logo
#

Latest news with #SiweiLyu

Fact Check: AI-generated video shared as footage of wildfires in Israel
Fact Check: AI-generated video shared as footage of wildfires in Israel

Reuters

time08-05-2025

  • Science
  • Reuters

Fact Check: AI-generated video shared as footage of wildfires in Israel

A compilation of videos generated using artificial intelligence was shared online as authentic footage of the wildfires that have hit Israel. Wildfires broke out on the outskirts of Jerusalem on April 30 and forced evacuations of three communities and the closure of a main highway. A video compilation showing massive fires was posted online, opens new tab on the same day with the caption, 'Israel declares a state of emergency as the fire engulfs it.' However, the video compilation was initially shared in a January 9 collaborative post, opens new tab on Instagram by two accounts, opens new tab that post AI-generated content, opens new tab. The caption said, 'AI-generated representation of fires that can cause global warming'. Neither of the Instagram accounts responded to requests for comments. Reuters previously addressed the same video compilation in January when it was shared as authentic footage of Los Angeles wildfires. Three digital experts told Reuters at the time that the compilation was AI-generated. At the time, Siwei Lyu, a professor of computer science and engineering at the State University of New York at Buffalo and James O'Brien, a professor of computer science at the University of California, Berkeley, told Reuters via email that the video was generated using AI. O'Brien pointed out that the video shows unrealistic-looking fire and smoke that look more like movie special effects. J. Rosenbaum, an AI researcher at RMIT University in Melbourne, also said the video was made using AI, highlighting signs such as warped power lines, flames merging into palm trees and inconsistent smoke clouds. False. The video compilation does not show authentic footage of wildfires in Israel, and three experts said it is AI-generated. This article was produced by the Reuters Fact Check team. Read more about our fact-checking work.

Footage bearing AI signs falsely presented as Myanmar quake aftermath
Footage bearing AI signs falsely presented as Myanmar quake aftermath

AFP

time02-04-2025

  • AFP

Footage bearing AI signs falsely presented as Myanmar quake aftermath

"Oh! Mandalay, it's like the city has fallen," reads the Burmese-language caption to the clip shared on Facebook one day after the quake. It has more than 11,000 views and depicts an aerial view of a devastated city with pagodas visible on the skyline. Text superimposed on the clip also says it shows the destruction in Mandalay. Image Screenshot of the false Facebook post taken on April 1, 2025 The video also surfaced in similar posts written in Chinese, English, Indonesian, Spanish, Tamil, Thai and Urdu as reports emerged about the damage in the city from the shallow 7.7-magnitude earthquake. The head of the ruling junta, Min Aung Hlaing, said on April 1 that 2,719 people were confirmed dead so far, with more than 4,500 injured and 441 still missing (archived link). The toll is expected to rise significantly as rescuers reach towns and villages where communications have been cut off by the quake. Hundreds of kilometres away, Bangkok authorities said the death toll there had risen to 20, the vast majority killed when a 30-storey skyscraper under construction in the Thai capital collapsed. Comments to the posts indicate social media users believed the video shows genuine visuals following the quake. "Oh! Mandalay, my heart is broken," one wrote. Another said: "There's hardly anything left in the entire town." 'Made with Wan AI' A reverse image search on Google using the video's keyframes found a nearly identical video on X on March 29. Although its caption does not directly indicate the clip was made with AI, the logo of the generative tool Wan AI is visible at the bottom right corner of the higher quality footage (archived link). Image Wan AI logo at the bottom right corner of the video highlighted by AFP Visual errors also suggest the video was generated using an AI tool, according to experts. Shu Hu, director of Purdue University's Machine Learning and Media Forensics Lab, told AFP that trees in the clip lack visible roots while a person walks in a way not consistent with natural human motion (archived link). Siwei Lyu, director of the University at Buffalo's Media Forensics Lab, separately pointed out the Wan AI watermark, and that the shadows of objects in the clip are inconsistent in direction and size (archived link). Image Visual inconsistencies in the video highlighted by Shu Hu (L) and Siwei Lyu While there is no foolproof method to spot AI-generated media, identifying watermarks and visual defects can help, as errors still occur despite the meteoric progress in generative AI. AFP has previously debunked misinformation related to the earthquake in Myanmar here.

