We all love to share GIFs – and there are many ways to do this, through portals or online portals – but often, because there is so much content, you end up showing a low fidelity. GIF.
There may be a lot of copies of the same video clips as a GIF, or maybe it's just hard to capture and download, but Gfycat hopes that it can be solved at the technical level. Gfycat is now making a great technical effort to make these GIFs more beautiful and more visible as creators seek to continue to download content, regardless of the type of quality or fidelity they are. And it's more a video problem than a problem of image recognition, said CEO Richard Rabbat.
"We have scaled [through] the creators by word of mouth, and they are just excited by the content of Gfycat and [creating]," said Rabbat. "In many cases, what we are developing from the point of view of AI and machine learning are additional tools to support their enthusiasm. We want to allow them to generate more virality for their content, and in this case, make their content even more easily detectable. It is something that is very important to us because we continue to focus on the creators. "
Rabbat said that Gfycat will be browsing the web to find the original version of a video of where the GIF comes from – in some cases it comes from YouTube – and analyze this video to find out which part comes from the GIF. The company then produces a superior GIF and exchange, making the wider version of the GIF a better quality version. The company is creating a kind of template for every GIF image, and then trying to match that with better quality videos, he said.
"What we noticed was that a number of users who downloaded GIFs were incredibly popular, but when they downloaded most of the time, they were really bad quality, "said Rabbat. "We have been studying AI and machine learning for some time, with respect to [to] our initiative to beautify the web with regard to GIFs."
After that, if a creator downloads a GIF that includes a celebrity, it may be that he does not mark that as having that celebrity. The company has therefore performed an internal analysis to identify the celebrity in this GIF file and identify it automatically. The hope is that while the company has an existing popular celebrity library, it will be able to identify celebrities with these tools and automatically start marking them when they come.
Rabbat stated that Gfycat built these two tools internally because the products available on the market did not work well with GIFs. Although GIFs are, of course, a series of images, he often said that many different elements (like many celebrities) would appear in sequence while standard image recognition technology could only identify one or two of them. Technology is video-based, he said.
"One of the big challenges is the sheer amount of information that a GIF includes," Rabbat said. "It's hundreds of frames, sometimes more.We have to identify at a very high rate these different celebrities that are created.We wanted to do it in real time.We were able to do it in a minute of people creating content, we were able to identify celebrity. "
Finally, with all these tools, Gfycat wants to identify the text in the various GIF legends as they come into play. Again, part of the challenge here was that a GIF could enter with a caption, but the text is grainy and not easily readable or identifiable. Gfycat has sought to build internal tools that help to understand what legends say, and then make GIFs easier to discover based on these legends.
While Gfycat is certainly not the only one trying to make it easier to discover short video content like GIFs – there are companies like Tenor and Giphy looking to build robust platforms – it is trying to address the problem with technical tools. And with more than 130 million active users per month (in comparison, Giphy has 300 million active users per day), this will become a technical problem because this type of content can not be organized at large ladder.