Blog
An AI image generator where you upload the style reference
Upload a picture, badge it as a style reference, and the generator takes its light, palette and surface onto your subject. One prompt run with none and with three, and the six roles an upload can have.
Upload a picture, tell the generator that it is a style reference and not a subject, and it will take that picture's light, palette and surface onto whatever you write. On aikizi you attach up to 3 references on the $8 plan and 10 on the $20 one, and each upload gets a role badge that writes an explicit instruction into the prompt. That badge is the part most tools do not have, and it is the part that makes the upload work.
What an uploaded reference does that a word cannot
We ran one plain prompt twice — "a plain unglazed clay water pot standing on a worn wooden shelf, a folded cotton cloth beside it, photographic" — with nothing else changed but three uploaded references.
Read the prompt again and then look at the picture. Nothing in those eleven words said dusk, rain, cobblestones, a storm, or a warm lamp against a cold sky. The upload — a photograph of a wet South Indian street at blue hour — carried every one of them, while the subject stayed the bicycle the words asked for. That is the whole mechanism: the words choose what, the upload chooses how it looks.
You could write "cold blue morning light on wet stone" and get a version of it. You cannot write this particular blue, at this wetness, falling off at this rate. That is what the upload is for.
The six roles an upload can have
| Role | Use it for | What it tells the model |
|---|---|---|
| Style | The look only | Take its palette, mood and rendering; ignore its subject and composition |
| Character | A face that must survive | Keep their exact face, features, hair and age |
| Object | A thing that must stay itself | Reproduce its shape, proportions, markings and materials |
| Scene | The place and framing | Keep its location, composition, camera angle and depth |
| Pose | Body position | Copy its body position, limb placement and gesture, nothing else |
| Lighting | Where light comes from | Copy its direction, quality, contrast and colour temperature |
The reason the badge matters: Gemini has no typed reference slots. Every uploaded image goes into one flat array with your prompt appended last, and nothing in the interface says "this one is the face". The model infers it from the words — so the role writes the words. This is also the answer to "why did my style reference change my subject": it was not labelled, and the model guessed.
Uploads, plainly
- JPG, PNG or WebP, up to 50 MB.
- References cost nothing extra. A generation is 8 credits at 1K on Nano Banana Pro whether it carries none or ten.
- 3 per generation on the $8 plan, 10 on the $20 plan.
- Ten unlabelled references are worse than three chosen ones — every image is another vote on an average. Ten is worth having when they do different jobs.
FAQ
Which AI image generator lets me upload a style reference?
aikizi does, with up to 3 uploads per generation on the $8 plan and 10 on the $20 plan, and each upload carries a role badge saying what job it has. JPG, PNG or WebP up to 50 MB.
Is there a limit on how many pictures I can upload?
Three per generation on the $8 plan and ten on the $20 plan, JPG, PNG or WebP up to 50 MB each. The uploads themselves add nothing to the price of the generation.
Why did my uploaded reference change the subject of my picture?
Because it was not labelled. The model has no typed slots and guesses each image's job from the words. Give the upload the Style role and it is told explicitly to ignore that picture's subject and composition.
How many reference images should I upload?
Three chosen ones beat ten similar ones, because every unlabelled image is another vote on an average. Upload ten only when they do genuinely different jobs — a face, a place, a light, a palette.
What does a style reference carry that a prompt cannot?
The exact light, palette and surface rather than a description of them. In our test three uploads added rain, moss, mist and water beads that the prompt never mentioned.
Attach one picture you like to a prompt in the Studio and give it the Style badge — that single step changes more than any rewording. The full comparison of 3 references against 10 is part 4 of the generation series.