Let It Flow: AI Researchers Create Looping Videos From Still Images

Originally published at: AI Researchers Create Looping Videos From Still Images

Researchers from University of Washington and Facebook used deep learning to convert still images into realistic animated looping videos. Their approach, which will be presented at the upcoming Conference on Computer Vision and Pattern Recognition (CVPR), imitates continuous fluid motion — such as flowing water, smoke and clouds — to turn still images into short…

You do not need AI to loop still images. There is a website that ha been doing this for 5 about years.


Your talking about multiple images. This article explains the use of a single image - get your head around that.


1 Like

Sorry you did not understand my post. I was (and am) talking about a single image. Single image = one photo. Take one photo and animate it. By the way, it’s not “your”, it’s “you’re”.

RE:English grammar
I am replying to michaelvandeelen’s comment, but also expanding on it for completeness, fairness, and entertainment value.

I agree that squeakycoder, in order to comply with grammatical convention of today’s English, needs to either use the contracted form of you are, or edit to follow “noun phrase” syntax. This is where complexity resides, since a noun phrase allows the possessive pronoun’s use as the headword of a noun phrase, as long as it is understood as a noun phrase (catch-22) or is unambiguous as such.

Here are some examples:
Your strategizing nonstop is difficult to watch, with all that pacing and paroxysms of profanity, but the results speak for themselves.
You’re at it again, your talking about god follows the same pattern as always.
Your talking about her betrays your misogyny, so your attempts to portray yourself as noble and her as evil only makes people take her side.

Squeakycoder omitted the 2nd part, which likely excludes it from the small percentage of grammatically correct copy, but I found it easy to understand the meaning he was communicating, and I did not detect any underlying motive other than helping you understand what his research is about.

If I apply the same analysis to your comments, right away I detected a thinly disguised snarkyness, or maybe professional jealousy if applicable. In any case, an undeniable negative sentiment drives your comments. Now let’s discuss grammar. I would first ask you to clarify what an about year is, as Google search engine was unable to help me understand this unit of metrology. Correct use of grammar would strongly suggest combining your two sentences into a compound sentence, resulting in something like this: Since myimaginarywebsite.bs has successfully harnessed the power of mechanical turks to output so-so animated still images since circa 2016, deep learning AI is obviously superfluous use of Nvidia’s amazing technology.

You did not specify which website, contrast their results with the images presented here by the UW team (who made me feel a hint of pride for the UW alumnin association card in my wallet).

And finally, in the ad hominem tradition, you neglected the standard politeness of any forum contributor, which frankly, when scientists and engineers present their work, you ought to treat them with the utmost respect. The work of scientists, inventors, researchers, and engineers is the business of creating knowledge, which humanity benefits from en masse. This is also a moment when many researchers feel shy, since they are opening the door to original material they identify with personally and emotionally, and the same door allows criticism of all variety to enter and attack.

1 Like

Regardless of all the rest, if there is really an algorithm that can create animated images with as convincing visuals as this algorithm, I would love to see it. I have seen quite a few previous methods use multiple images, and find nice ways to blend them at the in/out points to create a nice looping feel, but never have I seen starting with a single image, and create a realistic animation. I have seen some cheesy ones that are adding fake ripples in the image, etc. Really just putting a perspective sin wave displacement, not anything you would ever have a real use for…

This I could see possible uses for, but nothing else I have seen before I could same the same. That didn’t stop people from adding really cheesy animation to their online social media post, etc, but overall nothing that I would say directly compares. But I totally have not seen everything online so perhaps I am oblivious to this other technique… Can you post a webpage, or a paper? Again I am honestly interested if such an algorithm exists?

Is the model and code for this open source??

Any update or links to this ?