[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M77T6H6M9DXKJ96CKTTMP6Y7":3,"skill-same:01M77T6H6M9DXKJ96CKTTMP6Y7":56},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":24,"commitSha":25,"skillDirectory":26,"forksCount":27,"repositoryHtmlUrl":28,"repositoryHomepageUrl":6,"ownerHtmlUrl":29,"ownerBlogUrl":30,"ownerDescription":6,"updatedAt":31,"install":32,"resources":45},"01M77T6H6M9DXKJ96CKTTMP6Y7","is-this-photo-real",null,"Verify whether an image or video is authentic, original and correctly captioned — provenance checks, error level analysis, noise and JPEG compression analysis, clone and copy-move detection, lighting and shadow consistency, C2PA Content Credentials, deepfake and AI-generation tells, and the honest limits of AI-detector tools. Use when fact-checking a photo or video, checking for a deepfake or AI-generated image, spotting manipulation, or testing whether footage is recycled or miscaptioned. Applies to KYC and onboarding fraud, insurance claim review, disinformation analysis, and evidence admissibility. Reference at useosint.com\u002Fskills\u002Fis-this-photo-real.","Reihantt6\u002Fsns-myagent","Reihantt6","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F219822095?v=4","01M77KK94PV8B7C2D9XWXNS8DJ","Mohamad Reihan tegar prasetya","01M77KK958QFMTHYXGXFV7KZ79",2,"MIT","https:\u002F\u002Fgithub.com\u002FReihantt6\u002Fsns-myagent\u002Fblob\u002Fmain\u002Fagent\u002Fskills\u002Fis-this-photo-real\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\ndescription: \"Verify whether an image or video is authentic, original and correctly captioned — provenance checks, error level analysis, noise and JPEG compression analysis, clone and copy-move detection, lighting and shadow consistency, C2PA Content Credentials, deepfake and AI-generation tells, and the honest limits of AI-detector tools. Use when fact-checking a photo or video, checking for a deepfake or AI-generated image, spotting manipulation, or testing whether footage is recycled or miscaptioned. Applies to KYC and onboarding fraud, insurance claim review, disinformation analysis, and evidence admissibility. Reference at useosint.com\u002Fskills\u002Fis-this-photo-real.\"\n---\n# Is this photo real\n\nVerification order is the whole skill: **provenance first, pixels last.** Finding the\nearliest copy and reading its caption settles more cases than every forensic filter\ncombined, and it produces evidence you can show someone. Pixel forensics produces a\ncolourful heatmap and an argument.\n\nThe beginner mistake is running error level analysis on a downloaded JPEG and announcing\nthe image is fake. The second is looking for deepfakes: the overwhelming majority of\ndeceptive media is **real footage with a false caption** — right pixels, wrong war,\nwrong year, wrong country.\n\n## Triage: what question are you actually answering\n\n| The claim under test | Do this first | Not this |\n|---|---|---|\n| \"This shows event X in place Y\" | `find-the-original-image`, then `geolocate-from-pixels` | Any forensic filter. Recontextualisation leaves no pixel trace at all. |\n| \"This is an unaltered photograph\" | Signal-level analysis, on the least-processed copy you can obtain | Analysing a screenshot or a platform download; both destroy the signal. |\n| \"This person said this on video\" | Provenance, then audio-visual consistency, then face-boundary behaviour frame by frame | An AI-detector score. |\n| \"This image was AI-generated\" | Absent camera physics and incoherent object structure | A detector verdict on its own. |\n| \"This screenshot is genuine\" | Layout, font and interface-version consistency; the underlying record if one exists | Image forensics. Fabricated screenshots are made in a browser, not an image editor. |\n\nThen always ask what the image would look like if the claim were **true**, write it\ndown, and check for those things specifically. Verification tests a hypothesis;\nhunting for anomalies fails, because anomalies are everywhere.\n\n## Method\n\n1. **Get the best copy.** Every re-encode, resize and screenshot destroys forensic\n   signal. Chase the original upload or the agency version, not the platform rendition.\n   Hash it, work on copies. If all you have is a screenshot, say so and lower every\n   downstream conclusion.\n2. **Provenance.** Run `find-the-original-image`; for video, extract and search\n   keyframes. You want an earlier appearance, a different caption, a photographer\n   credit, and an on-page date corroborated through `read-deleted-pages`. An earlier copy\n   with a different caption ends the case.\n3. **Metadata.** Run `secrets-in-file-metadata`: editing chain, thumbnail-versus-image\n   comparison, timestamp inconsistencies, whether MakerNotes fit the claimed device.\n4. **Provenance signing.** Check for C2PA Content Credentials.\n5. **Internal consistency.