Add match_target_res for Qwen Image 2.1 so reference images are scaled to the target's pixel area with their own aspect ratio, consistently across text-embedding caching, training, and sampling, and key the embedding cache on it.

toolkit/data_loader.py

@@ -530,6 +530,7 @@ class AiToolkitDataset(LatentCachingMixin, ControlCachingMixin, CLIPCachingMixin
                     encode_first_frame_in_text_embeddings=getattr(self.sd, 'encode_first_frame_in_text_embeddings', False) if self.sd else False,
                     dopsd_self_ref=getattr(self.sd, 'dopsd_self_ref', False) if self.sd else False,
                     text_embedding_space_version=text_embedding_space_version,
+                    text_embedding_uses_target_size=getattr(self.sd, 'text_embedding_uses_target_size', False) if self.sd else False,
                     te_padding_side=self.sd.te_padding_side if self.sd else "right",
                     latent_space_version=latent_space_version,
                     temporal_compression=temporal_compression,