WaveSpeedAI announced angel funding for its image and video inference platform. Its announcement described a serving stack that combined GPU scheduling and fused inference operations and a catalog covering video models such as Wan and Kling.
WaveSpeedAI
Also known as: WaveSpeed AI · WaveSpeed · wavespeed.ai
Plain English
WaveSpeedAI offers one API for hosted models that generate and edit images, video, audio, and 3D assets. Its catalog includes models from different publishers as well as variants WaveSpeedAI has optimized for serving.
Technical Overview
Applications submit jobs to model-specific endpoints with a shared authentication and job interface. An asynchronous request returns an ID, which clients can use to poll for results or correlate a webhook. The input schema still depends on the chosen model, so switching endpoints can require changing the request payload.
Timeline (4)
2025
Milestone- Release
WaveSpeedAI announced a hosted 480p video-to-video variant of Wan 2.1 with custom LoRA support. Applications could pass a style adapter along with source footage and a prompt to customize the transformation on the hosted endpoint.
- Ecosystem
WaveSpeedAI partnered with Alibaba to present Wan 2.5 in a September 24 launch livestream. The partnership made WaveSpeedAI a distribution channel for the model, which could generate video with synchronized audio and dialogue.
2026
EcosystemWaveSpeedAI documented a hosted MCP endpoint for its media models. Compatible clients could connect to a single remote URL with browser authorization or an API token, then discover models and submit generation jobs through the same connection.
See also (7)
Open full graphWaveSpeedAI