** Signage language, plates, currency, uniforms, vehicle\n   models, season, weather and shadows against the claimed date and place, via\n   `geolocate-from-pixels`. Ordinary detective work, more productive than forensics.\n6. **Physical consistency.** Lighting, shadows, reflections, perspective.\n7. **Signal-level forensics.** Noise residuals, JPEG quantisation and double compression,\n   clone detection, colour-filter-array traces. Easy to over-read, worthless on a\n   platform-processed file.\n8. **Write what you verified**, not a verdict.\n\nTools and their failure modes: [reference\u002Ftool-catalogue.md](reference\u002Ftool-catalogue.md).\nOrdered by cost: [reference\u002Fverification-checklist.md](reference\u002Fverification-checklist.md).\n\n## Lighting, shadow and geometry — the checks that hold up\n\nNo tooling, reasoning you can explain to an editor or a court, immune to recompression.\nThis is where to spend your time. **Shadow convergence** is the strongest. Sunlight is\nparallel, so in a perspective image, lines drawn from each shadow's tip through the base\nof the object that cast it must all meet at one point — the projection of the light\nsource. Draw three or four. An object whose line refuses to meet the others was probably\nnot in the original scene; sloped ground is the confound, so use one plane only. Then:\nshadow direction and penumbra hardness should be consistent across an outdoor scene;\nspecular highlights in eyes, glass and polished metal should agree on where the lights\nare; a reflection must show what is in front of it, correctly placed and reversed;\nparallel lines should converge on a common vanishing point with eye level consistent for\npeople on one ground plane; and real lenses leave an optical signature — consistent depth\nof field, chromatic aberration at high-contrast edges, vignetting, a noise floor that\nvaries with brightness. An element carrying none of that, in an image that has it\nelsewhere, was added.\n\n## Error level analysis, and why it is mostly used wrongly\n\nELA re-saves the image at a known JPEG quality and displays the difference, on the theory\nthat a region compressed a different number of times responds differently. Four reasons\nit produces confident nonsense: it responds to **content**, so edges and texture light up\nwhile flat sky and skin go dark, meaning every image has \"suspicious bright regions\"; one\nre-save destroys it, so ELA on a social-media download describes the platform's encoder\nand nothing earlier; it cannot localise a modern edit, because content-aware fill,\ngenerative editing and a full re-save leave no differential history to find; and it fails\nin both directions, with bright regions on untouched images and clean output on\nmanipulated ones both routine.\n\nWhere it earns its place: on a single-generation JPEG straight from a camera, a pasted\nregion from a differently-compressed source can genuinely show up. Narrow case. Use it\nas one weak input, only on least-processed files, never as the basis of a published\nclaim. More defensible relatives — quantisation-table comparison against camera\nsignatures, and double-compression detection — are also defeated by platform processing.\n\n## C2PA and Content Credentials\n\nC2PA binds a cryptographically signed manifest to a file recording capture and edit\nhistory. Where it exists it is the strongest provenance evidence available, because it\nis verifiable rather than inferential. A **valid** manifest means the signer asserts\nthis history, the file is unchanged since signing, and you know who to hold\nresponsible — not that the content is true. A signed photograph of a staged scene is a\nsigned photograph.\n\n**Absence** means almost nothing: most cameras do not sign, most editing pipelines do\nnot preserve manifests, and platforms strip them during re-encoding. Missing\ncredentials are the default state, not a red flag. Same for the IPTC\ndigital-source-type field used to label synthetic media, and for model-specific\ninvisible watermarks — a positive is strong where you can check it, a negative only\nsays one vendor's mark was not found.\n\n## AI generation — durable tells and tells that rot\n\nAnything resting on a model's current weaknesses will be fixed. Prefer tells\ngrounded in physics and structure.\n\n**Durable, because they need a world model the generator does not have:** impossible\nlighting (inconsistent shadow directions, missing shadows under objects, a subject\nlit from a direction with no source); structural incoherence in background objects (a\nbicycle frame that does not connect, a railing whose baluster spacing changes, stair\ntreads that do not line up, a strap that vanishes and resumes, patterned fabric whose\npattern ignores the folds); text degradation, especially small, repeated or peripheral\ntext; contact and occlusion errors, such as a hand around a cup that does not enclose\nit, or feet not meeting the ground; absent camera physics — no sensor noise, no\nchromatic aberration, uniform focus, and too little high-frequency detail, which is\nwhat \"over-smooth skin\" actually is; and no plausible provenance at all.