inkloom README

The project's README from github.com/Inkloom-art/inkloom. As one markdown file: README.md.

Inkloom

Specialised AI models for logo design.

inkloom.art

What we are building

A logo is not a picture. It is a constructed object with rules: a mark that holds at sixteen pixels and on the side of a building, letterforms spaced by eye rather than by metric, clear space derived from the mark's own geometry, and lockups that still read when one of them is all you have room for.

General image models do not work this way. They produce something logo-shaped — a plausible arrangement of marks with no construction behind it, no reasoning about the business, and nothing you can hand to a printer or a sign maker.

Inkloom is building models that construct a mark the way a studio does, as a sequence of decisions that can each be explained:

Stage What it produces
Brand analysis Turns a description of a business — sector, audience, tone, competitors — into concrete constraints: stroke weight, width, geometry, counter shape, which symbol families fit
Typography Selects and fits letterforms against those constraints, then does the work that makes a wordmark: optical spacing, kerning at display size, a custom ligature where the name needs one
Symbol construction Composes geometric primitives under construction rules — shared radii, tangent junctions, consistent terminals — so the result is built rather than sampled
Composition Optical alignment rather than mathematical centring, clear-space ratios taken from the mark itself, and the lockup variants a brand actually needs

The output is meant to be a specification, not a bitmap: a mark you can describe, defend and reproduce.

Where we are

Early access is open at inkloom.art. Create an account, redeem a code, and credits are reserved against your account.

Generation is not live yet. We would rather say that plainly than imply otherwise: every page in the product says so, credits are described as reserved rather than spendable, and the feature flags that would switch generation on default to off and are not togglable from the console — because enabling a flag whose feature does not exist exposes a broken surface rather than a feature.

What runs today is the platform the models will ship on. Accounts and authentication, the credit ledger, the access-code system, the operations console, and the machinery around them: backups that are restore-tested rather than merely taken, an alerting pair where each half watches what the other cannot see, and a deployment path that refuses to migrate a database whose identity has not been confirmed.

What comes next

Generation itself, then the things that only make sense once it exists: export in the formats a designer and a printer each need, brand kits, revision history on a mark, and paid plans. None of it is claimed as present until it is.

About this repository

This source is published so the engineering can be read and audited — in particular the security and data-handling claims we make. It is not a distribution: see LICENCE.

Found a security issue? SECURITY.md says where to send it and what to expect. Please do not open a public issue.

Operational documentation — deployment, incident response, environment and runbooks — is kept internal. Source comments occasionally point at it by filename; that is a reference for the people who run the service, not a broken link. It describes how the service is operated, which is of no use to a reader and of some use to an attacker.

Built on Cloudflare Workers, Postgres and React Router, with the application and its API served from one origin — which is what makes the session cookie first-party and removes cross-origin handling entirely.

The test suite runs against a real database, a real browser and a real mail server rather than mocks of any of them, because the guarantees that matter here are transaction guarantees and a mock cannot have one. The test that matters most fires twenty-five simultaneous redemptions of a single code at a real database and asserts that exactly one redemption, one ledger entry and one balance exist afterwards.

Contact
General [email protected]
Security [email protected] — please read SECURITY.md first

Copyright © Inkloom. All rights reserved.

inkloom (Inkloom-art/inkloom)

Searching Database…

1.8K ·GitHub
Inkloom
这是什么？Inkloom 构建专门的 AI 模型，把标志设计成一系列可解释的阶段——品牌分析、字体排印、符号构建和构图——而不是直接生成一张图片。
这和一般的 AI 图像生成器一样吗？不一样——通用图像模型只会生成一张看似像标志的图片，背后没有任何构建逻辑，而 Inkloom 的流水线通过有理有据的阶段构建标志，产出可以直接交给印刷厂的结果。
怎么开始使用 Inkloom？访问 inkloom.art，创建账号，然后兑换访问码；完成后额度就会预留到你的账号。
需要邀请码才能进入吗？需要，产品处于抢先体验阶段，必须先兑换访问码，额度才会预留到你的账号。
现在真的能用它生成标志吗？还不能——生成功能明确尚未上线，产品里的每个页面都这么标注，生成功能的特性开关默认关闭，也无法从控制台打开。
平台如何处理账号数据和会话？应用及其 API 由同一个 Cloudflare Workers 源站提供，后端是 Postgres，这样会话 Cookie 保持第一方，无需跨域处理；兑换逻辑则针对真实数据库做了负载测试以保证正确性。
我可以自己部署或把它分叉为开源项目吗？不可以——公开源代码只是为了让工程和安全方面的声明可以被审计，项目明确声明这不是发行版本，版权保留所有权利。
Inkloom 的方式和一次性 AI 图像生成器有什么区别？Inkloom 把标志拆解为一个设定约束（笔画粗细、几何形状、符号族）的品牌分析模型、在约束内进行构建的字体与符号模型，以及一个处理视觉对齐和留白规则的构图引擎——最终产出的是一份站得住脚的规格说明，而不是一张位图。
在生成功能上线之前，平台测试有多严格？测试套件针对真实的数据库、真实的浏览器和真实的邮件服务器运行，而不是用模拟对象；其中关键测试会同时发起 25 次对同一个访问码的兑换，并验证最终恰好只有一次兑换、一条账本记录和唯一的余额结果。
生成功能上线后还计划了哪些特性？面向设计师和印刷厂的导出格式、品牌套件、标志的修订历史以及付费方案——这些目前都还不存在。
What are similar projects?react-starter-kit, ip-as-logo-skill, clevis, logo-design-skill, open-saas, terraink, simple-icons, agentic-inbox
Which topics is it in?branding, design-systems, security-audit, creative-tools, cloudflare-workers, typography, saas, design, authentication, artificial-intelligence, testing, postgresql
Creator
Inkloom-art
Size
3.7 MB
Stars
1.8K
Updated
CommitSep 26
Age
27d
Commits · 90 Days