example is an AI portrait in pastel-toned style, inspired by a K-Pop stage performance, without exactly copying a real person. It becomes problematic when real people are placed into scenes that never occurred, especially in intimate, nude, or demeaning contexts. This is called a deepfake: AI-generated media that looks like real photos or videos but is completely artificial ( ScienceDirect).

The AI can theoretically generate all three types of images. Whether the usage remains respectful or verges on abuse depends on the prompts and how the results are handled.

Example of harmless, own image editing:

"Bearbeite dieses Selfie so, dass die Farben weicher und der Hintergrund leicht unscharf sind, ohne mein Gesicht stark zu verändern."

Example of stylized fan art without a real person:

"Erzeuge ein Porträt einer fiktiven Pop-Sängerin in K-Pop-inspirierter Mode, im Neonlicht einer Bühne. Sie soll keiner bekannten Person deutlich ähnlich sehen."

Example of a problematic prompt toward a deepfake:

"Erzeuge ein hyperrealistisches Foto, in dem eine echte Sängerin und ich alleine in einem Schlafzimmer liegen."

Current State & Risks

Nano Banana is described by Google as an AI image generator that can transform photos into countless new creations, such as 3D figures or new scenes ( gemini.google). ). Everyday use has increased, with users transforming phone photos into analog-looking Polaroids, glamorous portraits, or pop culture scenes ( secureitworld.com).

At the same time, reports are increasing about K-Pop idols and other stars whose faces are manipulated with AI into romantic or sexualized scenes. Apps encourage fans to mix their own images with photos of K-Pop stars to create fake kiss videos ( dojeonmedia.com). ). Deepfakes increasingly affect minors as well, which in fandom sparks outrage, since intimate scenes are created and distributed that are hurtful to those involved ( PopRant).

K-Pop agencies are cracking down. JYP Entertainment announced the strongest legal steps against deepfake videos of their female artists. Studies show that especially women and K-Pop idols are the focus of sexualized deepfakes ( straitstimes.com, kci.go.kr).

Critical voices about Gemini itself warn of data privacy risks, such as long-term storage of photos and possible use for AI training, even though Google uses watermarks ( timesofindia.indiatimes.com). ). NDTV cites data breaches, sharing of personal information, and phishing attempts as risks ( ndtvprofit.com). ). Security blogs and police authorities warn about fake sites that collect data or distribute malware with free Nano Banana filters ( secureitworld.com, indiatimes.com).

Source: dnpindia.in

The viral Gemini AI photo trend, often depicted by women in traditional saris, has far-reaching ethical implications.

Motives & Interests

Photo-realistic AI images are attractive because they suggest closeness. Fans see themselves in situations with idols that are hardly possible in real life. Studies on deepfake pornography show that such content elicits strong emotions and caters to personal fantasies ( ScienceDirect, link.springer.com).

In addition, there is the creative lure: many use Gemini to experiment with aesthetics, try outfits, or reinvent themselves visually. Reports describe saree portraits, 3D figures, and nostalgic Polaroids from selfies ( secureitworld.com, timesofindia.indiatimes.com).

Platforms and providers also have interests. Google describes Nano Banana as a leading tool for image generation, uses watermarks like SynthID, and promotes creative use cases ( gemini.google, ai.google.dev). ). SynthID marks AI-generated content invisibly with markers that can be detected later ( deepmind.google, blog.google). ). This is an important step, but not a perfect protection, especially if images are further edited or uploaded ( theverge.com).

On the fan side, a tension arises: creativity vs. boundary-crossing. When P-P idols appear in AI fantasy images, some fans view it as harmless admiration, others as intrusive, because idols are real people ( PopRant).

Responsible use

The crucial question is how to use Gemini creatively without infringing on others. The selection of uploaded photos, the wording of prompts, and handling of the results are important.

Safe Foundation: Which photos you upload

Articles on Nano Banana advise against uploading intimate, compromising, or identifying photos, as data breaches and misuse can never be completely ruled out ( ndtvprofit.com, secureitworld.com).

Concrete, copy-paste-ready approaches:

"Nutze dieses Foto so, dass nur mein Gesicht und Oberkörper zu sehen sind. Schneide alle anderen Personen aus dem Bild."

"Erzeuge aus diesem Foto eine Version mit neutralem Hintergrund, ohne Details der Wohnung oder des Zimmers zu zeigen."

