Skip to main content
SynapsAI Cloud is built for wide model support. If the architecture loads through Transformers, Diffusers, or vLLM, you can usually launch it the same way you would an LLM: Hugging Face repo or artifact, then a private API. That includes text generation and chat, but also embeddings, ranking, images, speech, video, classification, question answering, fill-mask, and feature extraction. The table below is the source of truth for which pipelines are on. When you launch from Hugging Face, pipeline_tag in the model card must match a row. Artifacts set pipeline when you create them. Custom containers do not use this table — see Custom image. Use a row to pick an example and to see if streaming is on. Safetensors weights are supported. Custom code loads only the way Transformers, Diffusers, and vLLM load remote code. Unavailable Hub tags (text-to-video, image-to-video, video-to-video) are not listed. If your task is missing, contact support.

​
Tasks

text-to-speech is /v1/audio/speech. Other text-to-audio models use /v1/audio/generate.