~

I felt like we could all use some more capybaras in our life. 💖

Outlive Your Legend

STEP 1: Take a screencap of the video, preferably while it's playing so that the PLAY button isn't blocking the image.

STEP 2: Reverse image search of choice. To my chagrin, I tend to end up resorting to Google's Search by image because Tineye keeps failing me and I haven't gotten around to doing a deep dive into currently available reverse image search services.

If you use the Firefox desktop web browser (untested: other desktop web browsers, Firefox mobile web browser) you can combine STEP 1 & STEP 2: Feel free to repost the video to Tumblr first, preferably as a draft or private post to prevent people from reblogging the uncredited vid, because this method doesn't work as well on Reddit videos.

Right-Click on the video. In the right-click menu, look for "Image Search Options" and hover over it to bring up a list of search options. I'll have to try out some of the other options later, but for this demonstration I used Google.

The search engine should open in another tab.

STEP 3: The Search. This part will be a mix of luck, tenaciousness, and deductive skills.

For Google, I recommend going to the "Exact matches" tab because it presents matches and their information in a much more condensed format than the other tabs. "Exact matches" is a misnomer, as is apparent in the screencap below. This search actually pulled up images from at least two separate videos by the same woman hanging from the same tree: one where she's wearing a slate grey pencil skirt and blazer (as in the video reposted by OP) and one where she's wearing a pale grey long-sleeved dress.

Ideally, a functional search engine would let you sort results by useful variables such as "date posted". Google is not such a search engine, and with its progressive enshittification you may or may not be able to bully Search by image into limiting results to certain date ranges. Luckily, these videos have been less flagrantly reposted than some, so I didn't have to scroll through too many.

When searching for a source, one of the primary things to pay attention to is date. Obviously, the original source has to be older than any of the reports. Unfortunately, not all the Google results have dates, so the earliest dated video isn't guaranteed to be the original.

In this case, the earliest result was of the grey dress video variant posted on Instagram on Jan 17, 2026. However, upon checking the link it became apparently that the Instagram account was not the originator of the video and their repost was unsourced.

If the date check fails you, keep an eye out for any undated results that clearly stand out from the rest. In this case, I picked up on the following:

Which has the following unique traits:

A new outfit! This is the only result that features the woman in a pale brown blazer and pencil skirt.

Dimensions. This result is 1080 x 1920, while the majority of results were smaller in size. Reposts are more likely to be downscaled from the original than upscaled, so if the initial date-check fails you, you're better off checking the largest undated results first.

Specific username (Liang Li (@liangli521)). A lot of the other results just have captions or titles in various languages. Depending on the website, these are usually either captioned reposts or random search terms that don't even link to the actual video. Results with specific usernames are rarer and thus worth checking out, though some of them may turn out to be reposters.

In this case, I hit the jackpot! There they all were: OP's video, the grey dress video, the brown blazer & skirt video, and many, many more!

Legitimately, Liang Li's whole thing seems to be going ham in skirts, dresses, and frequently heels. Possibly for advertising purposes? Very beautiful. Very powerful.

Bonus videos:

Cool lights!

Obstacle courses and playgrounds with friend!

H-hot damn, ma'am...

i am not particularly interested in reading fanfiction generated by an LLM. this is because i am in fandom to read fic that engages interestingly with a canon i love, and an LLM definitionally cannot do this. there are also human-written fic that do not do this! this is not a unique quality of LLM-generated text. "there are fics not to my taste" was a problem long before LLMs entered the picture.

what is new is the scale. the quantity of text an LLM can produce will always outstrip what humans can write by orders of magnitude. maybe this is a problem. readers don't have infinite time -- what's the sustainability of backbuttoning from fics i don't find engaging for one reason or another, if the scale is so dramatically tipped to one side? i genuinely, deeply adore discovering fics with a unique voice or an unexpected perspective, but how much harder is that going to be to find as the volume of fics to sift through grows ever larger?

on the other hand: my preferences are not everyone's preferences. people clearly do read LLM-generated text, and like it.

i've generally found the morality dimension to this discourse somewhat unproductive because my question is always: how does an action materially lead to changes you want to see? and i don't think it's helpful to say, if you use a tool that is widely available then you are a bad person. famously this is not a tactic that makes people eagerly change their ways.

(i am always thinking about daryl davis, a Black man who deradicalized ku klux klan members by actively choosing to have conversations with them. i am thinking about how a child of the founder of stormfront disavowed the movement because of extensive discussion over shabbat dinners. people find it hard to be wrong. to change someone's mind you have to put in the work; you need to approach with empathy.)

i'm equally uninterested in saying that the problem with LLM-generated text is that it is bad. many fics are bad with no LLM involvement whatsover. probably the LLMs will get better, since a lot of money is being invested into the effort. but if you frame the conversation as, LLM-generated text is bad, and therefore liking it means you have bad taste, or you are a fool, or, or. again, what is the end goal here? what happens is that people get defensive. people say, i liked this thing i read, and therefore it is good, and therefore it can't be machine-produced. reading is an active act between the reader and the text -- people aren't wrong for bringing themselves to that conversation, regardless of the source. we can't make not recognizing LLM usage a moral failing.

on a third hand: claude has a voice. gemini has a voice. chatgpt has a voice. much as human writers have tics or an oft-used turn of phrase, these tools do, too. i say this not because i am seeking to identify specific texts as LLM-generated, but because i care about the effects in aggregate.

fandom is not the first space i wrote in, but it's one where i've written consistently, often badly, and over time developed a voice i feel confident about. i really think that's valuable and worth preserving. but the increasing volume of LLM-generated text makes this harder for someone who is just starting to write today. how do you develop your own voice when everything around you sounds the same? the first step to learning something is mimicry, but to build on that you need diversity.

i think i found my new favorite artist on twitter

But brushing your teeth for 30 seconds is better than not brushing them at all when that 2 minutes seems exhausting.  Doing ten minutes of yoga is better than 10 minutes of sitting when 30 minutes of cardio sounds impossible.  Changing my clothes is good when a whole shower is impossible.  Standing on the porch for a few minutes is worth it after being in the house for three straight days because I don’t have the energy to go anywhere.