False posts about 'Indonesian pyramid' share AI visuals
False posts about 'Indonesian pyramid' share AI visuals

AFP

time28-03-2025

  • Science
  • AFP

False posts about 'Indonesian pyramid' share AI visuals

"Gunung Padang: The World's Oldest Pyramid Hidden Beneath the Earth," reads a March 4, 2025 Facebook post from a Thailand-based user that has since been shared over 2,000 times. It features an image appearing to show an amphitheatre below a hilly and forested area. "What if the oldest pyramid in the world wasn't in Egypt but in Indonesia? Gunung Padang, a mysterious site in West Java, may be just that," part of the caption adds. Image Screenshot of the false Facebook post taken on March 26, 2025 Similar posts written Chinese, Thai and Indonesian repeat the claim and include images purportedly showing the same site from slightly different angles. Stone terraces Multiple scholars told AFP it is inaccurate to characterise Gunung Padang as a pyramid. Lutfi Yondri, Padjadjaran who has done several excavations at Gunung Padang since 1997, told AFP on March 20 that the only structure that remains at the site is a "complex of stone terraces" (archived link). Carbon dating shows the terraces were built around 117 to 45 BC, and are on top of rock columns formed through a natural geological process, he said. Truman Simanjuntak from the Center for Prehistory and Austronesian Studies separately said Gunung Padang "has absolutely nothing to do" with pyramids (archived link). "Claiming there are man-made chambers inside the hills is a hallucination," he said on March 20. "Let's think rationally, and talk data." A 2023 study in the Archaeological Prospection journal that described Gunung Padang as a "prehistoric pyramid" has been retracted after experts in geophysics, archaeology and radiocarbon dating raised accuracy concerns (archived link). 'Physically implausible' Moreover, the images circulating online do not resemble the actual site as seen in an AFP photo taken July 2011. Image Siwei Lyu, director of the University at Buffalo's Media Forensics Lab in the United States, said an analysis of the images showed a "close to 100 percent possibility" that they were AI-generated (archived link). He noted that one of the images violates perspective geometry, where four lines that should be parallel in the real world do not intersect. "This is a telltale sign of its AI-generative nature," Lyu said. Image Image in the false post, with visual inconsistencies highlighted by Siwei Lyu Shu Hu, director of Purdue University's Machine Learning and Media Forensics (M2) Lab, also said the shadows in the images are "physically implausible in a real scenario" as the lights point at different directions (archived link). Image Images in the false posts, with visual discrepancies in the shadows highlighted by Shu Hu There is no foolproof method to spot AI-generated media but identifying visual inconsistencies can help, as errors still occur despite the meteoric progress in generative AI.

Aerial image of where 'Manaus meets the Amazon' is AI generated
Aerial image of where 'Manaus meets the Amazon' is AI generated

