Home › News

Safe Use of Photo Editing Apps and Deepfake Risks

30.09.2026

Opening a photo editor to adjust lighting or remove a background feels harmless—and most of the time, it is. The same software categories that crop a holiday snapshot or smooth a complexion, however, also contain the tools capable of producing non-consensual intimate imagery. The line between routine retouching and the creation of a deepfake pornographic image is thinner than most casual users realise, and crossing it carries severe ethical, social, and legal consequences.

What deepfake technology actually does

The term "deepfake" combines "deep learning" and "fake." At its core, a deepfake is synthetic media generated or altered by a neural network—a type of machine-learning model inspired by the structure of the human brain. The model is trained on large collections of images or video frames. Once trained, it can swap one person's face onto another's body, alter expressions, or generate entirely fabricated scenes that look convincingly real.

Not all deepfakes are malicious. Film production uses similar techniques for digital de-aging or stunt doubling. Academic researchers generate synthetic faces to test facial-recognition systems under controlled conditions. The harm arises when the technology targets a real, identifiable person without their consent, particularly in intimate or pornographic contexts.

How photo editing apps intersect with deepfake creation

Modern photo and video editing applications exist on a spectrum. At one end sit lightweight tools that adjust exposure or apply pre-set filters. Further along are applications offering face-swap features, body-modification sliders, and AI-powered "enhancement" functions. Some market themselves openly as novelty face-swap apps; others bury these capabilities inside broader editing suites.

The mechanism is often the same: the user supplies a source image containing a target face and a destination image or video template. The software maps facial landmarks—points around the eyes, nose, mouth, and jaw—and uses a trained model to transplant the source face onto the destination body, blending skin tones and adjusting lighting to mask the seam. When the destination template is pornographic and the source face belongs to someone who has not consented, the result is classified as non-consensual intimate imagery, often referred to as image-based sexual abuse.

The real-world harm of non-consensual deepfake imagery

Creating or distributing deepfake pornography is not a victimless prank. Research into the impact on survivors of non-consensual intimate imagery documents psychological harm that mirrors the effects of physical sexual assault: anxiety, depression, post-traumatic stress, and profound social withdrawal. Targets frequently face harassment, reputational damage, and employment consequences. The knowledge that such images exist online—and may be re-uploaded faster than they can be removed—creates a persistent state of hypervigilance.

The harm extends beyond the individual. Normalising the fabrication of intimate material erodes trust in photographic evidence generally. When any image might be synthetic, genuine documentation of wrongdoing becomes easier to dismiss as a deepfake—a phenomenon researchers call the "liar's dividend."

Who is affected

Studies of deepfake pornography consistently find that the overwhelming majority of targets are women. A significant proportion involve public figures—politicians, journalists, entertainers—but an increasing share targets private individuals: colleagues, classmates, former partners, or acquaintances. The barrier to creation has lowered as tooling has become more accessible, meaning the threat is no longer confined to those in the public eye.

Legal landscape across jurisdictions

Laws addressing deepfake pornography vary widely, but the%20the trend is toward criminalisation. Several jurisdictions have enacted specific statutes:

• United Kingdom: The Online Safety Act 2023 criminalises the sharing of intimate deepfake imagery. Further legislation under the Sexual Offences Act has been expanded to cover the creation of such material even without distribution.
• United States: A patchwork of state laws addresses non-consensual deepfakes. Federal proposals have been introduced but, as of early 2025, no unified federal statute specifically names deepfake pornography, though existing harassment and revenge-porn laws may apply.
• European Union: The EU AI Act imposes transparency duties for certain deepfakes under Article 50. Separate privacy, harassment and criminal laws may apply to non-consensual intimate imagery, and national rules differ.
• Australia: Several states have enacted offences targeting the production and distribution of deepfake pornography, with custodial sentences available.

Even where specific deepfake laws are absent, existing legislation on harassment, defamation, voyeurism, and copyright may still apply. The person whose image is used retains personality or portrait rights in many civil-law systems, opening the creator to damages claims.

Recognising unsafe or unethical application features

Not every AI-powered editing feature is a risk. The danger signals are specific and recognisable:

• Face-swap onto pre-loaded nude or semi-nude templates: Applications that offer a library of pornographic bodies onto which a user can project any face are explicitly designed for non-consensual intimate imagery creation.
• "Nudification" or clothes-removal tools: Some applications market AI that synthesises what a clothed person might look like undressed. These tools fabricate intimate material from innocuous source photos and have no legitimate editing purpose.
• Absence of consent verification: Ethical face-swap applications require both the source face and the destination body to be provided by the same user or require documented consent. Applications that impose no such check invite misuse.
• Anonymised or ephemeral export with no watermarking: Responsible AI-generated or AI-altered media should carry metadata or visible indicators of manipulation. Tools that deliberately strip provenance information make detection and accountability harder.

