SenseNova-U1.5-8B-MoT

SenseNova-U1.5-8B-MoT

A unified multimodal model for creating and editing images, with legible Chinese and English text, native 4K output and control through boxes and reference images. For designers and researchers with a large GPU.

Made by sensenova on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: SenseNova-U1.5-8B-MoT is a unified multimodal model from SenseNova, built on the NEO-unify architecture. Its focus is visual creation, covering image generation and image editing. The card lists better composition, text rendering, native 4K output, editing that keeps the subject intact, and tighter control over regions and objects.

Who it suits: Designers, marketing teams and researchers who generate or edit images and have a GPU that can hold a 17.5B-parameter model.

Industry | Who uses it | How they use it | Example
Marketing | Teams making posters and brand assets | Generate text-dense designs in Chinese and English | A poster with a clear order of information
Visual design | Illustrators and designers | Edit an image while keeping the subject's identity | Replace one object in a scene
Research | Multimodal model researchers | Study a single model that generates and edits | Compare generation and editing in one checkpoint

You can skip it if your GPU cannot hold tens of gigabytes of weights, or if you only need text output.

Where to use it:

• Python with the transformers library, on a machine with a large GPU
• The reference inference code in the project's GitHub repository
• Windows, macOS or Linux, to download the files

Use case | Who | What they get back
Image generation | Designers | Pictures with better composition, lighting and detail
Posters and infographics | Marketers | Images with legible Chinese and English text
Image editing | Content teams | Edits that keep the subject and the untouched areas
Region control | Technical artists | Results shaped by boxes, markers and reference images

How to set it up: You need Python, a GPU with room for about 35.1 GB of BF16 weights, and the reference code from the project's GitHub repository. The card's installation steps were not visible in the text used here, so follow them from the model page.

• Python and pip
• A large GPU, with disk space for the weights
• The reference inference implementation

1. Get the reference inference code from the project's repository.
2. Install it as the card describes.
3. Download the model files.

pip install -U huggingface_hub
hf download sensenova/SenseNova-U1.5-8B-MoT

How to use it: You pass a prompt to generate an image. For editing, you pass an image with an instruction. The card says you can add bounding boxes, visual markers or reference images for finer control.

Capability | What it does
Image generation | Makes pictures from a text prompt, including native 4K output
Text rendering | Draws legible Chinese and English text in posters and infographics
Image editing | Changes local areas, inserts or replaces objects, and uses multiple references
Visual control | Directs a region or an object using boxes, markers or reference images

Example: You ask for a poster with a short Chinese headline and a smaller English line beneath it. What comes back is an image. The card says text in such designs is more legible than in the earlier release.

For newcomers and experts: If you are new, read the quick start on the model page and try one short prompt before you explore editing or boxes. Expect a large download, since the weights are about 35.1 GB.

For experts, the update reworks the patchify layers and the post-training steps, and it changes the training data and how tasks are written. The checkpoint has 17.5B parameters. A related release, TailSteer, adds steadier multi-turn editing while largely keeping this model's abilities.

Text, image or reference input
SenseNova-U1.5 unified model
Generated or edited image

Good to know:

• Size: 17.5B parameters, about 35.1 GB in BF16 and about 8.8 GB at 4-bit by estimate
• Languages: English and Chinese
• Related model: SenseNova-U1.5-8B-MoT-TailSteer targets repeated edits
• License: Apache 2.0
• Cost to run: no API key or paid service named, but it needs a capable GPU
• Data it reaches: downloads files from Hugging Face, nothing else is named
• Not for: the card names no limits, and it is aimed at visual creation

On Hugging Face, 13 November 2026

• Task: any to any
• Library: transformers
• Parameters: 17.5B
• Weights: about 35.1 GB in BF16; about 8.8 GB at 4-bit (estimate)
• Formats: Safetensors
• Downloads in the last 30 days: 5,201
• Likes: 275
• License: apache-2.0
• Languages: en, zh
• Quantized versions: 8
• Fine-tunes: 16
• Tags: transformers, safetensors, neo_chat, feature-extraction, image-generation, image-editing, any-to-any

Links

• Model: https://huggingface.co/sensenova/SenseNova-U1.5-8B-MoT
• Files: https://huggingface.co/sensenova/SenseNova-U1.5-8B-MoT/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download sensenova/SenseNova-U1.5-8B-MoT

Then follow the usage in its model card. Nothing in this kit runs by itself.