Upload distilled WSI diffusion model package

.gitattributes

@@ -34,3 +34,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
 compare.png filter=lfs diff=lfs merge=lfs -text
+tile.png filter=lfs diff=lfs merge=lfs -text

README.md

@@ -14,12 +14,23 @@ base_model:
 ---
 
 # W8Yi/distilled-wsi-diffusion
+![Teacher vs Student](./tile.png)
 
-Distilled WSI diffusion student model exported from local training checkpoint.
+`distilled-wsi-diffusion` is a distilled student model derived from PixCell for
+UNI-conditioned histopathology image generation. It is designed to preserve the
+visual behavior of the PixCell teacher while enabling substantially faster
+sampling with fewer denoising steps, making it practical for rapid research
+iteration, hypothesis testing, and interpretability workflows on WSI features.
+
+## Why Use This Model
+
+- Faster inference than full-step teacher sampling for UNI-conditioned generation.
+- Compatible with PixCell-based conditioning workflow already used in this repo.
+- Useful for pathology-focused generative experiments where turnaround time matters.
 
 ## What Is Included
 
-- `student_model.safetensors`: distilled student weights (no optimizer/EMA history).
+- `student_model.safetensors`: distilled student weights.
 - `inference_config.json`: base model IDs and loading config.
 - `training_args_full.json`: original training args captured from checkpoint.
 - `checkpoint_export_summary.json`: export metadata.
@@ -91,7 +102,6 @@ img = decode_latents_to_images(pipeline, latents)[0]
 `compare.png` (left = teacher, right = student):
 
 ![Teacher vs Student](./compare.png)
-The teacher-generated image is on the left, and the student-generated image is on the right.
 
 Teacher rollout (35 steps): 0.8908s  
 Student rollout (4 steps): 0.1137s  
@@ -218,4 +228,4 @@ print(f"End-to-end speedup: {teacher_total / max(student_total, 1e-9):.2f}x")
   - `StonyBrook-CVLab/PixCell-256`
   - `StonyBrook-CVLab/PixCell-pipeline`
   - `stabilityai/stable-diffusion-3.5-large` (VAE subfolder `vae`)
-- Please check and comply with upstream model licenses/terms.
+- Please check and comply with upstream model licenses/terms.