Is It Really Possible to Keep AI-Generated Characters Consistent Across Images?

alamin1321

alamin1321

Newbie
Joined
Dec 5, 2023
Messages
27
Reaction score
28
Hey,
I’ve been experimenting with AI image tools lately, but one thing I’m struggling with is how to maintain a consistent character appearance across different images. Like — same face and style, but in new poses or settings.

I’ve tried repeating the same prompt, even referencing the old image, but I still get different results each time.

If anyone here has cracked the trick or has a method that works, I’d appreciate it if you could share. Just want to make the character feel like the same person every time.

Thanks in advance!
 
Advertise on BHW
Why not develop a custom tool for that and integrate with the AI image generator API like Flux and define the variables and characters prompt field in such a way that each character will generate onto its own unique playground fully customizable and you can always replicate it each time.

If you are into WordPress, you may want to develop a custom plugin to handle it and render it on a frontend via shortcode. There are many AI image generator APIs out there to integrate with your custom code.

You should also consider using a JSON file defining each attribute for each character.
 
Last edited:
The way to go is to use frames- to video to generate, then upload the frame. Text to video will always give random output.
 
Have you tried creating a well detailed JSON profile of the character?
Use the JSON when testing a new image.
That should work
 
Copycrypto said:
Have you tried creating a well detailed JSON profile of the character?
Use the JSON when testing a new image.
That should work
Click to expand...
How can you share the process? It is possible with the help of google veo 3?
 
try using some image services on replicate they allow you to generate then "tweak" the prompt it gives almost the same person every time slightly changes but pretty close.
 
bonzo90 said:
try using some image services on replicate they allow you to generate then "tweak" the prompt it gives almost the same person every time slightly changes but pretty close.
Click to expand...
I tried generating a first image for VEO 3 with Replicate too, but I’m not getting good results. My prompts are in JSON, but I still keep getting inconsistencies, I don’t know what to do anymore…
 
Have you experimented with LORAs? If you make your prompt specific and detailed enough you can get enough images that look close enough as one person or so. Then train the model and now you have a more consistent model.
 
use json and create that person then you can recreate each time that person accurate
 
VEO3 frames i2v method works. You need to structure the scenes though, with specific inputs on what each scene will be, preferably using json. Claude code helped me vibe code a python script which does this automatically for some faceless AI video generation and it works fairly well. You need need access to VEO3 FAST credits (unless you have a big budget for VEO3 quality) API. You can get it through fal.ai or kie.ai.
 
Advertise on BHW
You must log in or register to reply here.
Sign up now!

Main Menu

Home Main Forum List Black Hat SEO White Hat SEO BHW Newbie Guide Blogging Black Hat Tools Social Media Downloads

Marketplace

Account Selling Content / Copywriting EBooks / Methods Hosting Misc Proxies For Sale SEO - Link building SEO - Packages Social Media Social Media - Panels Web Design More

Making Money

Affiliate Programs Hire a Freelancer IM Journeys Making Money Pay Per Click Site Flipping Want to Buy

BlackHat World

Dispute Resolution Forum Suggestions & Feedback Introductions Lounge
Back
Top