EmoStyle is a Z-Image-based framework for affective artistic image generation. An LLM reasoner predicts valence-arousal, dominant emotion, therapeutic-effect labels, and aspect ratio from the input prompt. These inferred variables are encoded as an affective condition vector and injected into denoising blocks through AdaLN-style modulation, rather than being appended only as text. The system also trains one LoRA adapter per artistic-style bucket and selects the relevant expert at inference time. A lightweight VLM ranks candidates for prompt alignment, style consistency, emotional expression, and visual quality. The abstract reports first place in Track 1 of the 2026 AffectiveArt Challenge.
No heat snapshots are available in the last 24 hours.