How Is Yang Mun Made? The AI Tools Behind the 2.5M-Follower AI Monk

Yang Mun looks like an elderly monk dispensing calm life advice from temples, gardens and mountain retreats. His main Instagram account has more than 2.5 million followers.

He is also entirely synthetic.

Yang Mun was created by digital creator Shalev Hani, and the strongest current evidence points to a surprisingly lean production setup. HeyGen is the only tool we can confidently place at the center of the current workflow. Hani says he uses its avatar delivery and voice tools, sometimes uses its script writer, produces videos in batches, and can complete a video in around 20 minutes. HeyGen says the character was created entirely with its platform.

There is one complication.

Other sources claim Yang Mun uses a larger stack involving ChatGPT, Nano Banana, ElevenLabs and additional video tools, while a Eurovision News Spotlight investigation used Google's SynthID detector to identify Google-generated AI media in Yang Mun content.

Those accounts do not fit neatly together.

So rather than give you a tidy five-logo stack that may be wrong, we are separating what is verified, what is plausible and what remains unconfirmed.

The Yang Mun AI stack: what is actually verified?

The distinction matters.

A creator's production workflow can change over time. A source image might come from one model and then be animated through another. A voice tool may have been used early and later replaced. Creator-stack databases also tend to turn partial evidence into clean diagrams because clean diagrams are easier to publish.

Choosely would rather leave a box marked unknown.

HeyGen is the clearest tool behind Yang Mun

HeyGen has published the most detailed first-party account of how Yang Mun is currently produced.

Its case study identifies Shalev Hani as the creator and says Yang Mun was built entirely with HeyGen. Hani specifically names avatar delivery and voice as the features he relies on most. When a script is not already prepared, he also uses HeyGen's script writer.

The production process described by Hani is unusually simple:

1. 1Start with a recurring audience problem or spiritual theme.
2. 2Write a short script built around one idea.
3. 3Use HeyGen to create Yang Mun's avatar performance and voice.
4. 4Produce several videos together in a batch.
5. 5Schedule the finished clips across the week.

Hani says keeping the scripts simple was one of the most important lessons he learned. HeyGen says each finished video can take roughly 20 minutes to produce, while the move to this workflow increased publishing frequency from a few videos per week to daily content.

That is the first useful lesson from the stack.

Yang Mun does not appear to require an elaborate AI production pipeline for every video.

For the basic talking-character format, HeyGen is doing several jobs that creators often split between separate subscriptions.

What HeyGen is actually replacing

It is easy to describe HeyGen as an AI avatar generator and undersell what that means operationally.

A conventional version of Yang Mun would need some combination of:

• a human presenter
• camera and lighting
• recording space
• voice recording
• repeated shoots
• editing
• reshoots when a script changes

The documented Yang Mun workflow compresses much of that into one generation environment.

HeyGen currently provides avatar generation, AI voices, talking-photo and avatar-video systems, text-to-video tools, lip sync and related production features. In Yang Mun's case, Hani specifically highlights the avatar and voice layers rather than complex effects.

That explains the unusually lean workflow.

Yang Mun is not trying to be a cinematic short film. Most videos need one synthetic character, one location, one voice and one clear message.

For that job, adding more tools can easily create more work rather than a better result.

Then why did Google SynthID detect Google AI?

This is where the evidence gets interesting.

In January 2026, Eurovision News Spotlight examined Yang Mun videos using Google's SynthID detector. It reported that tested visuals were identified as AI-generated using Google technology.

That appears to clash with HeyGen's later description of Yang Mun as created entirely with HeyGen.

Not necessarily.

A Google image model could have generated source imagery that was later brought into HeyGen. An earlier Yang Mun workflow may have relied more heavily on Google tools before production consolidated around HeyGen. Different posts may use different pipelines. HeyGen may also be describing the video-production workflow rather than every asset that ever enters it.

The available evidence does not tell us which explanation is correct.

And that is exactly why we would not write:

Yang Mun uses Google Nano Banana.

SynthID tells us Google-generated media was present in the material investigators tested.

It does not identify the exact Google model, prompt, production stage or current workflow.

That is a useful finding. It is not permission to fill in the blanks.

What about ChatGPT, Nano Banana and ElevenLabs?

If you search for the Yang Mun stack, you will find a much neater answer.

VirtualHumans, for example, describes Yang Mun as using tools including ChatGPT, ElevenLabs, HeyGen, Nano Banana and Infinitetalk.

It is a plausible stack.

ChatGPT could handle scripts. Nano Banana could create the character imagery. ElevenLabs could create the voice. HeyGen or Infinitetalk could animate the character.

The problem is that plausible is not the same as verified.

HeyGen's own case study says Hani writes the scripts himself or uses HeyGen's script writer, and names HeyGen's voice system as part of his core workflow. That directly reduces the need for both ChatGPT and ElevenLabs in the documented process.

Meanwhile, the Google provenance finding proves some Google-generated media existed without proving Nano Banana specifically produced it.

So the current Choosely position is:

HeyGen: verified.

Google-generated visual material: verified in tested posts.

ChatGPT, Nano Banana, ElevenLabs and Infinitetalk: possible, but not sufficiently verified as part of Yang Mun's current production stack.

That is less satisfying than a five-tool graphic.

It is also more useful.

Do you actually need five AI tools to make a character like Yang Mun?

Probably not.

The Yang Mun case suggests there are two sensible ways to approach this type of content.

Option 1: Keep the stack minimal

For a straightforward talking AI presenter:

Script → HeyGen → final edit

The script can be written manually or created inside HeyGen. HeyGen then handles the synthetic presenter and voice.

This is the closest match to the current first-party description of the Yang Mun workflow.

