AI Influencer: How I make mine

Here is a really simple method I use to create my AI Influencer.

First I look for someone who has the look I am after. For this let’s say Tricia Helfer. Famous people or models are good as they already have lots of photos out there.




Then head over to grok and put in this prompt

Create a json prompt to replicate this image without naming them

This is something like what you will get

{
"prompt": "Photorealistic studio portrait of a confident adult woman with long wavy honey-blonde hair cascading over her shoulders, intense blue eyes looking directly at the camera, defined facial features with high cheekbones and full lips, subtle professional makeup, serious and strong expression. She wears a fitted black racerback tank top that shows athletic shoulders and collarbones. Hands placed on hips, upper body framed tightly. Soft even studio lighting, clean light gray seamless background, high detail skin texture, sharp focus, professional fashion photography style, 85mm lens look, natural skin tones",
"negative_prompt": "blurry, low quality, deformed, extra limbs, text, watermark, logo, cartoon, anime, oversaturated, harsh shadows, busy background, jewelry, smile, open mouth, glasses",
"style": "photorealistic portrait",
"lighting": "soft studio key light with gentle fill",
"camera": "medium close-up, eye-level, shallow depth of field",
"aspect_ratio": "3:4",
"quality": "ultra detailed, 8k"
}

Paste this into your image generator of choice and you will get something along the lines of



Ask the Ai to improve the skin tone and make her less shiney.

Upscale as best as you can.

Go back into the AI and ask for a character sheet and then a full body character sheet. It will bring back something like



Head over to google flow and create a character with these images. You can now make consistent images and videos with this model.

It’s a simple method but works for me.

very nice. will give it a try

Using grok to reverse engineer the prompt is a solid shortcut to get the initial look. Only issue with the character sheet method is that SD still tends to drift a lot when you start changing the angles or outfits for different promo posts. If you take those generated images and train a quick lora on them instead it makes the face consistency way tighter. @MasonBookwright what are you using for the video generation part of the flow? runway or luma?

KimB184 said:

Using grok to reverse engineer the prompt is a solid shortcut to get the initial look. Only issue with the character sheet method is that SD still tends to drift a lot when you start changing the angles or outfits for different promo posts. If you take those generated images and train a quick lora on them instead it makes the face consistency way tighter. @MasonBookwright what are you using for the video generation part of the flow? runway or luma?

Click to expand...

For me on the image thing it’s change one thing at a time. Upscale and reupload. So if I want her is a pod t shirt I use flow to add it then download the upscale image of her wearing it. If you change lots of things there is a detoriation.

I am really not technical for training Lora or locally hosted AI.

For video I almost exclusively use veo 3.1 fast inside google flow. I use my own prompt and the agent mode to get a batch done together then stitch them together on my phone.

Having formats really helps with this as well.