Skip to content

Tencent EMMA

Generate images from text plus reference images

Visit official site

What Tencent EMMA is

The model is described as an image generation model built on the ELLA text-to-image diffusion model, able to take multimodal prompts that combine text with supplementary modalities such as reference images. It freezes all parameters of the original T2I diffusion model and adjusts only a few extra layers.

What you can do with it

  • Feed text and reference images as multimodal prompts
  • Combine text with other modalities through a special attention mechanism
  • Freeze the original T2I parameters and tune only extra layers
  • Generate high-fidelity, detail-rich images without extra training

Example uses

  • Generate images in several styles together with ToonYou
  • Keep portrait detail in images made with AnimateDiff
  • Build an image set that follows a storyline

Getting started

  • Visit the product page and read the basic introduction
  • Read the technical documentation on how the model works
  • Install the needed dependencies such as a Python environment and libraries
  • Write your own multimodal prompts following the example code

Pros & cons

✓ What we like

  • Handles different multimodal configurations without extra training
  • Adapts to existing frameworks for personalised, context-aware images and video

! What to watch out for

  • Prompt design combines text with supplementary modalities
  • The design keeps the original T2I diffusion model frozen

FAQ

What prompts does it accept?

The material states it accepts multimodal prompts such as text together with reference images.

What model is it built on?

It is built on the ELLA text-to-image diffusion model, with the original parameters frozen.

Last reviewed: 2026-11-15

More AI image generation tools

View all →

How we review