use exched/checkpoint_n_lora dataset

app.py

@@ -60,7 +60,7 @@ R2_NAMESPACE = "s01"
 # Model registry
 # =============================================================================
 # Two pipelines are exposed through a single UI via the selector below. Both now
-# run custom single-file checkpoints pulled from the nsfwalex/checkpoint_n_lora
+# run custom single-file checkpoints pulled from the exched/checkpoint_n_lora
 # dataset (populated by the model_downloader Space):
 #   * Moody Pro Mix     -> Z-Image-Turbo finetune (fast, guidance-free DiT)
 #   * One Obsession     -> Illustrious / SDXL anime model with Compel weighting
@@ -84,7 +84,7 @@ def _supports_negative(model_name):
 
 # Custom checkpoints live in this HF dataset; HF_TOKEN (a Space secret) grants
 # read access. Paths mirror the model_downloader index.json layout.
-CHECKPOINT_DATASET = "nsfwalex/checkpoint_n_lora"
+CHECKPOINT_DATASET = "exched/checkpoint_n_lora"
 # Per-model checkpoint paths in the dataset live in the IMAGE_MODELS registry.
 # Base Z-Image repo supplies the VAE / text encoder / tokenizer / scheduler and
 # the transformer config; the single-file checkpoint only carries the DiT.
@@ -202,7 +202,7 @@ def _apply_prefix(prompt, prefix):
 # Image-model registry + startup loader.
 #
 # Each entry says which pipeline *family* loads the model, its checkpoint in the
-# `nsfwalex/checkpoint_n_lora` dataset, sane sampler defaults, and whether it
+# `exched/checkpoint_n_lora` dataset, sane sampler defaults, and whether it
 # honors a negative prompt. Adding a model = one entry here + uploading its
 # checkpoint (via the model_downloader Space); no other code change.
 #