\n\n**Ages badly — check, but do not rest on:** finger and tooth counts, ear asymmetry,\ngarbled foreground text, mangled jewellery, suspiciously symmetrical faces.\n\n**Detector tools.** A confident score with no auditable reasoning. They false-positive\non compressed, resized, upscaled, heavily edited and low-light real photographs,\nfalse-negative against generators newer than their training data, and are adversarially\nfragile — mild recompression moves scores. Run more than one, treat them as a weak\nsignal, never publish a conclusion resting on one. If your evidence is a percentage from\na website, you have no evidence.\n\n## Video\n\n- **Container and encoder.** `ffprobe -show_format -show_streams` and MediaInfo give the\n  encoder string, frame rate, rotation matrix and track structure; values typical of a\n  platform re-encode mean you do not have an original. `ffprobe -show_frames` exposes\n  frame types — duplicated frames mean frame-rate conversion or inserted slow motion, an\n  unexplained keyframe mid-way through a static shot can mark a splice, and interlacing\n  or telecine artifacts reveal a pipeline nobody mentioned.\n- **Generation loss.** Blockiness, banding and mosquito noise stack with each re-encode.\n  Heavily degraded footage is old, widely copied, or both — and every signal-level test\n  on it is void.\n- **Audio.** Lip-sync drift, room acoustics that do not match the visible space,\n  ambience that does not change when the camera goes indoors, noise-floor jumps at edit\n  points. The weakest link in most fabricated video and the least examined.\n- **Face-swap behaviour.** Flicker or blur at hairline and jaw, face lighting not\n  tracking head movement, teeth and tongue degrading during speech, the face at a\n  different resolution from the frame, breakdown on profile turns and hand occlusion.\n- **Keyframes to reverse search.** The most common outcome of a video verification is\n  finding the video, older, elsewhere. Extraction commands live in\n  `find-the-original-image`.\n\n## Where this goes wrong\n\n- **Every filter has a base-rate problem.** Run six forensic tools on an authentic\n  photograph and something will look anomalous. Anomaly is the normal condition of\n  real images.\n- **The platform did it.** Resizing, re-encoding, chroma subsampling and metadata\n  stripping produce artifacts people attribute to manipulation. Establish processing\n  history before interpreting any artifact.\n- **You will be handed the worst copy** — a screenshot of a repost of a crop. Most\n  signal-level analysis is invalid on it, and the honest report says so.\n- **\"Not manipulated\" is not a finding.** Absence of detected manipulation is a\n  statement about your tests, not about the image.\n- **A real photo can be entirely misleading.** Selective framing, staged scenes and\n  a true image with a false caption all pass every forensic test.\n- **Debunking amplifies.** A detailed refutation spreads the original claim — an\n  editorial judgement worth making deliberately. Material is also sometimes seeded to be\n  discovered and debunked, or to see who investigates; see\n  `investigate-without-getting-made`.\n- **Identification from resemblance** is the highest-consequence error here — \"this is\n  person Z because they look alike\" is not a finding.\n\n## Confidence grading\n\nGrade each claim separately — one image can have a confirmed origin, a contradicted\ncaption and unconfirmed authenticity at once.\n\n- **Confirmed original and correctly described** — earliest copy located from a plausible\n  originator, date independently corroborated, location visually verified, internal\n  details consistent, nothing contradicted in metadata or geometry.\n- **Confirmed recontextualised** — the same image demonstrably published earlier with\n  a different, better-sourced caption. The most common positive finding here.\n- **Probably manipulated** — a specific, describable physical or geometric\n  inconsistency you can point at, ideally with a source for the inserted element.\n  Not a heatmap.\n- **Probably synthetic** — multiple durable generative tells, no provenance history,\n  no camera-consistent compression or metadata, no earlier copies. Name the tells.\n- **Unconfirmed** — no earlier copy found, nothing wrong found. Where most cases end.\n- **Cannot be assessed** — the available copy is too processed for the tests the question\n  needs. Say which tests were invalid and why.\n\n## Worked example\n\nA video circulates showing an explosion, captioned as a strike on a named city that\nmorning.