The authors developed a large unified multimodal model (UMM) trained natively on interleaved text-image trajectories, supporting image synthesis, text rendering, complex composition, and multimodal planning within a single autoregressive sequence.
Visual memory keeps characters and scenes consistent across long image generation trajectories
The authors train a unified multimodal model with persistent visual references and scene-aware retrieval to generate many interleaved images without losing long-range consistency.
Independent
Ziqiao Peng · Yutao Cui · Wang Lin · Zhong-Yu Li · Qingwen Gu · Tianhui Song · +9 more
Research Digest··3 min read
Peng et al.
Why this paper
Independent
In one line
A visual-memory framework enables long-horizon interleaved generation with strong character and scene consistency.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ·No benchmark numbers found
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§