Principles for safe and ethical use

Using photo editing applications responsibly does not require abandoning AI features altogether. It requires applying a consistent ethical framework before every edit:

Consent is the non-negotiable baseline

If an edit places a real person's likeness into a context they have not agreed to—particularly an intimate, degrading, or misleading one—the edit is unethical regardless of the technical ease of making it. Consent must be informed (the person understands what will be created), specific (they agree to this particular use), and revocable (they can withdraw permission). Consent to appear in one context does not transfer to another.

Consider the downstream effects

Even images shared privately among a small group can be screenshotted, saved, and redistributed without the creator's control. The internet lacks a reliable "delete" function. A responsible editing decision accounts for the worst-case distribution scenario, not merely the intended audience.

Preserve provenance

When using AI-powered tools, retain metadata that records the alteration. Some applications embed C2PA (Coalition for Content Provenance and Authenticity) signals that travel with the file, marking it as AI-altered. Where this is not automatic, adding a visible watermark or caption stating that the image has been manipulated is a reasonable safeguard.

Audit application permissions and data handling

Photo editing applications routinely request access to camera rolls and cloud storage. An application that processes images server-side—sending your photos to a remote computer for analysis—may retain those images for model training or other purposes. Review privacy policies for data retention periods, training-data usage clauses, and jurisdiction of storage. Prefer applications that process images locally on your device where possible.

What to do if you encounter deepfake imagery

Discovering that intimate deepfake imagery of yourself or someone you know has been created or shared demands a structured response:

1. Document before reporting: Take screenshots capturing the image, the URL, the platform, the uploader's account details, and the date. This evidence may be needed for law enforcement or civil proceedings.
2. Report to the platform: Most major social media and content-hosting platforms have specific policies against non-consensual intimate imagery. Use the platform's reporting mechanism, selecting the category that most precisely describes the violation.
3. Seek specialist support: Organisations such as the Revenge Porn Helpline (UK), Cyber Civil Rights Initiative (US), and similar bodies in other countries offer practical guidance on content removal, legal options, and emotional support.
4. Consider legal advice: Depending on the jurisdiction, civil remedies (injunctions, damages) and criminal prosecution may be available. A solicitor or attorney experienced in image-based abuse can clarify options.
5. Avoid direct confrontation with the perpetrator: Engaging with the creator or distributor can escalate harassment and alert them to destroy evidence before it is preserved.

The responsibility of application developers

The burden of safety does not rest solely on users. Developers of photo editing software that includes AI-driven face or body manipulation bear a duty to design hostile to misuse:

• Safety by design: Restrict face-swap and body-modification features to contexts where consent can be reasonably verified. Disable or withhold nudification tools entirely.
• Friction for high-risk actions: Introduce deliberate steps—confirmation dialogs, processing delays, or consent prompts—that give users a moment to reconsider before generating sensitive content.
• Provenance by default: Embed C2PA or similar provenance markers in all AI-altered outputs. Make it technically difficult to strip these markers.
• Moderation of shared content: Where the application includes a community or sharing feature, deploy automated detection systems trained to flag likely non-consensual intimate imagery for human review.
• Transparent terms: State clearly in user agreements that creating non-consensual intimate imagery using the tool is prohibited and will result in account termination and, where legally appropriate, referral to authorities.

Detection: how synthetic imagery is identified

As generation techniques improve, detection becomes an arms race. Current methods include:

• Artefact analysis: AI-generated faces often contain subtle inconsistencies—blurred teeth, mismatched earrings, irregular iris reflections, or unnatural skin texture at the blending boundary.
• Frequency-domain examination: Deepfake images may exhibit statistical patterns in their high-frequency components that differ from unmodified photographs. Forensic tools analyse these patterns to estimate the likelihood of manipulation.
• Provenance metadata: If the file retains C2PA or similar metadata, the alteration history is available for inspection. Absence of provenance data is not itself proof of manipulation, but its presence is strong evidence of it.
• Physiological inconsistency detectors: Some tools check for impossible geometry—shadows that fall in contradictory directions, reflections that do not match the scene, or facial landmarks that violate normal proportions.

No single detector is definitive. The most reliable assessments combine multiple signals and, where stakes are high, human expert review.

A framework for everyday decisions

Not every editing decision requires a formal ethical analysis. For routine adjustments—cropping, colour correction, minor blemish removal—common sense suffices. When an edit approaches the territory of altering someone's appearance, placing them in a new context, or generating synthetic content, a brief mental checklist is valuable:

1. Does this edit misrepresent the person's identity, actions, or circumstances?
2. Would the subject consent to this specific use if asked?
3. Could this image cause harm if distributed beyond my intended audience?
4. Am I using a tool whose design encourages or normalises non-consensual manipulation?
5. Would I be comfortable defending this edit in a public or professional setting?

If any answer gives pause, the edit should not proceed. The convenience of a tool never outweighs the dignity and autonomy of the person depicted.

The technology behind deepfake creation will continue to advance, and photo editing applications will continue to integrate more powerful AI features. Safety depends not on abstaining from image editing entirely, but on maintaining a clear ethical boundary: any manipulation that fabricates intimate or misleading representations of a real person without their informed consent is a misuse of the tool, regardless of what the software permits. The most important safeguard is a decision made before the first click.