\n\n`yt-dlp` pulls the best rendition rather than a screen recording. `ffprobe` shows a\nframe rate and encoder string typical of a platform re-encode, so signal-level tests\nare off the table — noted, not lamented.\n\nKeyframes extracted at scene changes. Frame four hits nothing. Frame nine, the only one\nwith a skyline, hits a news site from eighteen months earlier in a different country.\nProvisional answer in ten minutes. But the earlier page's video is shorter, so which is\nthe parent? The suspect version has a hard cut and a noise-floor jump in the audio four\nseconds in, and the ambience after the cut does not match the visible space. Added\naudio, not added video.\n\nDead end worth recording: an AI detector scored the frames as likely synthetic. They\nare a re-encoded crop of real broadcast footage — exactly the input that fools these\ntools. Discarded.\n\nFindings: **confirmed recontextualised**, with `read-deleted-pages` corroborating the\nearlier page's date; audio **probably manipulated**; authenticity of the underlying\nvideo **cannot be assessed** at signal level given the copy available.\n\n**Reporting standard.** State what you verified, what you could not, and what each\nconclusion rests on — never a bare \"fake\" or \"real\". A defensible line reads: *the\nimage was published at least three years before the claimed event, on a news site,\ncredited to a named photographer; the location matches that credit and not the claim; no\nmanipulation was detected, but the only available copy was a platform re-encode, so\nsignal-level tests were not meaningful.* \"Our analysis shows this is fake\" is not.\nInclude the tests you ran and their negative results, the file's processing history, and\nyour assumptions, then hand the package to `write-the-intel-brief`.\n\n## Pivots\n\n| What you got | Send to |\n|---|---|\n| Earlier copies, credits, original caption | `find-the-original-image` |\n| Editing chain, device, timestamps | `secrets-in-file-metadata` |\n| Location and date verification | `geolocate-from-pixels`, `where-was-this-taken` |\n| Deleted or altered source pages | `read-deleted-pages` |\n| Publishing or seeding domain | `who-owns-this-domain`, `recon-a-domain-passively` |\n| Accounts amplifying the media | `hunt-a-handle`, `pattern-of-life-from-socials` |\n| Coordinated network behind the spread | `graph-the-network`, `find-leaks-in-the-wild` |\n| Named individuals in or credited on the media | `find-anyone` |\n\n## Legal and ethical notes\n\nPublishing an accusation of fabrication against a named person carries defamation\nrisk in most jurisdictions, and \"our tool said so\" is not a defence. Uploading\nmaterial to online forensic services discloses it to those services and, for some, to\npublic galleries — check whether a tool publishes submissions before submitting\nanything sensitive. If the media depicts a crime, a victim, or intimate content:\nminimise copies, do not redistribute, and in the case of child sexual abuse material\nstop immediately and report to the appropriate authority rather than analysing it. See\n[..\u002F..\u002FETHICS.md](..\u002F..\u002FETHICS.md).\n","# Is this photo real\n\nVerification order is the whole skill: **provenance first, pixels last.** Finding the\nearliest copy and reading its caption settles more cases than every forensic filter\ncombined, and it produces evidence you can show someone. Pixel forensics produces a\ncolourful heatmap and an argument.\n\nThe beginner mistake is running error level analysis on a downloaded JPEG and announcing\nthe image is fake. The second is looking for deepfakes: the overwhelming majority of\ndeceptive media is **real footage with a false caption** — right pixels, wrong war,\nwrong year, wrong country.\n\n## Triage: what question are you actually answering\n\n| The claim under test | Do this first | Not this |\n|---|---|---|\n| \"This shows event X in place Y\" | `find-the-original-image`, then `geolocate-from-pixels` | Any forensic filter. Recontextualisation leaves no pixel trace at all. |\n| \"This is an unaltered photograph\" | Signal-level analysis, on the least-processed copy you can obtain | Analysing a screenshot or a platform download; both destroy the signal. |\n| \"This person said this on video\" | Provenance, then audio-visual consistency, then face-boundary behaviour frame by frame | An AI-detector score. |\n| \"This image was AI-generated\" | Absent camera physics and incoherent object structure | A detector verdict on its own. |\n| \"This screenshot is genuine\" | Layout, font and interface-version consistency; the underlying record if one exists | Image forensics. Fabricated screenshots are made in a browser, not an image editor. |\n\nThen always ask what the image would look like if the claim were **true**, write it\ndown, and check for those things specifically. Verification tests a hypothesis;\nhunting for anomalies fails, because anomalies are everywhere.\n\n## Method\n\n1. **Get the best copy.