Definition
Plain language
A unified vision-language model that interleaves text and image generation.
As stated in the literature
A unified multimodal model performing autoregressive text generation interleaved with diffusion-based image generation; used in VibeServe as a long-tail serving target.
Why it matters: Unified text-and-image models stress inference serving in unusual ways, making them a stringent test of how flexible a serving stack like VibeServe actually is.
For example, the model can be asked to write a short story and illustrate each paragraph with a generated image, all in one continuous output stream.