How to Create an AI Influencer and Make AI Influencer Videos, Step by Step

October 4, 2026

PlaybookMethod

To create an AI influencer, you need one image of a person, and then a steady run of posts in which that person stays the same. The image matters most. In our own work with the product, the more the person in the image looks like a real person, the more real the person in the video looks, and a too-perfect face gives you videos that look generated. This guide covers the whole path, start to finish: make the face, bring it into Riffkit as a character, then make the account's videos three ways, one for each kind of post an AI influencer needs on TikTok, Reels and Shorts.

One thing to know first: Riffkit starts from a face image you bring. Make it in any image tool, or use a real photo, and Riffkit takes over once you upload it.

Riffkit was designed for this job: one character, many videos. Characters have their own menu. You save a person once, with a face image and, if you like, a persona and a voice sample, then pick that character for each video. The same image, once it passes review, is the reference for the person in every video. In Adapt and Create, the two ways of making a video where Riffkit writes the script (step 3), the script is written with the persona in mind and the lines are voiced after the sample.

Every new account gets free credits with no card. A short video comes back whole, and most longer ones start with a free first section that a plan finishes from the same script.

How to create an AI influencer in three steps

Make or choose the face image. Any image tool, or a real photo.
Add it to Riffkit as a character. One form with four required fields, then an automatic review that usually takes under a minute.
Make the account's videos. Create for the character's own posts, Adapt for a format that is working now, Swap to keep a clip's exact shots. One page, one submit, a few minutes per video.

Already have the face? Start at step 2.

Step 1: Make or choose the face image

There are two routes, and Riffkit treats them the same way.

A new face. Make the image in whatever image tool you like. Generate a few candidates, pick one, and stop generating. From here on the person is that saved file, not the words you used to get it, so keep the original somewhere safe.

A face from a character builder works the same way. Higgsfield's AI Influencer, for example, returns a close-up portrait and a full-body shot of the character (its launch post, October 3, 2026): save them as one image, the close-up beside the full body, and upload it in step 2. We tried it on October 7, 2026 with one of its ready-made characters: the image passed review and the character starred in the first section of a Swap. Use a version without a watermark, since Higgsfield's help center says generations on a free account carry one. The character does not have to look photographic either: an illustrated or cartoon character works as a Riffkit character too.

A real photo. The person does not have to be invented. Your own photo works, and so does a photo of a model or creator who has agreed, so an AI influencer can be the on-screen version of a real person. Riffkit's Terms set the condition: for any real person other than you, you need that person's informed, written consent for this use, and you keep it on file. If the person is you, using the feature is your consent.

What makes an image work as a character's face:

Recommended layout: two views in one image. Use one image with two views of the same person on a plain background. On the left, a close-up of the face: front-facing, head and shoulders, the face filling most of the panel. On the right, the full body: standing, head to toe, in the outfit you want as the character's usual look. These are two large views, not a sheet of small poses.
One person, alone in the picture. A headshot, a half-body shot and a full-body shot are all accepted. A sheet with many small poses is not a good face image.
Facing the camera, and clear. The form asks for one clear, front-facing photo.
Real, not polished. Natural skin texture with pores and slight imperfections, not airbrushed skin. For a full-body image, a real body and not an idealized shape. The person in your videos takes after this image, so between two candidates pick the less retouched one.
Clearly an adult. The Terms do not allow depicting anyone under 18, real or presented as real, and that covers an invented face that looks underage.
Nobody else's likeness. No lookalike of a celebrity or of another creator.
A file that uploads. JPG, PNG or WebP, under 30 MB, at least 300 pixels on the shorter side, in a normal portrait or square shape (not a long strip). An iPhone HEIC photo has to be exported as JPG first.

If you plan to use Swap on close-up, talking-to-camera videos, go one step further: the face large in the frame, on a plain background. That is the picture Riffkit's page on replacing the person in a video says works. A card with several poses usually fails to change the face in a close-up.

Step 2: Add the face to Riffkit as a character

In the Riffkit app, open Characters and click New character. The form has four required fields and two optional ones:

Name.
The image from step 1.
Gender: Female or Male.
Age range: Young, Middle-aged or Senior.
Persona (optional): a few words on who the person is and how they talk. It shapes how the scripts sound, not how the person looks. How to write a good one is in the persona brief guide. If the character is a real person other than you, their written consent has to cover the persona you write for them too.
Voice sample (optional): if you want a specific voice, upload 4 to 15 seconds of clean speech as an mp3 or wav file, up to 5 MB, with no background music or obvious noise. The character's spoken lines in Adapt and Create videos are then voiced after that sample, and how closely it matches can vary by engine. A Swap does not use it: a Swap follows the source video's sound, and on some engines the swapped person's on-camera lines get a new voice. For one fixed voice, use Adapt or Create. Without a sample the videos use a default voice. The rule for the face applies here too: your own voice, or the voice of someone who has agreed in writing.