** Every re-encode, resize and screenshot destroys forensic\n   signal. Chase the original upload or the agency version, not the platform rendition.\n   Hash it, work on copies. If all you have is a screenshot, say so and lower every\n   downstream conclusion.\n2. **Provenance.** Run `find-the-original-image`; for video, extract and search\n   keyframes. You want an earlier appearance, a different caption, a photographer\n   credit, and an on-page date corroborated through `read-deleted-pages`. An earlier copy\n   with a different caption ends the case.\n3. **Metadata.** Run `secrets-in-file-metadata`: editing chain, thumbnail-versus-image\n   comparison, timestamp inconsistencies, whether MakerNotes fit the claimed device.\n4. **Provenance signing.** Check for C2PA Content Credentials.\n5. **Internal consistency.** Signage language, plates, currency, uniforms, vehicle\n   models, season, weather and shadows against the claimed date and place, via\n   `geolocate-from-pixels`. Ordinary detective work, more productive than forensics.\n6. **Physical consistency.** Lighting, shadows, reflections, perspective.\n7. **Signal-level forensics.** Noise residuals, JPEG quantisation and double compression,\n   clone detection, colour-filter-array traces. Easy to over-read, worthless on a\n   platform-processed file.\n8. **Write what you verified**, not a verdict.\n\nTools and their failure modes: [reference\u002Ftool-catalogue.md](reference\u002Ftool-catalogue.md).\nOrdered by cost: [reference\u002Fverification-checklist.md](reference\u002Fverification-checklist.md).\n\n## Lighting, shadow and geometry — the checks that hold up\n\nNo tooling, reasoning you can explain to an editor or a court, immune to recompression.\nThis is where to spend your time. **Shadow convergence** is the strongest. Sunlight is\nparallel, so in a perspective image, lines drawn from each shadow's tip through the base\nof the object that cast it must all meet at one point — the projection of the light\nsource. Draw three or four. An object whose line refuses to meet the others was probably\nnot in the original scene; sloped ground is the confound, so use one plane only. Then:\nshadow direction and penumbra hardness should be consistent across an outdoor scene;\nspecular highlights in eyes, glass and polished metal should agree on where the lights\nare; a reflection must show what is in front of it, correctly placed and reversed;\nparallel lines should converge on a common vanishing point with eye level consistent for\npeople on one ground plane; and real lenses leave an optical signature — consistent depth\nof field, chromatic aberration at high-contrast edges, vignetting, a noise floor that\nvaries with brightness. An element carrying none of that, in an image that has it\nelsewhere, was added.\n\n## Error level analysis, and why it is mostly used wrongly\n\nELA re-saves the image at a known JPEG quality and displays the difference, on the theory\nthat a region compressed a different number of times responds differently. Four reasons\nit produces confident nonsense: it responds to **content**, so edges and texture light up\nwhile flat sky and skin go dark, meaning every image has \"suspicious bright regions\"; one\nre-save destroys it, so ELA on a social-media download describes the platform's encoder\nand nothing earlier; it cannot localise a modern edit, because content-aware fill,\ngenerative editing and a full re-save leave no differential history to find; and it fails\nin both directions, with bright regions on untouched images and clean output on\nmanipulated ones both routine.\n\nWhere it earns its place: on a single-generation JPEG straight from a camera, a pasted\nregion from a differently-compressed source can genuinely show up. Narrow case. Use it\nas one weak input, only on least-processed files, never as the basis of a published\nclaim. More defensible relatives — quantisation-table comparison against camera\nsignatures, and double-compression detection — are also defeated by platform processing.\n\n## C2PA and Content Credentials\n\nC2PA binds a cryptographically signed manifest to a file recording capture and edit\nhistory. Where it exists it is the strongest provenance evidence available, because it\nis verifiable rather than inferential. A **valid** manifest means the signer asserts\nthis history, the file is unchanged since signing, and you know who to hold\nresponsible — not that the content is true. A signed photograph of a staged scene is a\nsigned photograph.\n\n**Absence** means almost nothing: most cameras do not sign, most editing pipelines do\nnot preserve manifests, and platforms strip them during re-encoding. Missing\ncredentials are the default state, not a red flag. Same for the IPTC\ndigital-source-type field used to label synthetic media, and for model-specific\ninvisible watermarks — a positive is strong where you can check it, a negative only\nsays one vendor's mark was not found.