Need for Character Consistency and Style Locking in Image Generation

ChatGPT
2

Hi welcome to the community!

You can use multi-turn techniques and use a DNA Template before each scene:

DNA Template

All images must be wide size 3:2 aspect ratio watercolor anime style images.
In this session, we are creating a sequence of visual scenes titled ‘The Harmony of Three’.
The focus is on capturing subtle emotional moments between triplet sisters Hi, Fu, and Mi.
Each image builds the atmosphere of quiet connection and narrative continuity.
Keep all visual elements stable across images: characters, style, tone, and background environment.
All scenes take place in a serene coastal village with sakura trees, paper lanterns, and pastel-toned skies.

This scene features three identical triplet sisters named Hi, Fu, and Mi.
All three girls have identical appearances:

  • Light tan skin, round face with soft cheeks, and large almond-shaped hazel eyes.
  • Straight, shoulder-length silky black hair with front bangs.
  • They are average height, slender with graceful posture, and have flat stomachs and elegant proportions.
  • They wear traditional pastel pink dresses with slight variation:
    • Hi wears a pink hair ribbon
    • Fu wears a silver pendant
    • Mi holds a small light brown diary
  • Their expressions are calm and emotionally reflective.
    Their appearance must remain perfectly consistent in every image.

Render in soft anime watercolor style, with gentle linework and muted pastel tones.
Use ambient diffused lighting, with no hard shadows or saturation changes between scenes.
Backgrounds should use soft, desaturated colors with a gentle painterly effect.
Maintain a cinematic horizontal composition with medium-wide shot unless otherwise specified.

I will provide scenes. and you will create one by one, and each one is wide size 3:2 aspect ratio watercolor anime style images.

Some related topica:

A test for "Multi-Turn Generation" with 4o Image Generation

Prompt to make exactly same image but different pose - #23 by polepole

Create images with text and without any spelling mistakes - #5 by polepole

Your DALL-E problems now solved by GPT-4o multimodal image creation in ChatGPT? - #44 by polepole

4o ImageGen: Share your best pictures - #132 by polepole

1 Like