After you save, the image goes through an automatic review before the character can make any video. It usually takes under a minute, and the status updates by itself. If the image is refused, you see the reason and can retry or upload a different one. A pass means the character can now make videos. It is not a check of your right to use the face, which stays with you.

At the time of writing, a new account holds one character with one face image. A refused image does not use up that place, and you can delete the image and upload a different one. An account that has paid holds more: 50 characters, with up to 10 face images each. These numbers can change, and the app tells you when you reach the limit.

Step 3: Make your AI influencer's videos, three ways

An account is built by posting, and its posts do not all start in the same place. Some belong to the character alone: an opinion said to camera, a day in their life, the next episode of a running bit. Some join a format that is working on TikTok, Reels or Shorts right now. Some put the character into a clip whose exact shots are the point. Riffkit has one way of making each, all on the Make videos page, and the same approved image is the reference for the person in all three. Pick the character, choose how to make the video, and submit.

	Starts from	What your character brings	Use it for
Create	A creative direction you write. No source video	The face. The persona shapes the script, the setting and the lines, and the voice sample voices them	The character's own posts: an opinion, a day in their life, the next episode of a series, a product they use
Adapt	A video that already performed: its hook, pacing and beats	The face. The persona shapes the new story, and the voice sample voices the lines	Joining a format that is working now, with the character's own story. Also the choice for another creator's video
Swap	A clip whose shots you keep: your own post or a library template	The face, in place of the person. Shots, cuts, timing and music follow the clip	Running a clip again with your character in it, when the exact shots are the point

Videos use credits, counted by the second, and the page shows what a video needs before you submit. A source for Adapt or Swap can be a template from Riffkit's library, a link to one TikTok video, or a video file you upload, which is how a Reel or a Short goes in. At the time of writing it can be up to 45 seconds long, and a longer one is refused, not trimmed. A Swap uses more credits per second than an Adapt, so make your first Swap a short one. For choosing a source, see how to find winning TikTok videos.

Create starts from the character, not from a source. Open Create, pick the character and write the creative direction in a sentence or two: who does what, and where. For example: "She packs the last orders of the day in her garage, ten minutes before the post office closes, and explains to her dog why he does not count as a coworker." Riffkit writes an original script around the persona, how the character talks and where they spend their days, and makes the video. Set the length from 4 to 45 seconds, or choose Auto length and Riffkit picks it, never over 45 seconds. This is where a series lives: the same premise, a new episode each time.

Adapt is how the character joins what is already working. It keeps what made a video perform, its hook, pacing and beats, and tells the character's own story with new footage. You pick the language (nine are offered) and the shape: 9:16 unless you choose 3:4, 1:1, 4:5 or a landscape ratio. When the script has captions, they are burned into the video and timed to the speech. To be sure of them, write what they should say in the optional direction box, which also lets you steer the story in plain words: write what changes shows how. Music can carry over: when the source has a music track, Adapt lays it under your new video by default. Choosing Reference original BGM asks for a new track in the same style, and with that choice the voice sample is not used for that video.

That is the method behind our own published numbers: 1,504 AI-generated videos posted to real TikTok accounts and tracked through August 2026, each one rebuilt on the structure of a video that had already won. The best passed 1.5 million views with an 18% engagement rate, where 4 to 6% is usually called strong. The structure of the source mattered too: videos built on a two-beat source, one that grabs attention and pays it off with nothing between, cleared 1,000 views about twice as often as the rest, 31.7% against 15.7%. That is a correlation, not a controlled test, and the median video reached 418 views. The whole distribution is in do AI videos actually get views.

Swap keeps the video and changes who is in it. Camera work, cuts, timing, language, music and shape follow the original, and the result is exactly as long as the source. The source has to show a person for your character to take the place of. The sound follows the source too: on some engines your character says their on-camera lines in a new voice, and on others the original voice stays. The whole video is made again, it is not a face pasted onto old footage, so watch it before you post: a face may not change fully. Riffkit's own advice is to keep Swap for a video you have the right to use, and to choose Adapt for another creator's video. The reasons are in can you legally remake a viral TikTok.

Click for the silent comparison. Left, our own 12-second clip. Middle, Swap: the same shots with a saved character in place of the woman. Right, Adapt: the same beats (the phone call, the shoe sole over the lens, the reveal, the walk) told as the character's own story, from her garage to a backyard birthday party for her dog. Both made on Seedance 2.5 at 720p.