\n\n## AI generation — durable tells and tells that rot\n\nAnything resting on a model's current weaknesses will be fixed. Prefer tells\ngrounded in physics and structure.\n\n**Durable, because they need a world model the generator does not have:** impossible\nlighting (inconsistent shadow directions, missing shadows under objects, a subject\nlit from a direction with no source); structural incoherence in background objects (a\nbicycle frame that does not connect, a railing whose baluster spacing changes, stair\ntreads that do not line up, a strap that vanishes and resumes, patterned fabric whose\npattern ignores the folds); text degradation, especially small, repeated or peripheral\ntext; contact and occlusion errors, such as a hand around a cup that does not enclose\nit, or feet not meeting the ground; absent camera physics — no sensor noise, no\nchromatic aberration, uniform focus, and too little high-frequency detail, which is\nwhat \"over-smooth skin\" actually is; and no plausible provenance at all.\n\n**Ages badly — check, but do not rest on:** finger and tooth counts, ear asymmetry,\ngarbled foreground text, mangled jewellery, suspiciously symmetrical faces.\n\n**Detector tools.** A confident score with no auditable reasoning. They false-positive\non compressed, resized, upscaled, heavily edited and low-light real photographs,\nfalse-negative against generators newer than their training data, and are adversarially\nfragile — mild recompression moves scores. Run more than one, treat them as a weak\nsignal, never publish a conclusion resting on one. If your evidence is a percentage from\na website, you have no evidence.\n\n## Video\n\n- **Container and encoder.** `ffprobe -show_format -show_streams` and MediaInfo give the\n  encoder string, frame rate, rotation matrix and track structure; values typical of a\n  platform re-encode mean you do not have an original. `ffprobe -show_frames` exposes\n  frame types — duplicated frames mean frame-rate conversion or inserted slow motion, an\n  unexplained keyframe mid-way through a static shot can mark a splice, and interlacing\n  or telecine artifacts reveal a pipeline nobody mentioned.\n- **Generation loss.** Blockiness, banding and mosquito noise stack with each re-encode.\n  Heavily degraded footage is old, widely copied, or both — and every signal-level test\n  on it is void.\n- **Audio.** Lip-sync drift, room acoustics that do not match the visible space,\n  ambience that does not change when the camera goes indoors, noise-floor jumps at edit\n  points. The weakest link in most fabricated video and the least examined.\n- **Face-swap behaviour.** Flicker or blur at hairline and jaw, face lighting not\n  tracking head movement, teeth and tongue degrading during speech, the face at a\n  different resolution from the frame, breakdown on profile turns and hand occlusion.\n- **Keyframes to reverse search.** The most common outcome of a video verification is\n  finding the video, older, elsewhere. Extraction commands live in\n  `find-the-original-image`.\n\n## Where this goes wrong\n\n- **Every filter has a base-rate problem.** Run six forensic tools on an authentic\n  photograph and something will look anomalous. Anomaly is the normal condition of\n  real images.\n- **The platform did it.** Resizing, re-encoding, chroma subsampling and metadata\n  stripping produce artifacts people attribute to manipulation. Establish processing\n  history before interpreting any artifact.\n- **You will be handed the worst copy** — a screenshot of a repost of a crop. Most\n  signal-level analysis is invalid on it, and the honest report says so.\n- **\"Not manipulated\" is not a finding.** Absence of detected manipulation is a\n  statement about your tests, not about the image.\n- **A real photo can be entirely misleading.** Selective framing, staged scenes and\n  a true image with a false caption all pass every forensic test.\n- **Debunking amplifies.** A detailed refutation spreads the original claim — an\n  editorial judgement worth making deliberately. Material is also sometimes seeded to be\n  discovered and debunked, or to see who investigates; see\n  `investigate-without-getting-made`.\n- **Identification from resemblance** is the highest-consequence error here — \"this is\n  person Z because they look alike\" is not a finding.\n\n## Confidence grading\n\nGrade each claim separately — one image can have a confirmed origin, a contradicted\ncaption and unconfirmed authenticity at once.\n\n- **Confirmed original and correctly described** — earliest copy located from a plausible\n  originator, date independently corroborated, location visually verified, internal\n  details consistent, nothing contradicted in metadata or geometry.