It has an obvious advantage: fewer handoffs.

Every time a production moves between tools, something has to remain consistent—the character's appearance, voice, timing, framing or emotional tone. If one platform can handle several of those jobs well enough, complexity falls quickly.

Option 2: Build a more controlled multi-tool stack

Creators wanting more control over the visual environment could split the workflow into specialist layers:

This is closer to the production logic Choosely found when reverse-engineering Promptgenix and Chloe vs History.

The difference is that Yang Mun's format does not demand as many moving pieces.

Promptgenix builds individual images, animates them shot by shot and manually edits the sequence around music. Chloe vs History depends on writing, historical environments, character consistency, voice, sound and vertical editing.

Yang Mun mostly needs the same person to deliver another short message convincingly tomorrow.

Different job. Different stack.

The real advantage is consistency, not generation quality

The most interesting thing about the Yang Mun workflow is not that HeyGen can generate a believable elderly monk.

Plenty of models can create convincing people now.

The operational advantage is that Hani can repeatedly produce the same recognizable character, speaking in the same broad voice and format, without reshooting a human presenter.

That turns character consistency into infrastructure.

A traditional creator's face is naturally consistent because it is their face. A synthetic creator has to recreate that identity every time.

If the production system cannot maintain the character, the illusion falls apart quickly.

That is one reason a purpose-built avatar system makes more sense here than simply asking a general video generator for an elderly monk sitting in a garden giving advice.

The first result might look excellent.

The twentieth needs to look like Yang Mun.

The other lesson: do not make the tool do the creative job

HeyGen's case study repeatedly returns to something Hani learned after building the account: simple scripts work, the audience's recurring problems provide the ideas, and the software should support the message rather than distract from it.

That is not a particularly dramatic secret. It is the useful one.

The tool can generate a face, voice and performance. It cannot decide whether an idea is worth hearing, whether the character has a clear job, or whether the next video earns another minute of attention.

The production stack reduced Hani's cost of consistency. It did not replace positioning, editorial judgment or audience understanding.

What creators should know about disclosure

Yang Mun's current website describes the character as a digital teacher, and Instagram has just renamed its profile-level disclosure from AI creator to AI-generated profile.

Instagram says the label is intended for profiles centered on a synthetic person, not human creators who merely use AI during production. It also says detected AI-generated profiles that remain unlabeled can lose recommendation eligibility, while correctly labeled profiles are not penalized simply for being synthetic.

Choosely has not independently verified whether @yangmunus currently displays the new label. The enforcement and notification rollout is new, so we are not making a compliance claim.

The practical lesson is narrower: if the apparent person at the center of the account does not exist, disclosure is now part of distribution strategy—not just an ethical footnote.

Choosely verdict

HeyGen is the only tool we can confidently place at the center of Yang Mun's current production workflow. It handles the documented avatar, voice and video work, with a built-in script writer available when Hani has not prepared the copy himself.

Google-generated media has also been detected in tested Yang Mun content. That evidence does not identify the exact Google model or prove that Nano Banana is part of the current stack. The wider ChatGPT, ElevenLabs, Nano Banana and Infinitetalk claims remain plausible but unverified.

The most useful lesson is not the brand list.

It is how few tools a consistent synthetic presenter may now require.

For a format built around one recurring face, one voice and one short message, a minimal stack can be an advantage. Fewer handoffs make identity easier to preserve, production easier to repeat and the workflow easier to operate every day.

Copy the discipline, not the ambiguity.

The tools are becoming easy.

Trust is still expensive.

Frequently asked questions

Is Yang Mun real?

No. Yang Mun is an AI-generated character created by digital creator Shalev Hani. He is not a real monk or human teacher.

What AI does Yang Mun use?

The strongest current evidence identifies HeyGen as the core system for Yang Mun's avatar, voice and video production. Hani also uses HeyGen's script writer when needed. Investigators separately detected Google AI provenance in some Yang Mun visuals.

Does Yang Mun use ChatGPT?

Some secondary sources say ChatGPT is used for scripting, but Choosely found no clean current first-party confirmation. HeyGen's case study says Hani writes scripts himself or uses HeyGen's own script writer.

Does Yang Mun use Nano Banana?

Google SynthID detected Google-generated media in Yang Mun content examined in January 2026. That does not identify Nano Banana specifically, so Choosely treats the claim as unconfirmed.

Does Yang Mun use ElevenLabs?

ElevenLabs appears in secondary descriptions of the stack, but HeyGen's first-party case study identifies HeyGen voice as part of the documented workflow. The current ElevenLabs role is unconfirmed.

How long does a Yang Mun video take to make?

HeyGen says the current production process takes roughly 20 minutes per video and allowed Hani to move from a few weekly posts to daily publishing.

Can one tool make an AI character like Yang Mun?

For a straightforward talking-presenter format, yes. A platform such as HeyGen can combine script assistance, avatar performance, voice and video generation. A multi-tool stack becomes more useful when you need greater control over source imagery, voice, movement or final editing.

Keep your AI stack grounded in evidence

Viral stack diagrams age quickly and often blur what is verified with what merely looks plausible.

Explore Stack Intelligence to save the tools you rely on and monitor the changes that can materially affect your workflow.

Sources

• HeyGen customer story: How Yang Mun built a following of 2.5M with AI video
• Eurovision News Spotlight investigation: Yang Mun: The Google AI-generated persona reaching millions in the wellness market
• VirtualHumans secondary stack report: Who is Yang Mun? The AI-generated wellness guru
• Religion News Service reporting: They look like religious teachers. But they are AI—and millions are listening
• Instagram policy reporting: Instagram puts new limits on undisclosed AI profiles