Dreamina 3.1, developed by ByteDance, is a text-to-image generation model that serves as the visual engine for the company's creative platform, Dreamina (known as Jimeng in China). Released in mid-2025 as an iterative update to the Image 3.0 series, this model is built on the Seedream architecture and emphasizes high-fidelity aesthetics, cinematic realism, and precise control over complex scene compositions.
The model is particularly noted for its Smart Image Reference capabilities, which allow users to upload up to six reference images to guide the generation's lighting, character consistency, and layout. Version 3.1 introduced significant enhancements in text rendering, supporting high-quality typography in both English and Chinese scripts directly within generated images. This makes the model effective for designing posters, e-commerce assets, and creative branding materials.
Dreamina 3.1 supports high-resolution outputs up to 2048px and features a built-in Prompt Enhancer that automatically refines short user descriptions into detailed visual instructions. It offers versatile aspect ratio controls and maintains style consistency across batches, catering to professional workflows in cinematography and digital art. The architecture is optimized for rapid inference, typically producing detailed visuals in under 20 seconds.