CRAFT: Constrained Reward via Attention Fine-Tuning for Subject Personalization without Composed Targets
Subject-driven image personalization---generating new images that preserve the identity of one or several reference subjects in novel scenes---is a foundational capability for modern visual content creation. It is currently dominated by generalized methods that fine-tune a pretrained multimodal diffusion transformer (MMDiT) on hundreds of thousands to millions of paired \emph{(reference, composed-target)} examples, where each composed target is a synthesized image of the subject in a novel scene. Producing such targets demands a costly multi-stage curation pipeline---LLM-based prompt generatio
Lineage graph
Paper → model → repo connections mined from source citations (Tier-1 exact match).
Why these links exist
Every edge carries a method, confidence, and the source snippet that justified it — so bad links are debuggable.
- PossiblePossibly related (embedding) · 51%Meta rolls out Muse, a new AI image generator →
- FuzzyOverlapping authors or contributors · 62%browser-use/browser-use →
“Shared author/contributor keys: lee”
- FuzzyOverlapping authors or contributors · 62%pytorch/pytorch →
“Shared author/contributor keys: zou”
- LinkedLinked via arxiv author · 85%Jihun Park →
“CRAFT: Constrained Reward via Attention Fine-Tuning for Subject Personalization without Composed Targets”
- LinkedLinked via arxiv author · 85%Kyoungmin Lee →
“CRAFT: Constrained Reward via Attention Fine-Tuning for Subject Personalization without Composed Targets”
- LinkedLinked via arxiv author · 85%Jongmin Gim →
“CRAFT: Constrained Reward via Attention Fine-Tuning for Subject Personalization without Composed Targets”
- LinkedLinked via arxiv author · 85%Hyeonseo Jo →
“CRAFT: Constrained Reward via Attention Fine-Tuning for Subject Personalization without Composed Targets”
- LinkedLinked via arxiv author · 85%Jaeyeul Kim →
“CRAFT: Constrained Reward via Attention Fine-Tuning for Subject Personalization without Composed Targets”