I’m experimenting with Grok’s image generation for a small comic project. I can get a character I like in one image, but when I ask for a new pose or setting, the face, clothes, and proportions tend to drift.

What has worked for maintaining character consistency across separate generations? Do you reuse a reference image, keep a fixed description, or build each prompt from the previous result? I’m curious how people handle this without making every panel look like the same pose with a different background.