IRL vs AI video: Eliza and Lil Auntie's full playbook

We got Eliza and Lil Auntie on a call for our last workshop, two of the biggest video creators on crypto Twitter, on camera together, walking through exactly how they make their content. Eliza covered the AI side, the videos where she snaps her fingers and turns into someone else. Auntie covered the IRL side, the talking-to-camera style and how she got comfortable doing it. We ran for nearly two hours and it was packed with stuff you can actually use. Here is everything worth keeping.

Eliza's AI video workflow is simpler than people think

The first thing Eliza said is that people gatekeep this stuff way more than they need to. Her real process is short.

It starts with an idea, not a tool. She never opens an app and starts generating. She already has the scene in her head, then she builds toward it. As she put it, you have to have the scenario ready first, you cannot just start making an image with nothing in mind.

From there:

• She uses ChatGPT to make the image for each scene. She drops in her own photo, or anyone's, and dresses the character however she wants. Find an outfit on Google, tell it to put her in this outfit, and you can turn yourself into a warrior, a Hollywood actress, whatever the scene needs.
• Then she moves the image into Super Grok for the video. If you pay for X Premium you already have Super Grok, so it costs nothing extra. She says the consistency is the best part, one image gives you clip after clip that actually match.
• She edits in CapCut and adds sound effects and background music to make it interesting.

Her advice on prompting goes against what most people do. Keep prompts short and simple, four or five words per scene, something like "make this girl walk towards camera." Do not ask ChatGPT to write fancy prompts for you, she says those never work. Add the things you do not want changed, like keeping the face structure intact, and add "ultra realistic" or "cinematic lighting" if that is the look you want.

The finger-snap transition trick

This was the part everyone wanted, the trick where she snaps and transforms into another person. Here is how she does it, with nothing held back.

• Record yourself in your room doing a movement and the snap. Keep filming for another four or five seconds after the snap, because you need that motion to copy later.
• Grab a picture of the person you want to become.
• Take a picture of the actual room you filmed in. This is the part nobody talks about. You use it to keep the background consistent.
• Remove the background from the person's photo and place them into your room.
• Give it to Kling AI using motion control, so it copies your exact movement.
• Drop both clips into CapCut and cut on the snap.

She uses Kling AI because it is the cheapest she has found, around six or seven pounds for 600 credits. She tested the expensive platforms and says you can get similar quality for a fraction of the cost.

Auntie on getting comfortable on camera

Someone asked how to push past the cringe of putting your face online for the first time, and Auntie's answer was the most honest part of the call.

Her first video was over two years ago, in her backyard, holding a cool cat toy, wanting to hide behind filters because she was not confident about her look. What changed everything was self-acceptance. When you are talking, people are not studying your nose or your skin, they are listening to what you say. Once you stop worrying about how you look, content creation gets easy, because that stops being your focus. Expressing the idea in your head becomes the focus instead.

English is her second language and she said that does not matter either. The energy and the personality carry the video, the polish comes second.

The tools they actually use

Between the two of them, here is the real stack:

• ChatGPT for generating the scene images.
• Super Grok for turning images into video, free with X Premium.
• Kling AI for the motion-control transition videos, the cheap option.
• CapCut for editing, sound, and clipping.
• Whisper Flow for voice-to-text. Auntie talks her messy ideas out loud instead of typing them.
• Claude to turn those raw thoughts into a structured post. She even pastes in the open-source X algorithm and asks Claude to shape the post so the algo is more likely to push it.

Keep it under 60 seconds and skip "hey guys"

Both of them landed on the same lesson. Short wins. Auntie said she will film 20 minutes, cut it to two, and it is still too long for her audience. Under 60 seconds performs better because the completion rate is higher, and a higher completion rate means the algo pushes it to more people.

Eliza shared one specific tip from analysing her own videos: cut the "hey guys" opener. Those first words are your one chance before someone scrolls, so do not waste them on a greeting. She stopped saying it entirely and it helped.

Auntie made the same point for IRL videos. Skip the "last week I went to Thailand" warm-up and get to the point. You are not Kim Kardashian, people are not there to watch you talk about your week, they are there for the thing you are showing them.

What actually grows an account right now

A lot of the call turned into a real conversation about growth, and a few things stood out.

Follower count means almost nothing. Eliza pointed out accounts with a million followers that cannot get ten likes. The content is what carries.

Replying still works, but the way you do it matters. Auntie made the case for the silent readers, the 90 percent who never like or comment but do scroll the comment section. A reply that brings a real thought or a real question gets noticed. A "nice post" does not.

Both of them, and Captain too, pushed back on the small-account habit of simping up to big accounts. Agreeing with someone just to get noticed gets your comment skipped. A genuine take, even a disagreement, shows you are actually thinking, and that is the person they follow back. So engage with big accounts as equals, even if their account is a thousand times the size of yours.

There was also a shared frustration with content about content, the endless metrics posts and PNL screenshots. It gets engagement, but it does not build you into anything a brand would want to work with. Auntie's mindset shift was to post things she finds meaningful and useful instead of things that just farm likes. That made her stress-free, and she actually learns from the content she makes.

The algo notes from the session

Chewie, who had gone through the X algorithm in detail, shared a few useful data points:

• Narrative videos get roughly a 10x boost, with a 50 percent bump if people complete them.
• Priority goes to 15 to 30 second videos, and it still works under 2 minutes 20 seconds.
• That 2:20 number is the max video length for non-premium accounts, which means you can get real reach without paying.
• The premium boost itself is small, somewhere around 1.2x. Most of premium's value is psychological, people just pay more attention to premium accounts.
• Non-shilly videos, where you talk naturally instead of showing the product, tend to get discovered more.
• Podcast clips are rising fast, up about 20 percent on TikTok last month. He thinks they are the next thing to get ahead of, and it ties straight back into the clipper economy.

Can AI replace real creators

The last big question was whether AI video creators make real ones replaceable. Auntie's take was the sharpest. AI cannot copy the trust layer. A real creator uses the hard wallet, reviews it because they actually use it, and answers questions in the comments. An AI creator only shows the product, so the trust is gone.

The group landed somewhere reasonable. AI will keep growing and it will affect the industry, so it is worth learning now even if you start with faceless AI videos. But the creators people actually trust get harder to replace, not easier. The more real you are in the timeline, the more that is worth.

Huge thanks to Eliza and Lil Auntie for sharing all of this so openly, and to everyone who showed up and asked great questions.