Skip to content

SEED

A pretrained model for multimodal understanding and generation

Visit official site

What SEED is

The model is described as a large-scale pretrained system trained on interleaved text and visual data, showing multimodal understanding and generation after instruction tuning.

What you can do with it

  • Generate images from text
  • Generate text outputs
  • Describe the content of images
  • Run multi-turn in-context generation

Model notes

  • SEED Tokenizer v1 and v2 turn text into images
  • Emergent abilities include multi-turn in-context generation
  • The supplement says the model gives an LLM the ability to see and draw

Pros & cons

✓ What we like

  • Instruction tuning follows large-scale pretraining
  • Emergent compositional abilities are described

! What to watch out for

  • The listed abilities come from emergent composition
  • Text-to-image output runs through the tokenizers

FAQ

What are the tokenizers?

The material names SEED Tokenizer v1 and SEED Tokenizer v2, which convert text into images.

What emergent ability is named?

The material lists multi-turn in-context multimodal generation, compared to an AI assistant.

Last reviewed: 2026-11-15

More AI image generation tools

View all →

How we review