AlginGen improves zero-shot personalized image generation by training a learnable token and a selective attention mask that align textual and visual priors, achieving the best balance of concept preservation and prompt following on DreamBench++.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
AlignGen: Boosting Personalized Image Generation with Cross-Modality Prior Alignment
AlginGen improves zero-shot personalized image generation by training a learnable token and a selective attention mask that align textual and visual priors, achieving the best balance of concept preservation and prompt following on DreamBench++.