# Photo Manipulation The deliberate alteration of photographic images to misrepresent the depicted scene — removing people from photographs, adding people who were not present, retouching to flatter or disparage, recropping to change meaning, layering composites to suggest events that did not occur, color-grading to amplify or dampen emotional response. The category's analytical interest for the disinformation lexicon is in the continuous lineage from 19th-c. darkroom manipulation through 20th-c. political censorship (Stalin's purges as canonical case) through the digital-era Photoshop generation and into the contemporary AI-generated [[Deep Fakes]] and *synthetic media* generation. Each technological generation has lowered the cost and raised the realism of the manipulation; the strategic logic is preserved. ## The Stalin canonical case The historically-canonical study of political photo manipulation is David King's *The Commissar Vanishes: The Falsification of Photographs and Art in Stalin's Russia* (1997). King, a British design-historian, assembled the photographic-evidence-record of Stalin-era systematic-removal of purged officials from previously-published photographs. The pattern: 1. **A high-ranking Soviet official appears in widely-distributed propaganda photographs** in 1920s–1930s editions of *Pravda*, *Izvestia*, books, monuments, posters 2. **The official is denounced, tried, executed, or sent to the camps** in one of the multiple purges across the Stalin period 3. **Subsequent editions of the same photographs** appear with the official *removed* from the image — usually badly, with visible artifacts at the seams; sometimes well, with the absence almost invisible 4. **Previously-published image-bearing artifacts** are recalled, destroyed, or modified — books with offending photographs are reissued; monuments with carved-relief images are altered; museum collections are systematically updated 5. **The official's existence is progressively erased** — from the photographic record, from the textual record, from the institutional memory King's book reproduces hundreds of paired before-and-after images documenting the practice. The book is itself an artifact of the post-Soviet archival opening that made the Western-scholarly reconstruction possible. The Stalin case is canonical because it shows the *systematic-political-purpose* deployment at maximum intensity. The mechanism — alter the visual record to alter the political-historical record — has appeared in many subsequent contexts at varying intensity and obviousness. ## The technological generations ### Pre-digital (1840s–1990) Photographic-manipulation history is approximately as old as photography itself. Early darkroom techniques (combination printing, double-exposure, retouching the negative, retouching the print) were available from the 1860s onward. The Cottingley Fairies hoax (1917–1920), the Stalin-era operations described above, the wartime propaganda manipulations across multiple combatants, and the Cold War-era US-Soviet image-warfare all operated within pre-digital constraints. The skill-and-equipment threshold was high; the volume was correspondingly low. ### Photoshop era (1990–2017) Adobe Photoshop's commercial release in 1990 dramatically lowered the cost-and-skill threshold for high-quality photo manipulation. By the late 1990s, magazine cover-retouching, advertising-image construction, and political-campaign image management had become widely-practiced. Throughout the 2000s, *Photoshop* became a verb in general usage (*photoshopped* as adjective for any altered image), and audience-side expectations shifted toward general skepticism of photographic evidence. The Photoshop-era manipulation infrastructure produced a substantial sub-literature on *image forensics* — Hany Farid at Dartmouth (later UC Berkeley) substantially founded the academic field through the 1990s–2010s. The forensic-analysis toolkit included error-level analysis, light-source-direction inconsistency detection, JPEG-compression-artifact analysis, and shadow-and-perspective geometric verification. The cat-and-mouse arms race between manipulation and detection has continued through every subsequent generation. ### The generative AI era (2017+) The post-2017 introduction of generative-AI image-synthesis (GAN-based architectures, then diffusion-model architectures) has been a discontinuous capability-jump. The 2017 *deepfakes* Reddit-user (subsequently the *deepfake* common noun) demonstrated face-replacement at quality-levels not previously available; 2022's DALL-E 2, [[Stable Diffusion]], and Midjourney moved into the photorealistic-text-to-image regime; 2024's video-generation models (Sora, Veo) extended the capability to motion content; 2025's increasingly-multimodal frontier models have continued to escalate the realism. The forensic-detection field's response has been mixed. Detection algorithms exist but have generally lagged the generation capability; the recent literature suggests the *digital-forensic* approach of analyzing the artifact itself is becoming progressively less reliable, and that contemporary verification work has to depend substantially on *provenance* (cryptographic signing of source authentic content; C2PA standards; cross-evidence triangulation with other documented sources) rather than on *artifact analysis*. ## Specific high-profile contemporary cases A non-exhaustive partial catalog: - **2003 *Los Angeles Times* [[Iraq War]] cover** — Brian Walski composite photograph of a British soldier and Iraqi civilians; the photographer was fired upon detection - **2008 Iranian missile-test image** — Sepah News released image showing four missiles in air; one was photoshopped from a previous image to disguise a failed launch; widely-reproduced by Western press before detection - **2014 Russian *NTV*-broadcast Donbass casualty photos** — multiple instances of recycled Syrian-conflict imagery presented as Donbass-conflict imagery - **2017 Trump inauguration crowd-size images** — photographic discrepancies between Spicer-claimed audience size and aerial-photograph evidence - **2019 Kashmir-coverage doctored Indian-government imagery** — multiple documented instances of altered military-presence imagery - **2024 Trump-rally documented image manipulation** — AI-generated *Swifties for Trump* / *Black supporters* / *crowd composite* imagery; the practice of using AI-generated *typical supporter* imagery has become normalized - **2024+ generative-AI political-attack imagery** — AI-generated Trump-arrest imagery (May 2024 fake images); AI-generated *Biden-confused* imagery; AI-generated [[Kamala Harris]] campaign-attack imagery; the volume substantially exceeds any prior period ## The contemporary detection-and-provenance infrastructure The post-2020 institutional response to the AI-generation challenge: - **Coalition for Content Provenance and Authenticity (C2PA)** — Adobe / Microsoft / Intel / BBC / Sony industry-standard for cryptographic-signing of content provenance; partial adoption - **Content Authenticity Initiative** — adjacent industry-standard organization - **Digital-forensic services** — Sensity, Truepic, Reality Defender provide commercial deepfake-detection - **Academic-research infrastructure** — Hany Farid's Berkeley group; multiple university labs; DARPA Media Forensics (MediFor) and Semantic Forensics (SemaFor) programs - **Journalist-side training** — Reuters Institute, *Poynter*, *Bellingcat* maintain training materials and verification toolkits - **Platform-side labels** — Meta / Google / [[YouTube]] / TikTok have varying *generated-with-AI* / *altered-content* labels; enforcement is inconsistent None of these is sufficient; in combination they provide partial defense against the underlying capability gap. ## See Also - [[Deep Fakes]] — the AI-generated-media successor category - [[Source Credibility]] — the audience-side perception infrastructure that photo manipulation operates against - [[Disinformation]] — the parent category - [[Misinformation]] — the unintentional-spread sibling category for circulated-without-context manipulated images - [[Viral Misinformation]] — the spread-dynamics category that amplifies manipulated imagery - [[Disinformation Hub]] — Subject hub - [[Disinformation Lexicon]] — alphabetical MOC