\n- **Confirmed recontextualised** — the same image demonstrably published earlier with\n  a different, better-sourced caption. The most common positive finding here.\n- **Probably manipulated** — a specific, describable physical or geometric\n  inconsistency you can point at, ideally with a source for the inserted element.\n  Not a heatmap.\n- **Probably synthetic** — multiple durable generative tells, no provenance history,\n  no camera-consistent compression or metadata, no earlier copies. Name the tells.\n- **Unconfirmed** — no earlier copy found, nothing wrong found. Where most cases end.\n- **Cannot be assessed** — the available copy is too processed for the tests the question\n  needs. Say which tests were invalid and why.\n\n## Worked example\n\nA video circulates showing an explosion, captioned as a strike on a named city that\nmorning.\n\n`yt-dlp` pulls the best rendition rather than a screen recording. `ffprobe` shows a\nframe rate and encoder string typical of a platform re-encode, so signal-level tests\nare off the table — noted, not lamented.\n\nKeyframes extracted at scene changes. Frame four hits nothing. Frame nine, the only one\nwith a skyline, hits a news site from eighteen months earlier in a different country.\nProvisional answer in ten minutes. But the earlier page's video is shorter, so which is\nthe parent? The suspect version has a hard cut and a noise-floor jump in the audio four\nseconds in, and the ambience after the cut does not match the visible space. Added\naudio, not added video.\n\nDead end worth recording: an AI detector scored the frames as likely synthetic. They\nare a re-encoded crop of real broadcast footage — exactly the input that fools these\ntools. Discarded.\n\nFindings: **confirmed recontextualised**, with `read-deleted-pages` corroborating the\nearlier page's date; audio **probably manipulated**; authenticity of the underlying\nvideo **cannot be assessed** at signal level given the copy available.\n\n**Reporting standard.** State what you verified, what you could not, and what each\nconclusion rests on — never a bare \"fake\" or \"real\". A defensible line reads: *the\nimage was published at least three years before the claimed event, on a news site,\ncredited to a named photographer; the location matches that credit and not the claim; no\nmanipulation was detected, but the only available copy was a platform re-encode, so\nsignal-level tests were not meaningful.* \"Our analysis shows this is fake\" is not.\nInclude the tests you ran and their negative results, the file's processing history, and\nyour assumptions, then hand the package to `write-the-intel-brief`.\n\n## Pivots\n\n| What you got | Send to |\n|---|---|\n| Earlier copies, credits, original caption | `find-the-original-image` |\n| Editing chain, device, timestamps | `secrets-in-file-metadata` |\n| Location and date verification | `geolocate-from-pixels`, `where-was-this-taken` |\n| Deleted or altered source pages | `read-deleted-pages` |\n| Publishing or seeding domain | `who-owns-this-domain`, `recon-a-domain-passively` |\n| Accounts amplifying the media | `hunt-a-handle`, `pattern-of-life-from-socials` |\n| Coordinated network behind the spread | `graph-the-network`, `find-leaks-in-the-wild` |\n| Named individuals in or credited on the media | `find-anyone` |\n\n## Legal and ethical notes\n\nPublishing an accusation of fabrication against a named person carries defamation\nrisk in most jurisdictions, and \"our tool said so\" is not a defence. Uploading\nmaterial to online forensic services discloses it to those services and, for some, to\npublic galleries — check whether a tool publishes submissions before submitting\nanything sensitive. If the media depicts a crime, a victim, or intimate content:\nminimise copies, do not redistribute, and in the case of child sexual abuse material\nstop immediately and report to the appropriate authority rather than analysing it. See\n[..\u002F..\u002FETHICS.md](..\u002F..\u002FETHICS.md).\n",{"description":7},"main","58843b159d7a4050d6b785a98b908cce505c0b92","agent\u002Fskills\u002Fis-this-photo-real",0,"https:\u002F\u002Fgithub.com\u002FReihantt6\u002Fsns-myagent","https:\u002F\u002Fgithub.com\u002FReihantt6","reihsns.web.id","2026-08-19 20:08:18",{"source":8,"skillName":5,"commands":33,"prompt":43,"zipAvailable":44},[34,37,40],{"tool":35,"command":36},"npx","npx skills add Reihantt6\u002Fsns-myagent --skill is-this-photo-real",{"tool":38,"command":39},"bunx","bunx skills add Reihantt6\u002Fsns-myagent --skill is-this-photo-real",{"tool":41,"command":42},"pnpm","pnpm dlx skills add Reihantt6\u002Fsns-myagent --skill