Prompt Builder: Respectful phrases for your own images

When working with your own images you can freely design, as long as you don't create fake situations intended to deceive others.

Examples of ready-to-use prompts:

"Bearbeite dieses Selfie so, dass meine Haut glatter wirkt, aber meine natürlichen Gesichtszüge erkennbar bleiben. Vermeide extreme Veränderungen."

"Erzeuge aus diesem Selfie ein analog wirkendes Polaroid-Foto im Stil der 90er Jahre, mit leicht verblassten Farben."

"Verwandle dieses Foto von mir in ein Bühnenfoto mit Lichtshow, ohne meine Körperform oder Kleidung stark zu verändern."

Fan art instead of deepfake: When idols and stars are involved

Once real, well-known people are involved, caution is warranted. Media reports and agencies react sensitively to intimate or sexualized scenes ( dojeonmedia.com, PopRant).

Responsible Prompts:

"Erzeuge ein stilisiertes Poster im K-Pop-Stil mit einer erfundenen Boygroup. Die Gesichter sollen generisch wirken und nicht wie echte Idols aussehen."

"Erstelle eine Illustration, in der eine Person, die mir ähnlich sieht, auf einem Konzert im Publikum steht. Die Band auf der Bühne ist frei erfunden."

"Generiere ein Fanart-Bild im Manga-Stil, inspiriert von Bühnenoutfits meines Lieblingsidols, aber ohne das Idol selbst zu zeigen."

Prompts describing intimate or sexual actions involving real people should be avoided. This applies to instructions that place real singers, idols or actresses in bedrooms, baths, or other private scenes, especially with phrasing about nudity or bodily closeness.

Closeness can be conveyed through symbols:

"Erzeuge ein Bild, in dem ein fiktiver Popstar eine Dankesbotschaft an die Fans auf einem Bildschirm zeigt, während ich im Publikum stehe."

"Erstelle ein Bild, in dem eine fiktive Figur ein Fan-Paket mit Album, Lightstick und Brief vorbereitet. Die Person ist eine generische Künstlerin, kein echtes Idol."

Interactions with friends, partners, and family

For images of people in private settings, their consent is crucial. AI pornography shows how hurtful it can be when others create and share AI versions without consent ( link.springer.com).

A copy-pasteable sentence:

"Bevor ich unser Foto in einem KI-Tool bearbeite und teile, möchte ich dein Einverständnis. Bist du damit einverstanden, dass aus diesem Bild eine stilisierte Version gemacht wird?"

A no is a clear stop signal.

Labeling and publication

Even with correct technical use, there is a responsibility when sharing. Experts recommend labeling AI images so that their synthetic origin is clear ( ai.google.dev).

Possible formulations for captions:

"Dieses Bild wurde mit KI bearbeitet (Google Gemini). Es zeigt eine Fantasie-Szene, kein echtes Treffen."

"Fanart, mit KI erstellt. Die dargestellten Figuren sind fiktiv und keine realen Personen."

The video shows how deepfakes and manipulated media are used for fraud and disinformation, and helps to understand why clear labeling and caution when sharing are important.

Source: user-added

A shared image shows on the left a blue, transparent head with music notes, and on the right a baby wearing headphones.

Open Questions & Conclusion

Despite watermarks, guidelines, and legal debates, questions remain. It is unclear how consistently platforms delete deepfake content and support those affected. Victims of AI pornography often face high hurdles to have content removed ( link.springer.com).

It is also unclear how far future laws will go: Will non-consensual AI fan art with stars be treated as strictly as explicit deepfakes? What role do watermarks like SynthID play in identifying those responsible ( deepmind.google, theverge.com)?

For fandoms, the question remains what rules communities will set: Will there be clear codes such as 'no sexualized AI images of idols,' will respectful fan art be praised and intrusive content shunned ( PopRant)? )? Here, fans, artists, agencies, and platforms are all called upon to work together.

Gemini and Nano Banana enable creative image editing. The same technology can also produce deepfakes that show real people in non-consensual situations. The difference lies in the choices: which photos are uploaded, which prompts are chosen and how the results are shared. If creativity focuses on your own images and fictional figures, consent is obtained, intimate scenes with real people are avoided, and AI images are clearly labeled, Gemini can be used without infringing on others ( ndtvprofit.com, gemini.google).

The prompts in this text can be adopted directly and adjusted. This keeps fan love visible, without it becoming a deepfake that harms someone.