Make an AI Influencer for Free in 7 Lessons

Picture the account first. Dario Vela is a fictional 33-year-old home cook who makes 20-minute dinners in an apartment kitchen barely wider than a hallway. He posts a couple of short videos a week, his face never changes, and his bio says he's AI-generated. This course builds a persona like him in seven lessons, following the free How to Use playbook in Spicy Studio (it opens with a free account) and its pipeline: Idea → First frame → Motion → Edit → Post.

Free. This course, the playbook, the planning in Lesson 1 and the 25 credits every new account gets at sign-up, no card required.

What 25 credits buy. One face reference and one finished first frame: 5 credits for the face (GPT Image 2 at 1K, low quality), 10 for the face-replaced photo (1K, medium) and 10 for an enhance pass in Nano Banana 2 at 2K. That's enough to see whether the persona works as a still.

Not free: video. The cheapest clip in the playbook is 5 seconds of Seedance 2.0 at 480p, for 35 credits; Lip Sync starts at 50 and a Kling 3.0 Motion Control remake at 60. Posting every week means paid credits: top-ups start at €2 for 90, and Spicy Basic is €10 a month for 600.

Pro only. The Complex motion and Sketch workflows need an active Spicy Pro plan (€25 a month, 1,800 credits) or higher. Everything else works on any plan.

This lesson costs nothing and saves the most credits later. A persona card is a half-page note you reread before every post; if a caption, outfit or location doesn't fit it, the post doesn't go up.

Identity. Name, age, one-line backstory. Fictional and clearly adult: Dario is 33, and his face must read as an adult at a glance, whatever age the prompt says.

Niche and promise. One subject a stranger gets in a second. “Twenty-minute dinners for one, in a tiny kitchen” works; “food content” doesn't.

Three repeatable formats. Posts you can make every week without a new idea. For Dario: a dinner race with a timer on screen, a “fridge rescue” from leftovers and a one-pan Sunday recipe.

Voice. Dry humor, captions under two lines, a question at the end, in captions and audio alike.

Sponsor fit. Which brands already pay human creators in the niche? For Dario: cookware, grocery delivery, kitchen gadgets. If none come to mind, the account will be hard to fund; How Do AI Influencers Make Money? covers the income side.

Hard limits. What the persona never does: - Wear a real person's face. Spicy Studio's Terms prohibit realistic images of identifiable real people without their consent. - Play an expert. YouTube won't monetize channels that present AI personas as human experts on health, legal, financial or political topics, so there's no “dietitian” Dario. - Borrow material. Every photo, clip and voice you build on belongs to you or comes with permission.

An 18+ subscription persona is a different business with different rules: see What Is AI OFM?, the Free AI OFM Course and the playbook's 18+ Spicy section, whose workflows are marked as coming soon.

Followers recognize an AI influencer only if the face never changes, and that depends on one file: a front-facing image of the persona, attached to every generation. The playbook calls it the face reference.

Make it with GPT Image 2 in the Image Lab. Write it like a casting note: age, hair, one or two details you'd recognize across a room, a plain top, soft light and a bare background.

Example prompt: “Vertical 9:16 phone photo, straight-on head-and-shoulders portrait of a 33-year-old man with short dark curly hair, a trimmed beard with a small gray patch on the chin and a thin scar through his left eyebrow. Plain navy T-shirt, relaxed neutral expression, looking into the lens. Soft window light from his right, off-white wall behind him. Natural skin texture, no retouching.”

Draft cheaply. At 1K, a low-quality draft costs 5 credits and medium costs 10. Keep the draft with details you'd spot instantly (here, the scar and the gray patch), and when credits allow, sharpen it once with the Lesson 3 enhance pass, which keeps the person the same. GPT Image 2.5 also works, at 9 credits for 1K medium plus 1 credit per reference image.

One file, for good. Keep it 9:16, like the playbook's own face reference, and never regenerate it. Attach that same file to every image you make, and begin each video from one of those images; a second face reference is a second person. For drift problems, read Consistent AI Character; for quality and size options, see the GPT Image 2 guide.

Scenes generated from scratch are where AI influencers start to look like renders. The playbook's first-frame workflow starts from a real photo instead, with real light, real framing and a pose a human actually held, and changes only the face.

1. Stage the base photo. Pose, outfit and setting carry over. For Dario, lean a phone against the spice rack, set the timer and photograph yourself (or a friend who agreed to it) stirring a pan. Your own photos and images licensed for editing also work; strangers' posts don't.

2. Attach two references. In the Image Lab, the base photo is reference 1 and the face reference is reference 2, at 9:16, on GPT Image 2, GPT Image 2.5, Nano Banana 2 or Nano Banana 2.1.