is-this-photo-real","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FReihantt6\u002Fsns-myagent\u002Fblob\u002Fmain\u002Fagent\u002Fskills\u002Fis-this-photo-real\u002FSKILL.md",true,[46,52],{"path":47,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":50,"sourceHtmlUrl":51},"reference\u002Ftool-catalogue.md","REFERENCE","参考文档",10989,"https:\u002F\u002Fgithub.com\u002FReihantt6\u002Fsns-myagent\u002Fblob\u002Fmain\u002Fagent\u002Fskills\u002Fis-this-photo-real\u002Freference\u002Ftool-catalogue.md",{"path":53,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":54,"sourceHtmlUrl":55},"reference\u002Fverification-checklist.md",8922,"https:\u002F\u002Fgithub.com\u002FReihantt6\u002Fsns-myagent\u002Fblob\u002Fmain\u002Fagent\u002Fskills\u002Fis-this-photo-real\u002Freference\u002Fverification-checklist.md",[57,64,71,78,85,92],{"skillNo":58,"name":59,"nameZh":6,"description":60,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":61,"categories":62,"keywords":63,"publishedAt":19},"01M77XDCYZ407XYMF2T4P6F7VT","youtube-transcript","Extract transcripts from YouTube videos. Use when the user asks for a transcript, subtitles, or captions of a YouTube video and provides a YouTube URL (youtube.com\u002Fwatch?v=, youtu.be\u002F, or similar). Supports output with or without timestamps.","https:\u002F\u002Fgithub.com\u002FReihantt6\u002Fsns-myagent\u002Fblob\u002Fmain\u002Fagent\u002Fskills\u002Fyoutube-transcript\u002FSKILL.md",[],[],{"skillNo":65,"name":66,"nameZh":6,"description":67,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":68,"categories":69,"keywords":70,"publishedAt":19},"01M77XD7PBC98J6MXGJJ2TJ8DX","youtube-thumbnail-design","YouTube thumbnail design with specific dimensions, contrast rules, and mobile preview optimization. Covers safe zones, text placement, face expression psychology, and A\u002FB testing. Use for: YouTube thumbnails, video cover images, click-through optimization. Triggers: youtube thumbnail, thumbnail design, video thumbnail, click through rate, ctr optimization, youtube cover, video cover image, thumbnail maker, thumbnail tips, youtube design, video preview image","https:\u002F\u002Fgithub.com\u002FReihantt6\u002Fsns-myagent\u002Fblob\u002Fmain\u002Fagent\u002Fskills\u002Fyoutube-thumbnail-design\u002FSKILL.md",[],[],{"skillNo":72,"name":73,"nameZh":6,"description":74,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":75,"categories":76,"keywords":77,"publishedAt":19},"01M77XD2541C8HXJ6RHYZP4NGJ","youtube-short-maker","Make YouTube Shorts that hold attention with Pexo — vertical 9:16, fast-paced, hook in the first 3 seconds. Describe your topic and Pexo writes the script, generates the shots, picks the models, and assembles a finished Short with music and captions. Use for YouTube Shorts: \"youtube short\", \"make a youtube short\", \"shorts video\", \"short-form video\". NOT for long-form or landscape video.","https:\u002F\u002Fgithub.com\u002FReihantt6\u002Fsns-myagent\u002Fblob\u002Fmain\u002Fagent\u002Fskills\u002Fyoutube-short-maker\u002FSKILL.md",[],[],{"skillNo":79,"name":80,"nameZh":6,"description":81,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":82,"categories":83,"keywords":84,"publishedAt":19},"01M77XCXKVSYK7KTZ3PQBVVGFD","xcode-project-setup","Safely modifies Xcode projects (.pbxproj) to add Swift Packages and link files. Use this skill whenever an iOS project needs dependencies installed (e.g. Firebase, Alamofire).","https:\u002F\u002Fgithub.com\u002FReihantt6\u002Fsns-myagent\u002Fblob\u002Fmain\u002Fagent\u002Fskills\u002Fxcode-project-setup\u002FSKILL.md",[],[],{"skillNo":86,"name":87,"nameZh":6,"description":88,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":89,"categories":90,"keywords":91,"publishedAt":19},"01M77XCR1DV8ZQK2NHWKVY4EWX","x402","Set up Browser Use Cloud payments with x402 — pay per request from a crypto wallet (USDC on Base mainnet), no signup or API key. Two setups it works out up front — \"just use it\" (set up a wallet so you or Claude Code can run cloud browser tasks paid from the wallet — Claude writes and runs throwaway scripts, nothing touches your codebase) or \"build it in\" (install the SDK and write the key + code into your project). Walks through wallet setup, funding, .env, and a ~$1 test run. Use when the user asks about x402, pay-per-use, USDC payments, or wants Browser Use Cloud without an API key. For the free-tier signup (reverse-CAPTCHA → API key), use `browser-use cloud signup` or the `cloud` skill instead.","https:\u002F\u002Fgithub.com\u002FReihantt6\u002Fsns-myagent\u002Fblob\u002Fmain\u002Fagent\u002Fskills\u002Fx402\u002FSKILL.md",[],[],{"skillNo":93,"name":94,"nameZh":6,"description":95,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":96,"categories":97,"keywords":98,"publishedAt":19},"01M77XCMD3R3ZBZQQSW6E0793B","x-twitter-scraper","Build GitHub Copilot workflows with Xquik X API SDKs, REST endpoints, hosted Apify Actor runs, MCP tools, TweetClaw OpenClaw plugin installs, signed webhooks, tweet search, user lookup, follower exports, media actions, and agent automation.","https:\u002F\u002Fgithub.com\u002FReihantt6\u002Fsns-myagent\u002Fblob\u002Fmain\u002Fagent\u002Fskills\u002Fx-twitter-scraper\u002FSKILL.md",[],[]]