PrismImage / README.md
jasonjuan05's picture
PrismImage: 4-bit image model, 3-bit text encoder, VAE, tokenizer, style LoRA
ee7a77f verified
|
Raw History Blame Contribute Delete
733 Bytes
---
library_name: mlx
license: apache-2.0
pipeline_tag: text-to-image
tags:
- mlx
- text-to-image
- image-to-image
base_model: black-forest-labs/FLUX.2-klein-4B
---
# PrismImage
On-device image model for **Prism Image Studio** (macOS / iOS), in MLX format.
| Folder | Contents |
|---|---|
| `transformer/` | 4B distilled image transformer, 4-bit (group 128), with the Jason-Photo style fused in |
| `text_encoder/` | Prompt encoder, 3-bit (group 128) — also powers the app's built-in chat |
| `vae/` | Image autoencoder |
| `tokenizer/` | Tokenizer files |
| `loras/master-drawing.safetensors` | "Expressive" style LoRA, 8-bit row-wise |
The app downloads these files on first launch; nothing here needs to be fetched by hand.