Upload README.md with huggingface_hub

README.md

@@ -16,31 +16,18 @@ size_categories:
 
 # Alucard Sprites
 
-A curated dataset of **162,337** pixel art sprites at 128x128 RGBA resolution with text captions, built for training the [Alucard](https://github.com/evilsocket/alucard) text-to-sprite generative model.
+A curated dataset of pixel art sprites at 128x128 RGBA resolution with text captions, built for training the [Alucard](https://github.com/evilsocket/alucard) text-to-sprite generative model.
+
+Each sprite has been individually normalized (multi-sprite grids split into individual sprites, tight-cropped and centered) and captioned using CLIP zero-shot classification describing sprite type, color, size, and view angle.
 
 ## Dataset Details
 
 | Property | Value |
 |----------|-------|
-| Total sprites | 162,337 |
 | Resolution | 128x128 RGBA |
 | Format | Parquet with embedded images |
 | Columns | `image` (PNG bytes), `text` (caption) |
 
-## Sources
-
-| Source | Count | License | Content |
-|--------|-------|---------|---------|
-| pixel-art-nouns | ~50K | HuggingFace | Captioned pixel art characters |
-| LPC 4-view | ~50K | CC-BY-SA 3.0 | Multi-view character sprites |
-| Kaggle Pixel Art | ~19K | Apache 2.0 | Characters, items, weapons, food |
-| PixelArt Multiview | ~12K | MIT | Multi-angle character sprites |
-| sprite_caption_dataset | ~13K | HuggingFace | Captioned sprites |
-| Kenney Assets | ~9K | CC0 | Platformer tiles, items, UI |
-| captioned-pixelart | ~5K | HuggingFace | Higher-res pixel art with captions |
-| GameTileNet | ~3K | CC0/CC-BY | Labeled game tiles |
-| DiffusionDB pixel art | ~2K | CC0 | AI-generated pixel art with prompts |
-
 ## Usage
 
 ### With Alucard (training)
@@ -49,31 +36,6 @@ A curated dataset of **162,337** pixel art sprites at 128x128 RGBA resolution wi
 pip install git+https://github.com/evilsocket/alucard.git
 ```
 
-Download the parquet and extract for training:
-
-```python
-import pandas as pd
-from pathlib import Path
-from PIL import Image
-import io
-
-df = pd.read_parquet("alucard_sprites.parquet")
-out = Path("data/train")
-out.mkdir(parents=True, exist_ok=True)
-
-for i, row in df.iterrows():
-    img = Image.open(io.BytesIO(row["image"]["bytes"]))
-    img.save(out / f"sprite_{i:06d}.png")
-    (out / f"sprite_{i:06d}.txt").write_text(row["text"])
-```
-
-Then pre-compute CLIP embeddings and train:
-
-```bash
-alucard-precompute --data-dir data/train
-alucard-train --data-dir data/train --epochs 200 --batch-size 64
-```
-
 ### With HuggingFace datasets
 
 ```python