oowa AI

upload a reference — a product, a character, a style. oowa remembers it. every image after the first one just fits. reference once. remix forever.

our tools:

→ layer editing Agent
powered by Claude. reasons through what you're asking, generates and edits multiple images, and uses your references like a creative director would.

→ oowa Spark
low-latency multimodal autocomplete for image prompts. helps you write better prompts without rewriting what you actually wanted. built on a multimodal MoE base, adapted with high-rank LoRA SFT, aligned via offline reward-model-guided RL using implicit product feedback. translation: it finishes your sentences. correctly. live for all free and paid users during limited preview.

→ reference-based generation
the wedge. the point. show, don't prompt.

who it's for: creators, marketing teams, DTC and e-commerce folks tired of AI output that screams "made with AI" from across the room.

made by Homi AI · built in Seoul