Use case

Create a consistent AI influencer

An AI influencer only works if the same face, body and style show up in every post. The practical way to get there is to stop prompting from scratch and work from reference images: one approved portrait feeds every photo, and every video starts from a photo.

Why this approach works

• Text prompts alone drift. A description like "brunette woman, 28" produces a different person each time; a reference image anchors the identity.
• Image-editing models take several references at once, so you can supply the face from different angles and lighting.
• Video models that start from a still inherit the face, outfit and framing of that still instead of inventing them.

Step-by-step workflow

1. 1

   Lock the base portrait

   Generate a handful of portraits with a quality text-to-image model and pick one you would be happy to see 100 times. This is your identity anchor: front-facing, neutral light, nothing covering the face.

2. 2

   Build a small reference set

   Create two or three more images of the same person from other angles (three-quarter view, profile, a smile). Edit models let you generate these from the base portrait. Discard anything that looks like a different person.

3. 3

   Generate scenes from references

   Use an image-editing model with your references attached and a prompt that states the scene, outfit and light. Always describe what must stay the same (face, hair) as well as what changes.

4. 4

   Animate the best stills

   Feed approved stills to an image-to-video model, or give a reference-driven video model several references directly. Describe the motion and camera, not the look: the image already carries the look.

5. 5

   Batch the week

   Plan content in one sitting: outfit, location and caption per post. Draft on cheaper models, regenerate only the keepers on premium ones.

6. 6

   Check platform rules

   Most platforms ask you to label AI-generated or synthetic content. Add the label to every post and avoid using a real person's likeness without consent.

Recommended models

Example prompts

What a typical week of content costs

A realistic weekly plan for one account: a set of everyday photos, a few hero shots and a handful of short clips. Adjust the quantities to your own posting rhythm.

This assumes one generation per asset. Budget extra credits for retries: AI generation is never one-shot, and drafting on a cheaper model first keeps retries inexpensive. Price your own mix with the video cost calculator.

Frequently asked questions

How do I keep the same face across AI images?

Work from reference images instead of text alone. Attach your approved portrait (plus two or three more angles) to an image-editing model and state in the prompt that the face and hair must stay unchanged.

Which model is best for a consistent character in video?

Gemini Omni Flash 1.1 accepts up to seven reference images and is designed for character consistency. Image-to-video models such as Seedance 2.0 and Kling 3.0 Pro also keep identity because they start from your approved still.

How many credits do I need per month?

The weekly plan above shows the credits for one week. Multiply by four for a month and compare with the plan allowances in the calculator table.

Do I need to disclose that the influencer is AI?

Platform rules vary, but most require labelling synthetic content. Label every post, and do not imitate a real person without their consent.

Related use cases

• UGC adsProduce product ads in the style of user-generated content: script, shots, models and cost per ad.
• Fashion lookbooksShow a garment on consistent models, in several settings, without a studio day.
• Storyboards for filmmakersPlan shots, pitch a film and test camera moves with consistent storyboards and animatics.
• A week of social media contentPlan and generate seven days of posts, stories and short clips for a brand in a single sitting.