How AI image generation is done: an 11-part guide made on aikizi

Eleven pages on how an AI image actually gets made, written by making them. Every picture on these pages was generated on aikizi while the page was being written, with the settings and the credits printed under it. Every price is read out of the code rather than remembered. Where something went wrong, or where another tool does it better, the page says so.

The series follows one question a person asks just before they spend money — what does this cost, which model, which size, how do I keep a face the same — and answers it with the picture that answering it produced. It cost about 200 credits to write. You can read it in any order; part 1 is the one to start with if you have not spent a credit yet.

The eleven parts

1. What one AI image costs on aikizi, and what that buys — the credit table, each plan divided into pictures a month, and why the free 5 credits will not reach a Nano Banana Pro picture.
2. Nano Banana Pro against Nano Banana 2 — one prompt, both models, 8 credits and 4. The cheaper one won on two counts we did not expect.
3. 1K, 2K, 4K: when 4K is worth 24 credits — the three sizes side by side, print sizes in centimetres, and the discovery that the three sizes are three different pictures.
4. Style references: 3 on Basic, 10 on Pro, and the six roles — what a reference carries that a prompt word cannot, and the exact sentence each role writes into your prompt.
5. Midjourney to aikizi — decode a picture you made in Midjourney, read its prompt and palette, rebuild the look on Nano Banana Pro. Where it holds and where it drifts.
6. Character consistency: realms and re-posing — a SUBJECT realm from a handful of references, new poses, a turnaround, and the parts that break first.
7. Prompts that work on Nano Banana Pro — the four-pass shape a decode writes in, why that shape suits this model, and three real gallery prompts run again.
8. Upscale or regenerate — 15 credits against 24, on the same picture, at 100%. When the upscale wins and when it cannot.
9. Voice, sound effects and music for a reel — a twenty-second soundtrack from three calls, inside the ten free minutes a day.
10. The gallery as a reference library — 16,000-odd published decodes, each with a prompt, a palette and a style DNA. How to search it and how to start from one.
11. Idea to reel: the whole workflow — one idea carried through decode, references, generation, a realm, an upscale and sound, with every step's credits summed.

What the series assumes

That you have made pictures somewhere before — Midjourney, the Gemini app, ChatGPT — and you are trying to work out whether a credit spent here buys more than a credit spent there. So the pages compare rather than introduce. There is no chapter explaining what an image generator is.

Two things the series will not do. It will not name a Midjourney style code, because those are paid and putting one on a free page gives away someone's work. And it will not talk about video generation, which aikizi retired: video and sound decode remain, generation does not.

The numbers that recur

Rather than repeat them on every page: Nano Banana Pro is 8, 16 and 24 credits for 1K, 2K and 4K. Nano Banana 2 is 4, 8 and 12. Nano Banana 2 Lite is 2, at 1K only. An image decode is 3, an upscale is 15, a download from Explore is 10 and carries the creative-use licence. Plans are Free 5 credits once, Basic $8 a month for 180, and Pro $20 for 540.

FAQ

What is this series?

Eleven pages on how AI image generation is actually done, each written by doing it on aikizi. Every picture carries the model, the size and the credits it cost, and every price is read out of the code rather than remembered.

Where should I start?

Part 1, what one image costs, if you have not spent a credit yet. If you already generate elsewhere and want the comparison, start at part 2 for models or part 5 for the Midjourney bridge.

Does the series cover video generation?

No. Video generation was retired on aikizi, so no page pretends otherwise. Video and sound decode remain, and part 9 covers voice, sound effects and music.

Are the pictures real?

Yes. Every image was generated on aikizi while the page was being written, then published before being embedded, and the caption gives the model, the size and the credits. Where we could not make something, the page says so instead.

The fastest way to read this series is with the Studio open in another tab, and the credit table beside it.