🇯🇵Built in Japan for creators everywhere

INFRA for AI Creators

Image, chat, voice, transcription and X reads — one key, OpenAI-compatible.Every app drifting behind this glass runs on this API.

Get an API keySee what runs on itRead the docs →
LLM API (OpenAI-compatible)
curl -X POST https://api.aicu.ai/v1/chat/completions \
  -H "Authorization: Bearer $AICU_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-5.6-sol",
    "messages": [{"role": "user", "content": "Hello"}]
  }'
TTS API (character voices)
curl -X POST https://api.aicu.ai/v1/audio/speech \
  -H "Authorization: Bearer $AICU_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"input": "こんにちは、AICUです", "voice": "nao"}' \
  --output voice.mp3

Paste this into your coding agent

Claude Code, Cursor, Codex, OpenCode — any agent that can read a URL. The prompt below points it at llms.txt and the skill pages, then asks for a first working script. Replace nothing; your key stays in an environment variable.

prompt
Read https://api.aicu.ai/llms.txt and https://api.aicu.ai/skills before writing any code.
Use the API key from the environment variable AICU_API_KEY (never hard-code it).
Then write a small script that:
  1. calls POST https://api.aicu.ai/v1/chat/completions (model "gpt-5.6-sol") and prints the answer,
  2. speaks that answer with POST https://api.aicu.ai/v1/audio/speech (voice "nao") and saves out.mp3,
  3. generates one image with POST https://api.aicu.ai/v1/images/generations (model "gpt-image-2.5-flare", size "1024x1024", quality "low") and saves out.png.
Print the AP cost of each call from the response (usage / ap_cost / X-AICU-AP-Cost). Stop and ask me if any call returns 402 or 403.
  1. 1Get a key at /dashboard/keys and export it as AICU_API_KEY
  2. 2Paste the prompt. The agent reads llms.txt and the skills, then writes the script
  3. 3Your first call grants $10 of AP; everything is metered per request on the dashboard
llms.txtAgent skillsOpenAPIGetting started

Draw your own character in one request

“LuC4 wants an illustration of LuC4 for a blog post. Can the Character API draw it with GPT-Image-2.5? What do I have to send?”

Yes. One JSON body to POST /v1/images/generations: a picture of the character, the scene, and the model. The image on the right is the unedited output of exactly this request.

request body
{
  "model": "gpt-image-2.5-flare",
  "character": "luc4",
  "reference_image": "<base64 of one picture of LuC4>",
  "prompt": "Anime style. LuC4: 20-year-old kind boyfriend, round chin, 1boy, solo, male focus, (streaked hair), red short hair with a light highlight streak in the fringe, red eyes, hoodie, short hair, hair between eyes. Scene: sitting at a sunny cafe table, typing a blog post on a laptop, looking up at the viewer with a big encouraging smile and a thumbs-up. Warm bright colors, soft lighting. No text.",
  "size": "1536x1024",
  "quality": "high",
  "visibility": "public"
}

Or hand it to your coding agent

prompt
Read https://api.aicu.ai/llms.txt first. Draw LuC4 for my blog post: POST https://api.aicu.ai/v1/images/generations with model "gpt-image-2.5-flare", character "luc4", reference_image = base64 of ./luc4.png, size "1536x1024", quality "high", and the scene I describe next. Use the key in AICU_API_KEY. Save data[0].b64_json as cover.png and print ap_cost.
LuC4 at a sunny cafe table, typing on a laptop and giving a thumbs-up — generated with gpt-image-2.5-flare via api.aicu.ai
  • gpt-image-2.5-flare · 1536×1024 · high
  • 5,100 AP (about $0.51) · 23 s
  • © AICU Inc. / LuC4 — credited automatically

Reference: one existing picture of LuC4. The response carries data[0].url, data[0].b64_json and ap_cost; the cost is also in the X-AICU-AP-Cost header. Generated 2026-09-14.

model
gpt-image-2.5-flare (omit it and you get gpt-image-2; -sunburst is the slower, finer variant). The aliases gpt-image-2.5 and flare also work.
character
Who the picture is of — and who gets credited. A registered slug needs no reference image; your own character's name needs one.
reference_image
One picture of the character, base64. The look comes from here, so the prompt only has to describe the scene.
prompt
The character's official appearance words, then the scene. The reference sets the look; the words keep details like the highlight streak. Write “No text” if you don't want lettering.
size / quality
These set the price: 1536×1024 high is 5,100 AP; 1024×1024 low is 700 AP. Only HTTP 200 is billed.
visibility
public lists the result in the gallery with the credit line. private keeps nothing and costs twice as much.
Registered charactersCharacter APIImage APIGetting started

Better than calling OpenAI or ElevenLabs yourself

Same upstream models, plus the parts you would otherwise build and tune yourself: character consistency, tuned Japanese voices, a budget that stops at zero, and one key for everything.