To swap the person, pick the character and leave Swap's optional "What to change" box empty. In the middle column her face is in every shot, and the shots, cuts and caption stay where the original has them. In the car and on the red carpet she wears the original's gown and gloves, and on the fire escape the outfit from her character image. Her character image uses the recommended layout from step 1.

For the Adapt on the right, the creative direction told her own story in a few sentences: where she is before the cut, what the reveal shows, and that her backyard party takes the place of the car and the red carpet. Our first Adapt of this clip had a direction that only asked for the person to be replaced, and it followed the original's story with a new face, which is what Swap is for.

A post for a product, your own or a brand's, works in all three. Save the product once with a clean photo (several on a plan). In Create and Adapt it can be on camera or only mentioned; in a Swap it goes on camera, so the product needs at least one photo. How placement works: AI product placement.

Each finished video comes with a suggested post caption, hashtags and a cover frame. You download it and post it yourself.

On consistency, the honest bar is recognizable, not identical in every frame. The same approved image is the reference for every video, which is what keeps the person from changing the way one described only in words does. More on that in keeping the same AI character across every video.

If your AI influencer sells on TikTok Shop

TikTok Shop's US seller rules speak to AI presenters directly. As of their last updates, September 2 and September 17, 2026:

Say it is AI, in the video. A video with a synthetic face, voice or digital human has to state that it is AI-generated, with on-screen text, a watermark, a sticker or a note. That comes on top of TikTok's own AI label, covered in the AI label guide.
Show the product as it is. AI-made scenes are allowed when they match the product's real size, material and function. Results the product cannot achieve are not.
No invented experts. No fabricated doctors or professors, and no digital human presented as a newscast or an expert lecture, to vouch for a health product.
No AI voices in a LIVE.

One question the pages leave open: the rules for high-quality videos ask for videos that feature your face and the physical product, and neither page says how a video presented entirely by an AI person is judged against that. If shoppable videos are how the account will earn, read both pages yourself, AI-generated content and high-quality videos and LIVEs, before you build the shop on an AI presenter alone.

Doing this responsibly

Four rules, all from Riffkit's Terms:

Say it is AI. Where a platform or the law requires AI-generated content to be labeled, applying the label is your job, and the videos must never be presented as real recordings. A natural-looking person is about quality, not about hiding anything. For TikTok, read the AI label guide before you post.
Adults only. No persona under 18, real or presented as real.
No impersonation. A real person appears only with their written consent, and never as a deceptive stand-in for a public figure.
Nothing sexual or intimate involving a real person.

One more honest note: a consistent person in a proven format is a good start, not a promise of views. Even in our own numbers, the median video reached 418.

Start with the image

Spend your time on step 1, because every video takes after that one image. Then add the character in Riffkit and make the first video: the character's own post in Create, or a format that already won in Adapt. If you work in a coding agent, the Riffkit skill makes the videos once the character exists. The character itself is set up in the web app.

Signing up is free with no credit card, and the first video runs on the free credits every new account gets. When you want whole videos at 720p and more than one character, the plans are here.

FAQ

How do I create an AI influencer?

In three steps. First, make one image of the person in any image tool, or use a real photo you have the right to use, such as your own. Second, upload that image to Riffkit as a character: a name, the image, a gender and an age range are required, a persona text and a voice sample are optional, and the image goes through an automatic review that usually takes under a minute. Third, pick that character each time you make a video, whichever of Riffkit's three ways you use, and the same person appears in each one. Riffkit starts from the image you upload.

How do I make AI influencer videos for TikTok, Reels and Shorts?

Set the person up once as a character in Riffkit, then make each post one of three ways. Create writes an original video around your character from one creative direction you write, 4 to 45 seconds long, with no source video. Adapt starts from a video that already performed: it keeps what made it work (hook, pacing, beats) and tells a new story with new footage starring your character. Swap keeps a clip's shots, cuts and timing and puts your character in place of the person. A source for Adapt or Swap can be a template from Riffkit's library, a link to one TikTok video, or an uploaded video file, which is how a Reel or a Short goes in. Create and Adapt videos are 9:16 unless you choose another shape; a Swap keeps the shape of its clip.

What should an AI influencer post?

A mix of three kinds of post, and Riffkit makes one kind of video for each. The character's own posts, such as an opinion said to camera, a day in their life or the next episode of a running series, start from a creative direction you write, in Create. Posts that join a format already working on TikTok, Reels or Shorts start from that video, in Adapt: it keeps the hook, pacing and beats and tells the character's own story with new footage. A clip whose exact shots are the point, such as your own best post or a library template, goes through Swap, which keeps the shots and puts your character in place of the person. The persona and the voice sample shape Create and Adapt videos; a Swap carries the character's face.

What kind of image works best as an AI influencer's face?

