BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:Asia/Tokyo
X-LIC-LOCATION:Asia/Tokyo
BEGIN:STANDARD
TZOFFSETFROM:+0900
TZOFFSETTO:+0900
TZNAME:JST
DTSTART:18871231T000000
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20250110T023309Z
LOCATION:Hall B7 (1)\, B Block\, Level 7
DTSTART;TZID=Asia/Tokyo:20241203T132300
DTEND;TZID=Asia/Tokyo:20241203T133400
UID:siggraphasia_SIGGRAPH Asia 2024_sess105_papers_423@linklings.com
SUMMARY:PALP: Prompt Aligned Personalization of Text-to-Image Models
DESCRIPTION:Technical Papers\n\nMoab Arar (Tel Aviv University), Andrey Vo
 ynov and Amir Hertz (Google Research), Omri Avrahami (Hebrew University of
  Jerusalem), Shlomi Fruchter and Yael Pritch (Google Research), Daniel Coh
 en-Or (Tel Aviv University), and Ariel Shamir (Reichman University)\n\nCon
 tent creators often aim to create personalized images using personal subje
 cts that go beyond the capabilities of conventional text-to-image models. 
 Additionally, they may want the resulting image to encompass a specific lo
 cation, style, ambiance, and more. Existing personalization methods may co
 mpromise personalization ability or the alignment to complex textual promp
 ts. This trade-off can impede the fulfillment of user prompts and subject 
 fidelity. We propose a new approach focusing on personalization methods fo
 r a \emph{single} prompt to address this issue. We term our approach promp
 t-aligned personalization. While this may seem restrictive, our method exc
 els in improving text alignment, enabling the creation of images with comp
 lex and intricate prompts, which may pose a challenge for current techniqu
 es. In particular, our method keeps the personalized model aligned with a 
 target prompt using an additional score distillation sampling term. We dem
 onstrate the versatility of our method in multi- and single-shot settings 
 and further show that it can compose multiple subjects or use inspiration 
 from reference images, such as artworks. We compare our approach quantitat
 ively and qualitatively with existing baselines and state-of-the-art techn
 iques.\n\nRegistration Category: Full Access, Full Access Supporter\n\nLan
 guage Format: English Language\n\nSession Chair: Kfir Aberman (Snap)
URL:https://asia.siggraph.org/2024/program/?id=papers_423&sess=sess105
END:VEVENT
END:VCALENDAR