The Underground Guide to Making AI-Generated Viral Videos

If you’re an everyday social media person, then you will have stumbled upon movies like this:

Or this:

Or this

Because of the U.S. commerce conflict, viral movies depicting chubby People toiling in fictional sweatshops have exploded throughout TikTok and X, racking up hundreds of thousands of views. These clips present exhausted employees stitching garments in dismal manufacturing facility situations—a satirical jab at Trump’s promise to deliver manufacturing jobs again to America by way of hefty tariffs on Chinese language items.

One 32-second video created by TikTok person Ben Lau has been considered hundreds of thousands of occasions, exhibiting AI-generated People in dire manufacturing facility situations, full with conventional Chinese language music and ending with a snarky “Make America Nice Once more” slogan.

AI has turn out to be the brand new political cartoon, a robust software for political protest, and savvy customers are developing with artistic methods to make use of their favourite fashions as means to unfold a message.

Ever puzzled how they do it? It’s really fairly simple. All you want is a robust sufficient PC—or be prepared to spend a couple of bucks/euros/pesos/kilos/yuan—to deliver your concepts to life. When you don’t wish to spend cash although, you higher have a variety of time to spare—we’re speaking hours for simply a few generations.

Right here’s find out how to make a video of anybody you’ll be able to consider doing absolutely anything:

The best way to make your personal AI video

Each video begins with one preliminary body. Consider it because the DNA that determines your whole creation’s feel and look. Movies are basically rapid-fire picture sequences (usually 24-60 frames per second), and your preliminary body units the visible customary for all the things that follows.

You have obtained two routes right here: snag an present picture, or prepare dinner up your personal. Actual-life images are the very best different, however because you’re unlikely to search out genuine images of Trump hunched over a stitching machine or Xi Jinping executing an ideal roundhouse kick to a peasant, AI picture mills are your finest guess.

In fact, if you wish to generate images of well-known individuals, your finest guess is an uncensored mannequin. Open-source options like Secure Diffusion 3.5, Hidream, or Flux are the very best. Your mileage might differ with some paid options, however usually Ideogram and Reve may match properly, with ChatGPT being probably the most censored mannequin of the bunch.

We selected Freepik Mystic (principally a Flux implementation by Freepik), and prompted “Donald Trump consuming a hamburger in a sweatshop.” Here is what we obtained:

Relying on the mannequin, your outcomes shall be higher or worse—so generate a couple of photos till you’re proud of the outcomes. When you discover your output doesn’t resemble the individual you wish to generate, then it’s secure to say it isn’t educated to acknowledge that face or has been censored. When you use an open-source mode, then you could wish to prepare a LoRA on that particular face, nevertheless it’s an excessive amount of of a problem—higher to strive one other mannequin or decide one other picture.

Protip: I invested in a Freepik subscription. Beginning at $69 per yr, I basically get entry to a one-stop store housing a pack of picture mills, video instruments, sound mills, and extra fairly than subscribing to completely different platforms. Opting at no cost options? Brace your self for critical processing calls for and potential multi-hour waits. For instance, Kling’s free tier would possibly take three hours to generate a mere 10 seconds of footage.

You’ll be able to run some fashions regionally in your PC, however as I discussed above, you will have a reasonably highly effective laptop with a GPU of not less than 12GB of VRAM to have usable outcomes.

Deciding on your video engine: Energy vs. freedom vs. price

There are completely different video mills, every one with its personal traits. The mannequin you select will dramatically influence your closing product’s look, capabilities, and price. These are a few of the finest fashions that can do the job with out too many content material restrictions:

Kling AI affords respectable high quality with considerably fewer content material restrictions, at a fraction of the fee—for instance, it is going to generate a video of Trump (which Google’s Veo2 censors), nevertheless it gained’t generate NSFW or excessive stuff. Its free tier crawls alongside at snail’s tempo however really works. That is seemingly the engine behind many viral Trump manufacturing facility movies circulating on-line.
Wan 2.1 shouldn’t be accessible on Freepik, however seems on platforms like Fal.ai, charging roughly $1 per 4-second clip. It is open-source, can run regionally with enough {hardware}, and operates with out content material restrictions—actually something goes.
Luma Dream Machine provides you with fast, reasonably priced outcomes that lean barely towards the fantastical. That is good for meme-worthy political concepts that do not require photorealistic high quality—like turning Trump right into a flying superhero, or making him eat a grossly exaggerated hamburger.

These different options are closely censored. You gained’t discover them helpful on your political memes, however may very well be value making an attempt if you wish to generate different issues:

Google Veo2 delivers superb realism, however enforces strict content material insurance policies—no Trump or superstar movies allowed. It is also a useful resource hog, generally demanding 3-4 occasions the computing credit of competing choices in cloud-based suppliers like Freepik.
Runway’s new V4 mannequin can be fairly good at realism with cinematic high quality, however shares Google’s restrictions. It is proprietary and resource-intensive, although barely inexpensive than Veo2.

For many political meme-makers, Kling 1.6 Professional strikes the candy spot on Freepik, whereas Wan affords most artistic freedom elsewhere. If you wish to see a comparability, on this overview, we generated the identical scene with all these fashions so you’ll be able to examine the outcomes.

Crafting your political meme with image-to-video

Along with your preliminary body (the picture you generated to start your video) and video engine chosen, you are able to deliver your political commentary to life. Whereas interface particulars differ between platforms, the core workflow stays constant throughout instruments.

Step 1: Find the beginning picture (or preliminary body) add subject—this important first body dictates your whole video’s aesthetic and high quality. Since every new body builds upon the earlier one, your preliminary picture successfully blueprints your entire manufacturing. Add your picture into the corresponding subject.

Step 2: Craft a exact scene description within the immediate subject. Preserve it centered on a single second fairly than a fancy narrative. These engines excel at creating cohesive snippets fairly than storytelling arcs.

The fields will differ relying on what you select, however the interface is considerably related for all fashions. Since we’re utilizing Freepik, it is going to appear to be this:

Right here is an instance with one other platform, Fal AI:

When working in your immediate, visualize and describe your meme as a single scene. Keep in mind that these are quick movies, so that you could be detailed by way of the scene components, however not likely by way of the scene improvement. Visualize your meme as a single scene.

For instance the immediate for our meme was “Shut-up shot: A person in a swimsuit is consuming a juicy hamburger, with melted cheese oozing out. The digicam gently zooms in to emphasise the small print of the burger and the person’s centered expression, whereas the background reveals employees stitching in a sweatshop.”

Step 3: Configure some bits of technical specs like period and determination—widescreen for desktop viewing, sq. for social feeds, portrait for TikTok or Instagram Tales.

Step 4: Growth—click on generate… then put together to attend since these fashions require vital processing time. Relying on the platform you employ, the movies might take a couple of minutes to yield some 10 seconds of video. Nevertheless different platforms might take as a lot as 2 hours to generate equally lengthy outputs.

Some suppliers allow you to generate movies in parallel. This can be a good concept, as a result of you could not get good leads to your first technology. It is advisable to generate a couple of, so you could have all kinds of options to select from.

When you’re carried out, all it’s a must to do is both share it as-is or add audio components that reinforce your message.

Congratulations—you have simply crafted digital ammunition on your personal aspect within the ongoing, endless world meme conflict.

Right here’s what our closing makes an attempt ended up wanting like: