TuZhan launches RabbitVis, an AI design tool with layer-level editing

Visual AI company TuZhan Intelligent has announced the official launch of RabbitVis, an AI design production tool positioned as letting AI complete the full design workflow. Beyond generating a single image, users can now continue with editing, layout changes, and adaptation work.

Over the past two years, AI models have reshaped content production, and visual models have kept pushing the boundaries of generation quality. But a persistent problem remains: when image generation ends, the design work is only beginning. Moving a logo to the left, enlarging a font, or producing an e-commerce version — the real time sink is not the creativity itself, but the endless rounds of revision, adjustment, and resizing that follow generation.

RabbitVis targets exactly that pain point. Through AI generation, editing, and layer capabilities, it reduces users' reliance on complex design software like Photoshop and Illustrator, bringing AI deeper into real design production workflows.

The editing capability rests on TuZhan's visual model foundation. Earlier, UniWorld-View, developed with Peking University and the Peng Cheng Laboratory, topped the WorldScore world model leaderboard. For design production, the company built UniWorld-Design, which uses layers as a key organizing structure for visual content, letting the model both generate images and understand the structural relationships within them.

Concretely, UniWorld-Design can generate ready-to-use transparent design elements from text descriptions and decompose full images into multiple editable layers for later adjustment, combination, and reuse. In evaluation, its I2L image-layering results include an RGB L1 of 0.1264, an Alpha Soft IoU of 0.7325, and an editability score of 1.32; its VLM subjective composite score reached 20.43 out of 25 with a normalized total of 0.817, and its T2RGBA transparent-material generation achieved a CLIP Score of 33.03.

TuZhan argues that AI competition is shifting from general capabilities to scenario-specific capabilities: the model provides the capability, while what actually creates value is the product. RabbitVis does not redefine the model — it redefines how AI models should enter the design industry and actually finish a job.

For professional designers, RabbitVis reduces repetitive operations; for business operators, content creators, and ordinary users without design backgrounds, it lowers the barrier to professional-grade design. The vision is to turn AI output from a one-shot image into design assets that can be further processed, adjusted, and reused, ultimately becoming deliverable design results.