Open AI has again teased us users with its highly anticipated text to video generator SORA. The company has highlighted again its ground-breaking potential to artificially generate human videos that looks ultra-realistic.
The two videos are made by NiceAunties and David Sheldrick who are both professional creators. The former is a Singaporean artist whereas Sheldrick is British Korean artist with graduation from London college of Fashion.
The video made by Nice aunties showcases Sora’s ability to generate realistic looking people of old age with wrinkles and other signs of ageing whereas the video created by Sheldrick shows young attractive women dancing in glamarous dresses.
Neither the video creators nor Open AI has released the text prompt that was used to create the videos depicting complex facial expressions, skin textures and even the artifacts.
While I may not be able to give my opinion on the art’s depiction but I can certainly say that the video got me emotionally engaged and I wouldn’t have differentiated the people from a real human being if the video would have been shown to me randomly by someone.
These two videos released by Open AI did not had any glitches which we have normally seen over the months from other text to video generators such as hands of woman holding a cell phone which appeared damaged.
However, at some instances the human face did looked noticably different (below image) but certainly not unrealistic that it would disrupt the engagement of the viewer.
Its not clear whether SORA will have an in-built mechanism that portrays slight unrealism of the humans as a way to tell that the video is artificially generated. It could be a way to mitigate the side effects of Generative AI such as deep fakes or voice replication.
Postponing the release of Sora until after the U.S. elections?
Sora was initially released on February 15, 2024 where many clips generated by it were shown on OpenAI’s website. At that time, the company announced that it is making the product available to “red teamers to assess critical areas for harms or risks”. The company has since remained tight-lipped about its debut.
As much as SORA’s ability is breathtaking and inspiring, the rise of deepfakes have shown the level of damage that can be done if tools like SORA or other text to video generators are used by the bad actors.
Fake videos with misleading information can be utilized by opponents in political scenario. The US 2024 presendential elections will be conducted on 5th November. Can it therefore be assumed that Open AI is holding the release of SORA till the presedential elections.