ArkAiLab-Adl/nexora-vector-v0.1
TEXT GENERATIONConcurrency Cost:1Model Size:4BQuant:BF16Ctx Length:32kPublished:Mar 27, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Cold

Nexora-Vector-v0.1 by ArkAiLab-Adl is a 4 billion parameter experimental text-to-vector model, based on Qwen3-4B, designed to generate structured SVG graphics from natural language prompts. This supervised fine-tuned model specializes in producing simple vector graphics, geometric shapes, and basic illustrations for rapid prototyping. It is intended for research and early-stage design workflows, translating textual instructions into SVG code.

Loading preview...

Nexora-Vector-v0.1: Text-to-Vector SVG Generation

Nexora-Vector-v0.1, developed by ArkAiLab-Adl, is a 4 billion parameter model built on Qwen3-4B, specifically fine-tuned to generate structured SVG graphics from natural language prompts. This beta release focuses on translating textual instructions into SVG code for vector-based design workflows.

Key Capabilities

  • Generates SVG markup for simple vector graphics.
  • Produces geometric shapes and basic illustrations.
  • Creates lightweight icons and minimal design assets.
  • Supports rapid prototyping in vector-based design.

Limitations and Considerations

As an early-stage beta model, Nexora-Vector-v0.1 has a high hallucination rate, often producing invalid or non-renderable SVG. Its limited generalization stems from a small training dataset (~1,500 samples), affecting consistency and making it weak at handling complex scenes. Outputs require manual validation and post-processing, and the model is not production-ready.

Intended Use Cases

  • Academic and applied research in text-to-vector generation.
  • Experimental AI-assisted design systems.
  • Educational exploration of structured output generation.
  • Lightweight SVG prototyping and ideation.

Usage Recommendations

To optimize results, users should employ concise and specific prompts, validate all SVG outputs, and expect imperfections. Iterative prompting and post-processing are recommended. Quantized versions (GGUF, MLX 4-Bit) are available via Open4bits for efficient local inference.