VallottonDocs

Vallotton

Your AI drawing studio

Vallotton is, at heart, an AI drawing workflow. Its goal is to drive a tag-based image model (NovelAI V5) through natural conversation, while solving some of the problems such models have on their own, such as keeping a style consistent and reusing original characters, and adding planning, continuous generation and self-reflection on top.

Quick start User guide Deployment GitHub
The painter Vallotton at work in her studio late at night

What Vallotton is

Vallotton is a drawing studio. It looks and works much like ChatGPT: every piece of work is a conversation. Vallotton paints what you ask for, edits pictures already in the conversation, discusses them with you, or simply chats. (Early in the month: GPT Astra, recite the multiplication table for me. Late in the month: you know, Vallotton is not bad at programming either.)

Underneath, Vallotton is a workflow that joins a large language model (DeepSeek V4 flash by default) to NovelAI V5. The core loop is simple: the language model translates what you say into the Danbooru-style prompt that NovelAI understands, and NovelAI paints it. Done naively, that runs into trouble fast: the same character looks different in every picture, the details of her clothes drift, the style wanders, and there is no way to change just one part of a picture. Vallotton adds a fairly involved process around the loop to avoid these problems as far as it can.

Understanding this core loop matters, because what Vallotton can do is bounded by what its language model and NovelAI can do. It is very good at characters from popular anime franchises and supports artist-style tags1, because NovelAI was trained for exactly that; NovelAI also has dedicated landscape and furry datasets, and Vallotton does well there too. What it cannot do is produce photorealistic images the way ChatGPT (with gpt-image-2.5 behind it) can.

What you need to know and have before you start

NovelAIA NovelAI subscription and its persistent token. Vallotton's pictures use the subscription's allowance.
Language modelA DeepSeek API key (Grok is supported too; see the deployment guide).
Running locallyTo run Vallotton on your own computer you need a Windows, macOS or Linux machine with an internet connection. The start script downloads a Python runtime by itself if the computer has none.
Deploying to a serverTo run Vallotton on a server you need a Linux machine reachable from the internet, and a domain name.

Pick the guide that fits what you want to do.

I

Quick start

Start Vallotton on your own computer and paint your first picture.
II

User guide

How to use everything, and the words for it: editing pictures, the two-layer library, variants and tag lines, reflection and continuous generation, the inspector.
III

Deployment

Put Vallotton on a server so you can reach it from anywhere, and configure it through environment variables.

1Drawing characters from a particular franchise, or imitating the personal style of a living artist, is contested in many places: legally it may touch intellectual property, and ethically many artists have said clearly that they do not want AI imitating their style. Vallotton exposes this part of NovelAI's ability as it is, with no filter added and no judgement made on your behalf. Where the pictures go, whether they are published or sold, and whether an artist's name is attached, is for you to weigh against the rules and the ethics that apply, and NovelAI's terms of service. ↩

* Vallotton is released under the Apache-2.0 license. NovelAI, DeepSeek and xAI are services of their respective companies; the accounts, the costs and the generated content are the user's own responsibility.