3. Prompt. Adapt the playbook's one-line prompt to your persona: “Same as image ref 1, but with the man from image ref 2”. The playbook writes it for its female example character, so use the word that fits yours. GPT Image 2 at 1K medium costs 10 credits.

4. Enhance. Open Nano Banana 2 with the result attached and type “Enhance the image quality” (10 credits at 2K, or 12 on Nano Banana 2.1). If it changes too much, the playbook's longer prompt holds the person, face, pose, outfit and background fixed and improves only detail, skin and light.

5. Inspect. The playbook's six checks, condensed: a sharp face, the usual face reference, a vertical frame, a real phone look with no gloss, an outfit and place that fit the niche, and nothing that will move cut off at the edge. Our Free AI Video Course for Beginners runs through the same check, with examples, in its first-frame lesson.

No base photo? Attach the face reference to Nano Banana 2 and describe the shot plainly:

Late dinner: “Phone photo, vertical 9:16: the man from the reference image tastes sauce off a wooden spoon in a narrow apartment kitchen at night, warm under-cabinet light, steam over the pan, cluttered counter, real skin.”

Market run: “Phone photo, vertical 9:16: the man from the reference image holds a paper bag of tomatoes at an outdoor farmers' market on a gray Saturday morning, canvas stalls behind him, candid daylight.”

A video model animates the still you give it and can't rescue a weak one, so every route here starts from a frame that passed the Lesson 3 check. Pick the route by what the shot needs.

One small moment: animate. A lid lifting, a first bite, a raised eyebrow. In the animate workflow, attach the frame in the Video Lab and choose Seedance 2.0 or 2.5, which the playbook prefers to the Kling models here. Seedance 2.0 defaults to 16:9, so switch it to 9:16; Seedance 2.5 takes its shape from the frame. Write concrete actions in order, in simple words, and keep the motion small: plain image-to-video stops looking real when you ask for a whole scene.

Prompt for Dario: “The camera stays completely still. The man lifts the lid off the pan and steam rises. He leans in, breathes it in, nods slowly, then looks into the camera and raises one eyebrow. Ultra realistic. Natural movements.” Those last two sentences are the playbook's ending for believable motion. Five seconds at 480p costs 35 credits on Seedance 2.0, or 50 on Seedance 2.5, which adds sound by default.

A movement you filmed: motion control. A knife-skills flourish, a dance at the stove. Film it yourself, or use a clip you have permission for, with one person, a simple background and the face visible throughout. The motion-control workflow has you screenshot its first frame, face-replace it and open Character Swap with Kling 3.0 Motion Control: your frame is the character image, the clip is the motion video, and the keep-original-audio toggle saves your voice or the sizzle. Standard is 20 credits a second, billed for at least 3 seconds, so a 10-second remake costs 200.

Camera moves or fast action: complex motion. Regular motion control is the weaker option once the camera travels. In the complex motion workflow, part of Spicy Pro, Seedance copies a reference video's motion, camera work and audio onto your character. Seedance 2.0 at 480p charges 4 credits per second of output plus reference, so copying a 10-second clip costs 80. As with every route, use footage you shot or have permission for: trends are for ideas, not for re-uploading.

How you make an AI influencer talk depends on who supplies the voice.

Your voice, word for word: lip-sync. In the lip-sync workflow you record the line yourself, or use a voiceover you hold the rights to, as an MP3: one clean voice, little music, trimmed to the line, since shorter audio syncs more accurately. Make a frame of the persona facing the camera with lips closed and load both into Lip Sync. Keep the default prompt, “The camera remains still at all times.”, or add body language for an emotional line: “He shrugs, laughs halfway through and points at the pan. The camera remains still at all times.” Audio can run up to 60 seconds. Standard (480p) is 10 credits a second and Pro (720p) 18, billed per whole second with a 5-second minimum, so a 15-second voiceover on Standard is 150 credits. More in Make a Photo Talk With AI.

Lip-sync also rescues failed motion control: when the person in your reference clip turns away or hides their face, keep the audio and sync it onto a clean frame.

AI-generated voices: a sketch. Street interviews, POVs, podcast-style clips and comedy bits, with dialogue and background sound generated in one pass by Seedance 2.0 or 2.5, make up the sketch workflow, part of Spicy Pro. Keep scripts short, with a hook up front and a punchline at the end. Seedance 2.0 clips run 5 to 15 seconds and 2.5 up to 30. Seedance 2.0 starts with audio off, so switch it on; ten seconds at 480p with sound costs 70 credits. The Seedance Prompt Guide has a free template.

For Dario, lip-sync carries the recipe voiceovers, where quantities must be exact, and sketches carry the comedy, like a fake food critic stopping him on the street.

A raw generation isn't a post yet. The playbook's edit checklist is short on purpose: - Open on a caption or a POV line in frame one, since plenty of people scroll with the sound muted. - Emojis and on-screen text on the moments that matter. - A meme sound on the reaction. - A fast punch-in when the joke lands; it can also mask a small AI glitch. - No heavy effects.

