How to Create Consistent AI Avatars in Bulk with AutoWhisk
Why AI characters change between images, and how I keep one character consistent across 20 images with ChatGPT, Gemini and the free AutoWhisk extension.
Key takeaway
Build one reference character, write all prompts at once separated by blank lines, and let AutoWhisk batch them.
In this article8 sections · 6 min read
The video plays from YouTube when you press play.
Typing the same character description over and over is the slow part of AI image work. And even when you do, the character often looks slightly different in every image. This workflow keeps one character consistent across 20 images and generates them all in one batch.
I use it for storyboards and story-style videos. Here is each step, and why it works.
Why AI characters change between images
Image generators do not remember earlier images. Each prompt starts fresh, so small differences in wording, or none at all, can change the face, hair or clothes. Two things fix most of this: using the exact same character description in every prompt, and giving the generator a reference image of the character to follow.
The workflow
Three tools, one consistent character
- ChatGPT
- Gemini
- AutoWhisk
What you need
- ChatGPT, or any AI writing tool, to write the story and prompts.
- Google Gemini, to create the base character image.
- The free AutoWhisk browser extension, to send many prompts to Google's Whisk image tool in one batch.
Step 1: Write the story and scenes
Ask ChatGPT to write a short story, describe one main character in detail and list 20 scenes from the story. One fixed character description is what keeps every image consistent.
Make the description concrete: age, hair color and style, eye color, clothing with colors, and one or two unique details, such as round glasses or a red scarf. Vague descriptions like "a young man" give the generator too much freedom.
Step 2: Create the base character
Use the character description in Gemini to generate a single clear image of your character. A plain background and a front-facing pose work best. This becomes your reference image, so generate a few and pick the one you like most.
Step 3: Prepare the prompts
Ask ChatGPT to turn the 20 scenes into 20 image prompts and separate each prompt with two line breaks, which means one blank line between prompts. AutoWhisk uses those blank lines to split them into separate images.
Ask ChatGPT to start every prompt with the same short character description, then add the scene. For example: "A young woman with short black hair, round glasses and a red scarf, walking through a rainy market at night."
Step 4: Generate in bulk
- Add the AutoWhisk extension to your browser and open it.
- Paste all 20 prompts.
- Add your base character image as the reference.
- Start the batch and let it generate every image.
Leave the browser tab open while it works. Depending on the number of prompts and the service's limits, it can take a while.
Tip: Keep the character's key details, such as hair, clothing and colors, in every prompt. Results stay mostly consistent. Regenerate any outliers one by one.
Character descriptions
Vague versus specific
Tips for better consistency
- Keep the art style the same in every prompt, for example "3D animated style" or "watercolor illustration".
- Avoid changing the character's outfit unless the story needs it.
- For close-ups, mention the face details again. For wide shots, focus on clothing and silhouette.
- Check that you have the right to use the images for your purpose, and follow the image service's terms.
The short version
Write one detailed character description, create a reference image in Gemini, turn your scenes into prompts separated by blank lines, and let AutoWhisk generate them all with the reference image attached. Regenerate the few that drift.
