Mew Design – Natural language to editable UI/graphic design

Hi HN, I’m the founder of Mew Design.

Like many of you, I love tools like Midjourney and DALL-E, but I was frustrated that they output flattened raster images. If the text is misspelled or the layout is slightly off, the whole image is unusable.

We built Mew Design to solve the "text hallucination" problem in AI graphics.

The Tech: Instead of generating a single pixel layer, we use a Multi-Agent System (currently running on Gemini 3.0 and customized models).

One agent acts as the "Art Director," parsing your prompt to understand intent (e.g., "minimalist event poster").

It dispatches tasks to specialized sub-agents: one generates the background visuals, while another calculates the typography hierarchy and layout vectors.

The result is a fully editable design where text, images, and shapes are separate layers—not a flat JPEG.

What makes it different:

Editable Text: You can actually correct typos or change fonts after generation.

Vector/Layout Awareness: It understands "logo top right" or "large headline" better than standard diffusion models.

We are currently exploring how far we can push agentic workflows in design. I’d love to hear your feedback on the generation quality and the editing experience!

(We just integrated Gemini 3.0, so speed should be improved.)