Yahoo

time18-03-2025

  • Science
  • Yahoo

Aerial image of where 'Manaus meets the Amazon' is AI generated

"Clear division between the city and the forest in Manaus, Brazil," reads a Thai-language Facebook post shared on March 14, 2025. Manaus, in the heart of the Amazon rainforest, is the capital and largest city of the Brazilian state of Amazonas (archived link). An image attached to the post, which was shared more than 4,800 times, purports to show an aerial view of where the city meets the rainforest. The same image was also shared alongside similar claims elsewhere on Facebook in Thai, as well as other languages. "Their country is so beautiful and well-organised," read a comment on one of the posts. Another said: "The condominium with the forest view must be so expensive because there won't be any new buildings blocking them." The image, however, was generated by artificial intelligence. A reverse image search on Google led to several other posts that used the picture, but the website's "About this image" feature revealed a label indicating it was made using Google's AI tools. A Google spokesperson told AFP that a SynthID watermark had been detected in the circulating image, "meaning the image has been generated or modified with AI". SynthID, which was launched by Google's DeepMind AI lab in 2023, watermarks and identifies images generated with Google AI (archived here and here). "This is exactly the sort of problem SynthID was intended to address, so we're glad to see it being put to good use here," the spokesperson said in a January 28 email. Experts in the field also told AFP the image bore signs of being generated by AI. Shu Hu, director of Purdue University's Machine Learning and Media Forensics Lab, said vehicles in the image are "uniformly white", an anomaly that is "highly improbable in real-world scenarios" (archived link). He also identified an "absence of shadows under the vehicles" and "shadows cast by buildings (that) lack clearly defined shapes". Siwei Lyu, director of the University at Buffalo's Media Forensic Lab (UB MDFL) pointed out to other inconsistencies, including misshapen windows on several buildings in the image (archived link). AFP photos from Manaus are similar to the falsely shared image, but show different buildings separating the built-up environment from the rainforest. Google Earth's satellite and street view imagery also shows how buildings at the boundary of the city differ from the high-rises depicted in the falsely shared image. AFP has previously debunked other AI-generated images that used Google AI.

Aerial image of where 'Manaus meets the Amazon' is AI generated
Aerial image of where 'Manaus meets the Amazon' is AI generated

AFP

time18-03-2025

  • Science
  • AFP

Aerial image of where 'Manaus meets the Amazon' is AI generated

An AI-generated image of a straight divide between a city and a verdant forest has been shared thousands of times around the globe in posts that falsely claimed it was a photo of where the Brazilian city of Manaus meets the Amazon rainforest. While the image resembles genuine photos of Manaus in the heart of the Amazon, the picture was in fact generated using Google's AI tools. "Clear division between the city and the forest in Manaus, Brazil," reads a Thai-language Facebook post shared on March 14, 2025. Manaus, in the heart of the Amazon rainforest, is the capital and largest city of the Brazilian state of Amazonas (archived link). An image attached to the post, which was shared more than 4,800 times, purports to show an aerial view of where the city meets the rainforest. Image Screenshot of the false Facebook post, captured on March 17, 2025 in Thai, as well as other languages. The image, however, was generated by artificial intelligence. 'Made with Google AI' A reverse image search on Google led to several other posts that used the picture, but the website's "About this image" feature revealed a label indicating it was made using Google's AI tools. Image Screenshot from Google Images, with the AI label highlighted by AFP A Google spokesperson told AFP that a SynthID watermark had been detected in the circulating image, "meaning the image has been generated or modified with AI". SynthID, which was launched by Google's DeepMind AI lab in 2023, watermarks and identifies images generated with Google AI (archived here and here). "This is exactly the sort of problem SynthID was intended to address, so we're glad to see it being put to good use here," the spokesperson said in a January 28 email. Experts in the field also told AFP the image bore signs of being generated by AI. Shu Hu, director of Purdue University's Machine Learning and Media Forensics Lab, said vehicles in the image are "uniformly white", an anomaly that is "highly improbable in real-world scenarios" (archived link). He also identified an "absence of shadows under the vehicles" and "shadows cast by buildings (that) lack clearly defined shapes". Siwei Lyu, director of the University at Buffalo's Media Forensic Lab (UB MDFL) pointed out to other inconsistencies, including misshapen windows on several buildings in the image (archived link). Image Screenshot of the falsely shared image, with visual elements indicative of an AI-generated image highlighted by AFP Authentic Manaus photos AFP photos from Manaus are similar to the falsely shared image, but show different buildings separating the built-up environment from the rainforest. Image Screenshot comparison between the AI-generated image (left) and an aerial photo taken by AFP (right) Google Earth's satellite and street view imagery also shows how buildings at the boundary of the city differ from the high-rises depicted in the falsely shared image. Image Screenshot of the Google Earth imagery, taken March 18, 2025 AFP has previously debunked other AI-generated images that used Google AI.

DOWNLOAD THE APP

Get Started Now: Download the App

Ready to dive into the world of global news and events? Download our app today from your preferred app store and start exploring.
app-storeplay-store