AI influencer guides

One original face, a 50-image dataset, a character LoRA you own, the AI label on, then growth and Fanvue. Real steps, settings and GPU costs.

The Z-Image Turbo settings our face workflow runs: 9 steps, CFG 1, shift 3, dpmpp_2s_ancestral, 1024×1536. Why each value, and what breaks.

Prompts, reference images, a character LoRA, or a LoRA trained on a dataset you generate. What each holds, what it costs, and when you need a LoRA.

A character LoRA needs 20 to 50 images. We train on 50. What they should show, how to caption them, how to pick a trigger word, and what goes wrong.

Compare the 2000, 2500 and final 3000-step files with the same prompt and seed, as video 3 of our course does. How to judge them and pick the keeper.

LoRA doing nothing, giving the wrong face or not loading in ComfyUI? Seven checks, cheapest first, from the file to the dataset.

A trigger word is a made-up token plus a class word, like zvx woman. How to pick one, where it goes, and what to check when a LoRA ignores it.

We run both as paid engines in our Dataset Maker: $0.14 vs about $0.03 a photo, which versions you actually get, and the job only Seedream would do.

Three ways to animate an original AI character: Kling (rented, fast), Wan 2.2 (open-weight on RunPod) or LTX (cheap batches). What each needs and costs.

Make AI influencer dance videos from clips you filmed or licensed: one sharp image of her, two LTX prompts, batches of ~20 clips, and the AI label.

Kling is rented and fast, Wan 2.2 is open-weight on a rented GPU, LTX makes cheap batches. Trade-offs, real timings and versions checked 2 Oct 2026.

Rebuild the clip's first frame with your AI character, then run motion control with one prompt. Kling's clip rules and credit costs, checked Oct 2026.

Rent a GPU by the hour instead of buying one. Which RunPod GPU to pick, what storage costs, how to run ComfyUI, and the stop vs terminate trap.

Fix ComfyUI missing models and "Value not in list": the right folder for each file, names that match exactly, and uploading models to a RunPod pod.

A RunPod pod that won't start is usually still downloading. Read the logs, wait out first boot, then check ports, GPU count and the CUDA filter.

The pod's image needs a newer CUDA than the machine's driver. On RunPod you redeploy with the CUDA Versions filter, e.g. 13.0 for our LoRA trainer.

What actually pays for an original, disclosed AI character: Fanvue, AI UGC for brands, brand posts. Real costs, a worked example, no fake numbers.

OnlyFans needs a verified human behind every account. Fanvue allows disclosed AI creators. What that means for an AI model, with sources and dates.

Fanvue allows AI creators who disclose it. The ID check, the 20% fee, $20 payouts, where the AI disclosure goes, and the settings we use.

Fanvue allows $3.99 to $100 a month. What the course's Fanvue lesson sees, what Fanvue's Aug 2026 data says works, bundles, trials and price rises.

Collab, trend and main accounts, a 7-day warm-up, a reel and carousel every 2 days, an optional $20/day boost on your 3 best posts. AI label on.

Instagram, Meta, TikTok, Fanvue, OnlyFans, the FTC, the EU AI Act and the TAKE IT DOWN Act: what each requires from an AI influencer. Checked 2 Oct 2026.

The FTC doesn't ban AI avatars in ads, but fake testimonials are illegal. TikTok Shop's AI rules, ad disclosure and the EU AI Act. Checked 2 Oct 2026.

TikTok's Creator Rewards rules don't mention AI: 10,000 followers, 100,000 views in 30 days, original videos over 1 minute, labelled. Checked 2 Oct 2026.

Instagram doesn't hide AI for being AI. It stops recommending unlabelled AI people, suggestive posts and spammy accounts. How to check and fix it.

Since 31 Aug 2026 Instagram limits AI personas without the AI-generated profile label. Who needs it, how to turn it on, and what it does to reach.