Prompt: Create a Character Identity Reference Sheet
Turns one or two photos of a person into a labelled multi-angle reference sheet for keeping a face consistent in later AI images. Works with Gemini or ChatGPT when you attach the photo; use only photos you have permission to use.
Using the attached reference photo(s), create ONE image: an identity reference sheet of this exact person, laid out as a labelled {{GRID — e.g. 3x3}} grid, {{ASPECT RATIO — e.g. 16:9}}.
Panels (each labelled; "left" and "right" mean the subject's own left and right):
1. Front
2. Three-quarter left
3. Left profile
4. Rear three-quarter left
5. Straight back
6. Rear three-quarter right
7. Right profile
8. Three-quarter right
9. Close-up front portrait
Identity: reproduce the person exactly as photographed, including face shape, hairline, brows, eyes, nose, lips, cheeks, jaw, chin, ears, skin tone and texture, facial hair and hairstyle. Keep natural asymmetry and apparent age unchanged. Infer unseen angles conservatively from the visible structure.
Constraints:
- Every panel must show the same person, age, hairstyle and {{CLOTHING — e.g. plain grey crew-neck t-shirt}}, framed as {{FRAMING — e.g. head-and-shoulders}} at the same camera distance.
- Neutral expression, relaxed posture, soft even studio lighting and a plain {{BACKGROUND — e.g. light grey}} background in every panel.
- Do not beautify, idealise or symmetrise the face, and do not change the hairstyle or facial hair.
- Realistic photograph with natural skin texture and sharp facial detail; no illustration, anime or CGI look.
Before finalising, check that the face is identical in all panels. Return only the image, with no accompanying text.
Example
Input
GRID — e.g. 3x3: 3x3 ASPECT RATIO — e.g. 16:9: 16:9 CLOTHING — e.g. plain grey crew-neck t-shirt: plain grey crew-neck t-shirt FRAMING — e.g. head-and-shoulders: head-and-shoulders BACKGROUND — e.g. light grey: light grey
Output (a written description, not a generated image)
What a good result should look like: A single 16:9 image arranged as a labelled 3x3 grid of nine panels, each showing the same person's head and shoulders at identical camera distance. The panels are labelled Front, Three-quarter left, Left profile, Rear three-quarter left, Straight back, Rear three-quarter right, Right profile, Three-quarter right, and Close-up front portrait. In every panel the person wears a plain grey crew-neck t-shirt, has a neutral expression and relaxed posture, and is lit with soft, even studio light against a plain light-grey background. Facial features — face shape, hairline, eyes, nose, jaw and hairstyle — read as consistent across all nine panels, with natural asymmetry preserved rather than smoothed into a symmetrical, idealised face. The rear and profile panels show plausible, conservatively inferred hair and ear shapes consistent with the visible front and three-quarter views. The overall look is a realistic photograph with visible skin texture, not an illustrated or CGI style.
Why it works
- Fixing the grid layout and labelling nine specific angles removes the ambiguity a vague 'different angles' request leaves open.
- Locking clothing, framing, lighting and background as constraints means only the angle varies across panels, isolating the one thing the sheet needs to show.
- The final identical-face check catches drift between panels before the image is returned, not after.
Why the basic version fails
“Make a reference sheet of this person from different angles” gives you a random collage, a single portrait, or a slightly different person in every view. Nothing tells the model how many panels to draw, how they are arranged, what stays fixed between them, or what “left” means.
What this prompt fixes
It fixes the layout (one labelled grid), names every angle, states once what must be identical across panels, and fixes the clothing, framing, lighting and background so that only the angle changes. The final check asks the model to compare panels before it returns the image, which catches drift in the face or hair.
Tips per model
- Gemini and ChatGPT need the photo attached to the same message. Two or three photos work better than one, ideally including one side-on shot, because the back and profile panels are inferred from what the model can see.
- Midjourney does not accept a photo this way; use its own reference feature (
--crefor--oref) and drop the “attached photo” wording. - The rear views are informed guesses, not facts. Check the back of the head and the ears against real photos before treating the sheet as a reference.
Use it responsibly
Use photos of yourself, or of people who have agreed to it. Some image tools will refuse to work from a photo of a real person, and that is expected.