Keep the same AI character across different scenes
Recognition depends on specific visible traits. Separate identity from action instead of relying on a name. Text-based image generation can support resemblance, but it cannot guarantee exactly the same person.
A guide by OpenPix
Create an unchanged identity block
Describe age range, face shape, hair, eyes, clothing and one distinctive small feature. Repeat a few unambiguous traits word for word in every prompt. A name only helps you organize the series.
- Face and hair: oval face, green eyes and a straight black chin-length bob.
- Clothing and palette: navy work jacket, mustard scarf and black trousers.
- Distinctive feature: a small triangular silver brooch on the left lapel.
- Consistent presentation: the same style, similar framing and similar lighting.
Change the action and prop separately
Copy the identity block unchanged. Replace only the action sentence: holding a map becomes holding a lantern. Say which hand holds the object. Initially keep background, viewpoint and light identical so identity is easier to compare.
Once a variation works, change a second factor such as location or camera angle. Large viewpoint changes can conceal identifying traits.
What a text prompt can and cannot control
Use the same OpenPix style and image format. Initially turn off prompt refinement for comparisons, or carefully inspect the rewritten text. Save successful images to your favorites and compare face, hair length, clothing and brooch side by side.
Reference images, fixed seeds or specially trained models may offer additional control in suitable tools. This guide does not promise those controls in the current studio. A seed alone cannot guarantee identity when the action changes. Exact repeatable brand characters can require selection and editing.
Detailed example prompts
Scene A: Mira holds a map
Mira, an adult woman around 30 with an oval face, green eyes and a straight black chin-length bob. Navy work jacket, mustard scarf, black trousers, small triangular silver brooch on the left lapel. She stands in a quiet old railway hall holding a folded paper map in her right hand at chest height. Medium three-quarter view, soft light from a tall window on the left, muted warm colors, natural skin texture, photographic style.
The first part defines identity. Action and presentation follow. Brooch, scarf and hair provide visible comparison points.
Scene B: Mira holds a lantern
Mira, an adult woman around 30 with an oval face, green eyes and a straight black chin-length bob. Navy work jacket, mustard scarf, black trousers, small triangular silver brooch on the left lapel. She stands in a quiet old railway hall holding a small unlit brass lantern in her right hand at chest height. Medium three-quarter view, soft light from a tall window on the left, muted warm colors, natural skin texture, photographic style.
Only the prop changes. The lantern is unlit so it does not introduce another light source or change the face’s appearance. This makes resemblance easier to assess.
Best practices for your images
- Keep the identity block separate from the editable action sentence.
- Compare visible traits rather than only the overall impression.
- Keep style and image format consistent and expand the series gradually.
- Use reference material only when you can responsibly use the images.
Common mistakes and limitations
- The same name or same person as before is not enough: a new text prompt does not automatically include the previous image.
- Do not expect a guaranteed face copy. Simplify the scene and correct specific traits when they drift.