Personalized Image Generation with Reasoning and Reflection
PEARL personalizes image generation from user histories via a reason-reflect loop, beating baselines by about 15% on a new benchmark.
The paper introduces a unified benchmark for personalized image generation from rich user histories such as reviews, posts, images, captions, and metadata, rather than curated visual exemplars alone. Tasks are Personalized Scene Generation, placing a given object in a lifestyle-aligned scene, and Personalized Creative Generation, producing a novel image faithful to a user's aesthetic identity. PEARL couples a multimodal reasoner with a frozen image generator in an interleaved reason-reflect loop trained with differential data reward. Across both tasks it outperforms strong baselines, with an average 15% gain on personalization metrics.
- New benchmark covers scene generation and creative generation from user histories.
- Evaluation covers fidelity, quality, distinguishability, history alignment, and utility.
- PEARL pairs a multimodal reasoner with a frozen image generator.
- An interleaved reason-reflect loop is optimized with differential data reward.
- PEARL improves personalization metrics by about 15% over strong baselines.
From the original article
1 min read at huggingface.co
Personalized image generation has remained narrowly focused on conditional synthesis from curated visual exemplars, rather than capturing who a user is. In practice, however, a user's personal context is much richer, comprising reviews, posts, images, captions, and metadata accumulated over time.
By Bo Ni, Ngoc N. Tran, Qinwen Ge · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.