Remake the format, never the video. Find what keeps working in your niche and write your own lines for your own character. Near-copies of your own clips don't count as new either: the same script on a new background gives viewers nothing, and YouTube won't monetize content that looks mass-produced from a template.

A first month on Spicy Basic. One way 600 credits (€10) could cover Dario's first month, before retries: - 8 first frames at about 20 credits each (face-replace plus enhance): 160 credits - 4 animated moments, 5 seconds each on Seedance 2.0 at 480p: 140 credits - 2 lip-synced voiceovers, 10 seconds each on Standard: 200 credits - 1 motion-control remake, 5 seconds on Standard: 100 credits

That's seven short videos plus a spare frame for a photo post, roughly two uploads a week. Retries eat into it, so treat it as a ceiling; plans and top-ups are on the pricing page.

Let results pick the format. After a few weeks, compare watch time and shares across the three formats and make more of the winner. No schedule promises views. If clips stall, Do AI Videos Get Less Views? walks through the diagnosis, and six Reels formats give you more to test.

AI influencers are allowed on Instagram, TikTok and YouTube if they're honest about two things: being AI, and being paid. Here's what to switch on, as of October 2026.

Instagram. On August 31, 2026, the “AI creator” label became AI-generated profile, for accounts that feature an AI-generated person; turn it on in Edit profile. Instagram says profiles that add it themselves keep their reach, while unlabeled AI-person profiles it detects can lose recommendation eligibility until they add it or win an appeal in Account Status. Creators who only use AI tools don't need it. Meta also requires its AI disclosure on posts with photorealistic video or realistic-sounding audio that was digitally created or altered.

TikTok. Its Community Guidelines require clear labels when AI realistically depicts people or scenes, which covers every Dario clip. Turn on the AI-generated content setting when you post; TikTok can also add the label from Content Credentials metadata.

YouTube. When YouTube Studio asks about AI use at upload, answer yes for realistic scenes that never happened, a realistic fictional person included. YouTube's disclosure page says disclosing doesn't by itself limit a video's audience or its eligibility to earn; consistently skipping it can bring an added label, removed videos or a Partner Program suspension.

Hiding it isn't a plan. Spicy Studio's Terms say generated content carries machine-readable metadata identifying it as AI-generated, and TikTok and YouTube both say they can label posts from AI metadata.

Sponsored posts in the US. The FTC's 2023 Endorsement Guides update made clear that endorsements include virtual influencers, so paid posts need a clear disclosure such as “Ad”, and a platform's built-in tool may not be enough by itself. The FTC's consumer reviews rule (16 CFR Part 465, in force since October 21, 2024) bans fake reviews and testimonials, and FTC staff say it has no blanket ban on AI avatars or virtual influencers. Dario can show a pan and repeat claims the brand can back up. He can't say he's cooked with it every night for a month.

New York. Since June 9, 2026, an ad must conspicuously disclose a “synthetic performer”, an AI-made person not recognizable as a real one, when the advertiser knows it includes one. Civil penalties: $1,000 for the first violation, $5,000 for later ones (bill S8420A). Expect brands to ask for it.

Canada and Australia have their own rules, summarized in How Do AI Influencers Make Money?, and Is AI OFM Legal? covers label appeals. This is general information, not legal advice.

Only when it teaches something up to date that free material doesn't. Whether you're reading sales pages or AI influencer course threads on Reddit, four checks sort most of them.

Promises of followers, deals or income. No course controls an algorithm or a sponsor's budget, and income screenshots say nothing about typical results. How to Make Money With AI lists the red flags regulators have acted on.

Old rules. A course that skips Instagram's August 2026 label, TikTok's AI-generated content setting and YouTube's July 2026 monetization clarification is teaching last year's game.

A hidden lesson list. If you can't see the curriculum before you pay, assume it's thin.

Automation. Posting bots and “done-for-you” accounts make the kind of mass-produced uploads YouTube won't monetize.

“Free download” copies of paid courses tend to be old versions, the files can hide malware, and logging in to shady sites puts your accounts at risk. Nothing here needs downloading. For 18+ personas, the Free AI OFM Course lists red flags specific to OFM courses.

This course teaches a skill. It is not a promise of income, followers or sponsorships, and it isn't legal or financial advice. Many accounts, human or AI, never earn meaningful money. Results depend on your niche, content, consistency, costs and luck, and credits, platform fees and anything else you buy are real expenses. If you run sponsored posts or make claims about earnings, the FTC in the US, the Competition Bureau in Canada and the ACCC in Australia treat misleading claims and hidden ads as illegal. None of these regulators has reviewed or endorsed this guide.