Characters stay themselves

Pass reference images and GPT-Image-2.5 keeps face, hair and outfit. In our public AiCutyBench, 6 characters × 10 images passed 60 of 60. Multiple references per request; royalty ledger for licensed characters.

Voices already tuned for Japanese

The AiCuty character voices are tuned on ElevenLabs by our team — you call them by name (`voice: "nao"`), not by voice ID. Same text, same settings: cache hit, no charge.

A budget that stops at zero

Prepaid AP (1 USD = 10,000 AP). Only HTTP 200 is billed, errors and cache hits are free, and when the balance hits zero the key stops — no surprise invoice, no subscription.

One key, and the things a single vendor won't give you

Chat (OpenAI, Gemini, Kimi on Sakura AI Engine), images, voices, transcription and X post reads behind one base URL, with Japanese docs and support, and inference in Japan where it matters.

The money already flows the right way

AICU holds login and billing, so developers focus on building and character rights-holders get paid every time their character is used.

  1. 1

    End user

    Signs in, tops up AP as needed, and is charged only for what is used. When the balance reaches zero, it simply stops.

  2. 2

    App developer

    No auth, billing or invoicing to build. One API key reaches chat, voice and image generation, and revenue is recorded as usage.

  3. 3

    Character rights-holder

    Every use of your character is appended to a ledger and priced by your rate. Who used what, and when, is verifiable afterwards.

  4. 4

    Upstream cost

    What we pay the model providers. Which model cost what is traceable from the usage record.

Developers build

Login, billing, usage history and rights handling are provided. No per-provider keys, no home-made invoicing.

Rights-holders don't miss a use

Each generation is appended to an append-only ledger. Totals come out per payee, so they are already the unit of settlement.

Card details never pass through this API

Payments complete on the payment provider's side. This API holds your prepaid AP balance and the record of what was used for what — nothing else.

Sound familiar?

An API for hackathon teams, indie developers and agent builders

🏃

I need an MVP by tonight

One key covers LLM, TTS, transcription and image generation

⏰

No time to sign up for five vendors

OpenAI, Gemini, OpenRouter, Sakura AI Engine and ElevenLabs are already wired

🤖

I just want to hand my agent a SKILL.md URL

Agent skills for Claude Code and Cursor at /skills

💸

I won't pay a subscription for a hobby site

No monthly fee — prepaid AP, spent only on what you use

🔧

I don't want to babysit the API layer

Runs on Cloudflare's edge; you keep building

📊

I want to see every request

Every call is recorded with model, tokens and AP on your dashboard

Built on it ourselves

Every AICU service runs on this API. We sell what we use.

YOUKAI

Web manga viewer. Per-page billing and rentals go through this API.

Character IP

Reference-image generation with royalties split to the rights-holder.

AICU Slack bots

Articles, covers and X quote cards from a Raspberry Pi — no per-provider keys on the device.

Article & cover generation

Blog covers with GPT-Image-2.5 and Nano Banana.

More services are added as they go live.

Services

OpenAI-compatible LLM, character TTS, transcription, image generation and X reads — one key

LLM API

Beta · metered

OpenAI-compatible chat completions. Every model we serve behind one interface (the list below is synced with the API).

gpt-6-astragpt-5.6-solgemini-3.6-flashkimi-k2.7-code
LLM API details

TTS API

Beta · metered

ElevenLabs-quality multilingual speech with the official AiCuty character voices — one key, OpenAI-compatible. Same input, same settings: cache hit, free.

elenameiminanaosakimarsha
TTS API details
🎨

Image API

Beta · metered

GPT-Image-2.5 (Flare / Sunburst) and Nano Banana with reference images, character presets and a royalty ledger. OpenAI SDK compatible at /v2.5.

gpt-image-2.5-flaregpt-image-2.5-sunburstnano-banana
Image API details

Transcription API

Beta · metered

Speech-to-text with whisper-large-v3-turbo, 99 languages, strong on Japanese. Video, streams and meetings. OpenAI-compatible.

whisper-large-v3-turbo99 languagesinference in Japan
POST /v1/audio/transcriptions
🐦

X read gateway

Beta · metered

Read posts (text, author, date, attached media), profiles, timelines and search with your AICU key — no X developer account.

