Style references on aikizi: 3 on Basic, 10 on Pro, and the six roles

Guides

Style references on aikizi: 3 on Basic, 10 on Pro, and the six roles

One prompt run with no references and with three, plus the six reference roles and the exact sentence each one writes into your prompt. What a reference carries that a prompt word cannot.

7 min readAIKIZI

Part 4 of 11 · ← previous · all parts · next → Midjourney to aikizi

A reference image carries what a prompt word cannot: the exact light, the exact palette, the exact degree of wear on a surface. Basic lets you attach 3 per generation and Pro lets you attach 10. Below is one prompt run with none and with three, so you can see what the references did — and the part most people miss, which is that on aikizi you tell each reference what job it has.

The same prompt, no references and three

The prompt was deliberately plain: a plain unglazed clay water pot standing on a worn wooden shelf, a folded cotton cloth beside it, photographic. Nothing about light, place, weather or palette.

No references: a terracotta pot and a cream waffle-weave cloth on a wooden table in a sunlit cottage interior, warm daylight from a window in a limestone wall
Nano Banana Pro, 1K, no references, 8 credits. The model picked its own world: a warm cottage interior, daylight through a window, cream linen. Its decode.
Three style references: the same pot and cloth on a wooden bench against wet mossy stone in cold blue mist, water droplets on the terracotta, coarse grey-brown linen
The same prompt, three style references attached, 8 credits. Same pot, same cloth, same shelf — different world entirely. Its decode.

The three references were the temple-lamp pictures from part 3. Nothing in the prompt said rain, stone, moss, mist or cold. The references carried all of it: the wet blue-grey masonry, the moss, the water beads now sitting on the terracotta, the cool slate palette against warm earth, even the same shallow focus falling away into haze. The cloth changed from cream waffle-weave to coarse grey-brown linen because that is what belongs in that light.

That is the thing a prompt word cannot do. 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. The reference is the only way to say it.

The part people miss: every reference has a job

Gemini has no typed reference slots. Every image goes into one flat array and the prompt text is appended last — nothing in the interface tells the model "this one is the face". It works it out from the words alone. So the only way to make a role stick is to say it in the prompt.

That is what the role badge on each thumbnail does. Pick a role and aikizi writes the sentence for you, with the reference's position number in it, above your prompt. There are six, plus custom:

RoleUse it forThe sentence it adds
CharacterA person whose face must survive"Reference n is the character — keep their exact face, features, hair and age. Do not replace them with anyone else."
StyleThe look only, not the subject"Reference n is a style reference only — take its palette, mood and rendering quality, and ignore its subject and composition."
ObjectA thing that must stay the same thing"Reference n is an object — reproduce it exactly: its shape, proportions, markings and materials."
SceneThe place and the framing"Reference n is the scene — keep its location, composition, camera angle and depth."
PoseBody position and gesture"Reference n is a pose reference — copy its body position, limb placement and gesture, nothing else."
LightingWhere the light comes from"Reference n is a lighting reference — copy its light direction, quality, contrast and colour temperature, nothing else."
CustomAnything elseYour own words — "the second character", "the pattern on the fabric".

Read the Style clause again, because it explains our pot: "ignore its subject and composition". That is why three pictures of a brass lamp produced no brass lamp. Had we labelled them Object instead, we would have got a lamp.

Three against ten, honestly

Basic carries 3 references per generation and Pro carries 10. The gap matters, but not the way a feature table suggests — ten weak references are worse than three chosen ones, because every image you attach is another vote on an average.

How manyWhat it is good forWhat goes wrong
1One exact thing: a face, a product, a specific lightThe model may copy composition too, unless you label it
3A look with room to move — the Basic ceiling, and enough for most workLittle. This is the practical sweet spot
10Several distinct jobs at once: a character, a scene, a lighting, a paletteUnlabelled, ten references average into mush; label them or use fewer

The real argument for 10 is not "more style". It is that you can hold several different roles in one generation — this face, in this place, under this light, holding this object — which three slots cannot express.

The limit worth knowing before you subscribe

If references are how you work, the reference count is the line between Basic and Pro, not the credit count. Basic is $8 for 180 credits and 3 references; Pro is $20 for 540 credits and 10. Someone doing product or character work hits the reference ceiling long before the credit one. Someone making single illustrations never hits it at all.

An honest note on this page: we showed none against three, not three against ten. The role clauses above are copied out of the code that writes them, so they are exact; the ten-reference behaviour is described from use rather than demonstrated here.

FAQ

What does a style reference do that a prompt word cannot?

It carries the exact light, palette and surface rather than a description of them. You can write "cold blue light on wet stone" and get a version of it; you cannot write that particular blue at that wetness falling off at that rate. In our test three references added rain, moss, mist and water beads that the prompt never mentioned.

How many reference images can I use on aikizi?

Three per generation on Basic and ten on Pro. Free cannot use generation with references. If references are central to your work, that count is the real difference between the plans, not the credits.

Do reference images cost extra credits?

No. A generation costs the same whether it carries no references or ten — 8 credits at 1K on Nano Banana Pro either way. You pay for the picture, not the inputs.

Why did my reference image change the subject of my picture?

Because the model was not told what job that image had. Gemini has no typed reference slots, so it guesses from the words. Label the reference Style and aikizi writes "take its palette, mood and rendering quality, and ignore its subject and composition" into the prompt for you.

Are ten references better than three?

Only if they do different jobs. Every unlabelled image is another vote on an average, so ten similar references blur rather than sharpen. Ten is worth having when you need a character, a scene, a lighting and a palette in one generation.

Attach one picture you love to a prompt in the Studio and give it the Style badge; that single step changes more than any wording you will try. For holding a face across many pictures, references alone are not the tool — part 6 uses realms instead.

Part 4 of 11 · ← previous · all parts · next → Midjourney to aikizi