# Sumi 2.5: new models, character voices and better layouts on every screen

Published 2026-11-11 by Sumi Studio Team.

Source: https://sumistudio.ai/en/blog/sumi-2-5-update

Sumi Studio 2.5 is live. This update adds six new models, lets you give each character element its own voice, fixes several canvas and editor problems, and makes Sumi Studio easier to use on phones, tablets and ultrawide monitors. Here is what changed.

## New models: Kimi K3, Qwen Image 2.1-turbo, Grok Imagine Video 1.5 Lite and Qwen-Audio TTS

Kimi K3 and Kimi K3.1 Pro are now available for the Sumi agent and in Text nodes. K3 can search the web, and K3.1 Pro supports structured output.

For images, we added Qwen Image 2.1-turbo. For video, Grok Imagine Video 1.5 Lite.

There are two new voice models: Qwen-Audio 3.1 TTS Plus and Qwen-Audio 3.1 TTS Next. Before TTS Next starts, it estimates the cost from your prompt and reserves those credits.

Gemini 3 and later versions now have an Output length setting, which is useful for long scripts and other long text. If you don't pick an output length, the default is used.

[Grok Imagine Video 1.5 Lite next to Grok Imagine 1.5 and Veo 3.1 Lite, generating the same rally car shot.](https://sumistudio.ai/brand/blog-sumi-2-5-models.mp4?v=2026-10-21.1)

## Give your characters a voice

You can now bind a reference voice to a character element. Any generation that uses that character will use the same voice, so it sounds like the same person from one shot to the next.

![Creating a character element: the Voice and Character voice reference settings, and the dialogs for cloning your own voice or designing one from a description.](https://sumistudio.ai/brand/showcase-sumi-2-5.jpg?v=2026-10-21.1)

To add one, go to Assets, open [Elements](https://sumistudio.ai/en/assets/elements), create a new element and set the category to Character. Under Character voice reference, click Choose or upload audio and add a clip between 2 and 10 seconds long. If you don't have a clip yet, set Voice to My cloned voice, then click Clone my voice to record your own or Design a voice to create one from a short description. Seedance, Wan and Hailuo use the reference clip when you @ the character. For Kling, pick a Kling-cloned voice under Voice.

How many voices you can clone and how many custom voice slots you get now depends on your plan.

## Sumi through MCP

If you use Sumi Studio from an agent over [MCP](https://sumistudio.ai/en/connect), the agent now tells you how to renew or top up when your subscription has expired or you're low on credits. Once that's done, you can keep working in the same session.

## New product page

The site has a new [product page](https://sumistudio.ai/en/?landing=1) that shows what you can make with Sumi Studio and which models it offers, in 13 languages. If you're already logged in, you go straight to your workspace. You can find the product page any time from the Product overview link in the footer.

![The new product page: the three ways to create (chat with Sumi, the canvas or MCP), the four steps from idea to delivery, and use cases.](https://sumistudio.ai/brand/blog-sumi-2-5-product-page.jpg?v=2026-10-21.1)

## Fixes and improvements

### Models

In Kling 4.0 you can now use @ in the prompt to reference your elements. Audio settings and Omni reference keyframes also work correctly now.

Qwen-Audio system voices now include word-by-word timestamps (for the voices where Qwen supports them), and Qwen-Audio 3.1 Flash now has its full list of voices.

### Canvas and editor

Image and video tasks that are still running no longer get cancelled by mistake.

When you trim a clip on the timeline, the cut now accounts for the clip's playback speed.

The editor no longer redraws the whole frame while paused, so it puts less load on your computer.

When you scroll to the end of the prompt box, the canvas stays where it is.

The Add node menu now fits on laptops with shorter screens.

### Phones, tablets and large screens

We fixed about 20 layout problems across phones, tablets, smaller desktop screens, 1920 widescreens, ultrawide monitors and phones with a notch. Footers, grids, menus and input boxes no longer get cut off or covered.

Problems on touchscreens, in Safari and with the "reduce motion" setting are fixed.

The smallest text on the site is now 12px, so it's easier to read.

### Other

The setup guide for Claude now includes the step where you turn on the connector inside the chat, which was missing before.

We rewrote the interface text in all 15 languages so it reads more naturally.

Account security is stronger: email verification codes are better protected, and the password reset page no longer shows whether an email is registered.

[Try Sumi 2.5](https://sumistudio.ai/en)