One clear, front-facing image of one person that looks real and unretouched. In Riffkit's own experience the image decides the video: the more the person in the image looks like a real person, with natural skin texture and slight imperfections, the more real the person in the video looks, and an airbrushed, too-perfect face makes the videos look generated. A headshot, a half-body shot and a full-body shot are all accepted, and the layout we recommend is one image with two views of the same person: a close-up of the face on the left and the full body on the right. Avoid sheets with many small poses. The file should be JPG, PNG or WebP, under 30 MB, and at least 300 pixels on its shorter side. A real-looking person is about quality, not about hiding that the video is AI: where a platform requires an AI label, apply it.

Where does my AI influencer's face come from?

From one image you bring. Make the face in any image tool, or use a real photo you have the right to use, and upload it to Riffkit as a character. Riffkit does the video side: it uses that one image as the reference for the person in every video you make, whether the video follows a winning source (Adapt or Swap) or starts from one written idea (Create).

Can I use a character I made in Higgsfield's AI Influencer to make videos in Riffkit?

Yes. Higgsfield's AI Influencer returns a close-up portrait and a full-body shot of the character (Higgsfield's launch post, October 3, 2026). Save them as one image without a watermark, the close-up beside the full body, and upload it to Riffkit as a character image. It goes through the same automatic review as any other image, usually in under a minute. From then on you pick that character for each video: Adapt for a format that won on someone else's account, Swap for a video that is yours to use, up to 45 seconds, and Create for an original video from one written idea. Realistic faces and illustrated or cartoon characters both work as characters.

Can I base an AI influencer on myself or on another real person?

Yes for yourself: upload your own photo as the character image in Riffkit, and under Riffkit's Terms using the feature is your consent. For another real person, such as a model or a creator, the Terms require that person's informed, written consent for this use, including any persona you write for them, which you keep on file and provide if asked. Not allowed in any case: anyone under 18, impersonating a person, a deceptive depiction of a public figure, and sexual or intimate depictions of a real person. Every uploaded face goes through an automatic review before it can be used, and passing that review does not clear your right to use the face.

Can my AI influencer have its own voice?

Yes, in Riffkit's Adapt and Create videos. When you create the character, upload a voice sample: 4 to 15 seconds of clean speech as an mp3 or wav file, up to 5 MB, with no background music or obvious noise. The character's spoken lines are then voiced after that sample, and how closely the voice matches can vary by engine. One exception: if you set the background music to 'Reference original BGM', the sample is not used for that video. A Swap video does not use the sample: a Swap follows the source video's sound, and on some engines the character's on-camera lines get a new voice, so for a persona with one fixed voice use Adapt or Create. Without a sample the videos use a default voice. Use your own voice, or the voice of a person who has given written consent.

Can an AI influencer remake a trending TikTok, Reel or Short?

Yes, and how depends on whose video it is. For a format that worked on another creator's account, use Adapt in Riffkit: it keeps the hook, pacing and beats and makes a new story with new footage and your character. Keep Swap, which follows the original's shots, cuts, timing, language and music and changes the person, for a video you have the right to use, such as your own post, or for a template from Riffkit's library. A Swap makes the whole video again, so watch the result before you post it: the face may not change fully, and on some engines the voice changes too.

Do AI influencer videos get views?

Some do and most do not, the same as filmed short video. In our own published output of 1,504 AI-generated videos posted to real TikTok accounts and tracked through August 2026, every one made by rebuilding the structure of a video that had already won, the best passed 1.5 million views with an 18% engagement rate, the median reached 418 views, and about one in five cleared 1,000. Videos built on a two-beat source, one that grabs attention and pays it off with nothing between, cleared 1,000 views about twice as often as the rest: 31.7% against 15.7%. That is a correlation in published data, not a controlled test.

Can an AI influencer sell on TikTok Shop?

Yes, within TikTok Shop's own seller rules (US Seller University pages last updated September 2 and September 17, 2026). A video with a synthetic face, voice or digital human has to state that it is AI-generated, with on-screen text, a watermark, a sticker or a note. The product has to look and work as it really does. Invented doctors or professors, and digital humans presented as a newscast or an expert lecture, may not vouch for health products, and AI-generated voices are not allowed in a LIVE. The same rules ask high-quality videos to feature your face and the physical product, and neither page says how a video presented entirely by an AI person is judged against that, so read both pages before you build a shop on an AI presenter alone.

Do I have to disclose that my AI influencer is AI-generated?

Where a platform's rules or the law require AI-generated content to be labeled, yes, and applying the label is your responsibility, not the tool's. Riffkit's Terms say this directly and also forbid presenting a generated video as a genuine recording. In practice, when a video shows a realistic AI-generated person, use the platform's built-in AI-content label when you upload it, and check each platform's own help page for its current rule.