About the Models

About the models

We currently use our Gen 7.2 model, and unlike others, we don't resell model access. All our models originate from the open-source community or our own R&D — we then disassemble, modify, fine-tune and optimize them to align with our legacy and 2DAI❤️ART lines. They also run on our own private cloud network.

Gen8 Preview video — the first public release of our next video engine is live behind videoModel: 'next': clips with sound, from a first frame plus up to 6 references that stay consistent, and in the studio an audio reference (music, or a voice to clone) with TIXI writing the scenario. fast quality and 1 to 10 s from Holder; every preset and clips up to 20 s at Founder. Creations it renders are stamped aiModel: 'gen8.0' (raw.videoModel: 'next'), and their queue state carries videoModel: 'next'.

Gen8 Preview images — the image side of the same generation is live behind imageModel: 'next' on generate.image: up to 6 references kept natively, legible text in the picture, long literal prompts honoured, the same quality ids (Fast to Max from Holder, Ultra from Believer, Ultimate from Supporter), priced on render time, roughly four to seven times the Gen 7.2 preset; me().canUseImageNext and me().qualities.imageNext say what the account may submit. Creations are stamped aiModel: 'gen8.0' with imageModel: 'next'.

What that means in practice

• Same weights everywhere. The images and videos this SDK produces come off the exact same Gen 7.2 and Gen8 stack as the ones you get from the 2DAI studio at 2dai.io — no watered-down "developer tier", no throttled model.
• In-house LLM rewriter. Prompt enhancement (the "TIXI" pass that expands short prompts) runs on a proprietary rewriter we operate — not a re-sold third-party LLM.
• Private cloud. Both the image/video generator and the rewriter run on network infrastructure we own end-to-end. Your prompts and outputs never transit through a public multi-tenant model API.

Where the model version shows up

The model version is stamped on every generated Creation:

const gen = await client.generate.image({ prompt: 'a red origami fox' });
gen.aiModel; // 'gen7.2'

It also comes back on client.finance.tier() implicitly (tier is versioned against the current active model) and on the studio-side dashboard.

Versioning

The model version bumps forward when we retrain — Creation.aiModel on the returned record is always authoritative. The SDK follows its own semver independently.

The related MCP server

If you'd rather let an agent (Claude Desktop, Cursor, Cline, Windsurf) drive your 2DAI account instead of scripting the SDK yourself, the same stack is exposed as an MCP server: 2dai-mcp-server. Same weights, same private cloud, same rules.