GET /v1/x/status/:idX gateway skill
</>

OpenAI-compatible

Your existing OpenAI SDK works as is

⇄

Multi-provider

OpenAI, Gemini, OpenRouter, Sakura AI Engine and ElevenLabs

🌐

Cloudflare edge

Low latency from a global network

🇯🇵

Japanese-first support

Docs and support in Japanese and English

Models and pricing

Beta is open and metered. Prices are public, revised weekly, and increases are announced seven days ahead.

AICU API is in beta until Oct 31, 2026 — prices are unchanged by the beta calendar.

Getting started
Sign up, issue a key, and your first API call grants you $10 worth of AP. Trial keys run for 14 days.
Beta calendar
Sep 23, 2026 public demo at Generative AI Expo · Oct 7, 2026 AP packs on sale · Oct 14, 2026 last beta keys · Oct 21, 2026 production (new keys need a card) · Oct 31, 2026 beta keys expire.
Pricing
Per-model rates are published and revised weekly (1 USD = 10,000 AP).See the rates
How you pay
Prepaid AP. Only HTTP 200 is billed; errors and cache hits are free. Auto-recharge is optional.
Support
Reply to any mail from api@aicu.ai, use the contact form, or join the community on Discord.Discord

Accounts may be suspended without notice in the event of abuse or misuse.

What is coming next is on ournews page. For reselling or high-volume use, pleaseget in touch.

Prepaid, spent only on what you use

No subscription and no unused allowance to lose. Usage is recorded per request and visible on the dashboard at any time.

The record grows with use

Which model cost what, per request. Service status is public.

Current models

The newest models from OpenAI, Google and Sakura AI Engine through one OpenAI-compatible interface. Additions show up as they land.

One key, one endpoint

Chat, character voices, transcription, image generation and X reads from the same key and the same base URL（https://api.aicu.ai/v1）. No per-provider keys.

AICU members get daily seed credit

Sign in and points arrive every day; writing a post or sending emo adds more. Spend them on the API, so prototypes cost you nothing out of pocket.

Just sign in
Daily points, granted even on days you don't use them
Write and earn
Publishing a post or sending emo adds points, up to a daily cap
Use characters too
Generate with registered characters; royalties are split to the rights-holder

Need more? Buy AP, or redeem a campaign coupon code.

Join (free)See Character IP
モデル提供元最大トークン特徴
deepseek-v3OpenRouter8,192汎用、コスパ良好
gemini-flashOpenRouter8,192軽量・低コストの Gemini。大量処理・定型タスク向け（原価×3）
llama-3.1-8bOpenRouter8,192軽量
llama-3.1-70bOpenRouter8,192高品質
qwen3-32bOpenRouter32,768大規模コンテキスト
gpt-4oOpenAI16,384OpenAI 高品質マルチモーダル（Alpha無料）
gpt-4o-miniOpenAI16,384OpenAI 軽量・高速（Alpha無料）
sakura-kimiさくらAI8,192国産推論 さくらのAI Engine 上の Kimi-K2.6（マルチモーダル・reasoning）。推論は国内・学習に不使用。
gpt-5OpenAI128,000最上位。推論が要る作業向け。入力は gpt-4o の半額、出力は同額（旧世代。新規は gpt-5.6 系を推奨）
gpt-5-miniOpenAI128,000gpt-5 の 1/5。gpt-4o-mini より賢く、gpt-4o より大幅に安い（旧世代。新規は gpt-5.6 系を推奨）
gpt-5-nanoOpenAI128,000最安の GPT-5。分類・抽出など定型の作業向け（旧世代。新規は gpt-5.6 系を推奨）
gpt-5.6-solOpenAI128,000最上位。難しい推論・長い文脈向け
gpt-5.6-terraOpenAI128,000sol の半額。日常的な作業の主力
gpt-5.6-lunaOpenAI128,000軽量で速い。対話・要約向け
gpt-5.4-miniOpenAI128,000安価。定型の生成や抽出向け
gpt-5.4-nanoOpenAI128,000最安。分類・判定など短い作業向け
kimi-k2.7-codeさくらAI8,192さくらのAI Engine 上の Kimi-K2.7-Code。コーディング特化・マルチモーダル（パブリックプレビュー）。OpenCode/Cline 等の OpenAI 互換ツールから利用可。国内推論・学習不使用。
gemini-3.6-flashOpenRouter8,192最新の Gemini 3.6 Flash。高品質マルチモーダル。高度な推論・長い文脈向け（原価×3）
gpt-6-astraOpenAI128,000次世代の最上位。難しい推論・長い文脈向け（gpt-5.6-sol の上位）

現在 19 モデルを提供中。この一覧は GET /v1/chat/models と常に同期しています。
利用可能モデルと価格については随時見直しを行います。