Researchers waste time setting up multimodal models before they can experiment with them. A playground would let them immediately test embeddings.
Build a web app with file uploads, similarity visualization, and API snippets. Support image, text and video inputs.
Monetize through compute credits for larger-scale experiments.
Launch with a basic demo using small test files.
The biggest risk is Tencent changing API access terms.