Chameleon
Images and text in one token stream, trained from scratch
Latest: Chameleon 7B/34B (Jun 2024 weight release; image-understanding only)
FAIR's early-fusion mixed-modal foundation model (May 2024): a single token-based transformer trained from scratch on interleaved images and text, able to understand and generate arbitrary interleavings of both. Meta released 7B and 34B weights under a research license in June 2024 — a precursor to natively multimodal product models.
Why it matters
Chameleon was Meta's first early-fusion model: one token-based transformer trained from scratch on interleaved images and text, able to understand and generate either, with the 34B version beating much larger models on interleaved-generation human evals. It is the research precursor of the natively multimodal Llama 4 and Muse Spark lines.
Facts
- The released checkpoints deliberately disabled image generation for safety; the paper reported the 34B model beating much larger models on interleaved-generation human evals.
Try it yourself
chameleon-7b on Hugging Face ↗ Code on GitHub ↗ Read the paper ↗