Deepfake vs Morphed Image: Difference, Law and the 2-Hour Clock

The morphed image is decades old, a real photograph edited so a face sits on another body; the deepfake is its machine-age successor, a model trained on the person generating face, voice and motion outright. The law's design is worth stating precisely: for the victim, technique is irrelevant, the same offences and the same two-hour clock answer both, while for the platforms and tools, the synthetic form alone carries the SGI labelling regime. Topics 75 and 92 built the regimes; this note, as asked, is the dedicated comparison.

1. The Technique Line

Generated against edited

Figure 1: Generated against edited

The morphed image. A real photograph altered by hand with editing tools: the cut-and-paste of a face onto another body, retouched into apparent unity. It is bounded by its source material, classically static, and it leaves edit traces, mismatched lighting, edges, resolution, the traditional forensic tells.
The deepfake. Content synthesised by a model trained on images or recordings of the person: a face generated in motion, a voice cloned into new speech, video with expression and lip-sync. Nothing was cut from anywhere, the machine made the frames, and realism improves with every model generation, shifting detection from visible artefacts to statistical and provenance analysis.
Why the line matters twice. For harm, barely: both put the person where they never were. For regulation, decisively: the deepfake is synthetically generated information, so the 2026 labelling, declaration and verification duties attach to the tools and platforms in its chain, while the hand-morph, not being synthetic, rides the general due diligence categories (Topic 75)

2. One Victim, One Law

Figure 2: Shared offences, shared clock, parted regimes

The shared offences. s.66E for the violation of bodily privacy, ss.67 and 67A where the content is obscene or sexually explicit, s.67B where a child is depicted, however the depiction was made, with BNS defamation, forgery, intimidation and extortion as the deployment supplies: the offence looks at the image's effect, never its software (Topics 60, 61)
The shared clock. The intermediary rules place complaints of content exposing private areas, or in the nature of impersonation including artificially morphed images, on the short clock, compressed by the 2026 amendment to two hours, and the ordered-removal clock to three: the victim's remedy is identical for deepfake and hand-morph (Topics 74, 75)
The extra regime for the synthetic form. The deepfake additionally engages the SGI pipeline, creation-tool labels, upload declarations, SSMI verification and prominent display, so an unlabelled deepfake is a due diligence failure in the chain even before any victim complains; the morph triggers none of this (Topic 75)
Remedies beyond the criminal law. Personality rights injunctions for public figures against both forms, the civil defamation and privacy actions, and the grievance-officer-to-GAC ladder for takedown (Topics 88, 92)
The forensic difference. Proof is where technique returns: the s.79A examination of a morph hunts edit history, composite boundaries and source images; of a deepfake, generation artefacts, model fingerprints and provenance metadata, with the hash and custody disciplines common to both (Topic 82)

⚠ Exam trap

State the design in one sentence: the offences and the two-hour clock are technique-neutral, while the SGI labelling regime attaches only to the synthetic form, and support it from the rule text, which expressly names artificially morphed images on the short complaint clock alongside impersonation content. For children, close every answer with s.67B's indifference to how the depiction was made, and in proof questions, split the forensic fingerprints: edit traces for the morph, generation artefacts and provenance for the deepfake.

3. Frequently Asked Questions

What is the difference between a deepfake and a morphed image?

A morphed image is a real photograph manually edited, classically a face cut and pasted onto another body, bounded by its source material and betrayed by edit traces. A deepfake is synthesised by an AI model trained on the person, generating face, voice or full video anew, capable of motion and speech and increasingly free of visible artefacts. Both wrong the victim identically, and the offences, Sections 66E, 67 to 67B, defamation and forgery, apply to both; the deepfake alone is synthetically generated information attracting the 2026 labelling and verification duties.

How quickly must platforms remove deepfake or morphed intimate imagery?

On complaint, within two hours under the 2026 amendments to the intermediary rules, which expressly cover content exposing private areas and impersonation content including artificially morphed images; ordered removals generally run on a three-hour clock. Failure forfeits the Section 79 safe harbour through Rule 7, and the underlying content remains prosecutable under Sections 66E, 67, 67A or 67B with the BNS offences the deployment attracts.

4. Related Topics

Topic 75: Synthetically Generated Information. The regime the deepfake alone attracts.
Topic 92: AI and Indian Cyber Law. The wider AI harm map.