Skip to content

Prism — Vision Models

Small, self-contained image-to-image models for upscaling, denoising, and steganography — general-purpose utilities that run on modest hardware.

pip install olaverse[vision]

Each Prism model ships its own small model.py architecture file alongside the checkpoint on its Hugging Face repo. Loading a Prism model downloads and executes that model.py from the corresponding olaverse/prism-* repo. All Prism repos are published by Olaverse under Apache-2.0.

PrismUpscaler — Super-Resolution

Model Cards: olaverse/prism-upscaler-2x · olaverse/prism-upscaler-4x · olaverse/prism-upscaler-max

Which size?

size= Architecture Scale Pick when
"2x" (default) FSRCNN (~25K params) Fixed 2x Fast, single forward pass
"4x" FSRCNN (~25K params) Fixed 4x Bigger jump, accepts smoothing tradeoff
"max" LIIF (RRDB + implicit MLP) Any continuous resolution Exact target size needed 
from olaverse import PrismUpscaler

# Fixed scale
upscaler = PrismUpscaler(size="2x")
upscaler.upscale("input.jpg").save("output.jpg")

# Arbitrary target resolution
upscaler_max = PrismUpscaler(size="max")
upscaler_max.upscale("input.jpg", target_size=(1024, 1024)).save("output.jpg")

All three were trained with realistic degradation (blur, sensor noise, JPEG re-compression) rather than plain bicubic downsampling — built for real-world low-quality input.

Known limitations

  • 4x over-smooths fine/curly hair and other high-frequency texture — a consistent tradeoff at this scale.
  • None of the three have been evaluated against standard academic benchmarks (Set5/Set14/BSD100/Urban100) — comparisons on each model card are informal checks against a bicubic baseline.

PrismDenoiser — Noise/Blur/Compression Removal

Model Card: olaverse/prism-denoiser

Removes Gaussian noise, blur, and JPEG-like compression artifacts using a compact U-Net. Output resolution matches input (128x128 in, 128x128 out).

from olaverse import PrismDenoiser

denoiser = PrismDenoiser()
denoiser.denoise("noisy.jpg").save("denoised.jpg")

Reduces, doesn't eliminate, noise

On complex scenes, denoising achieves +3-4 dB PSNR but is typically incomplete — some residual grain remains. On near-grayscale images, the model can render a faint color tint, since it was trained predominantly on full-color photos.

PrismSteganography — Hide/Recover Messages

Model Card: olaverse/prism-steganography

Hides a recoverable message (up to 8 ASCII characters / 64 bits) inside a cover image imperceptibly, using a jointly-trained U-Net encoder / CNN decoder pair. A differentiable noise layer at train time means the decoder recovers the message even after distortion — not just from a pristine copy.

from olaverse import PrismSteganography

steg = PrismSteganography()

stego_image = steg.hide("cover.jpg", "hi there")
stego_image.save("stego.jpg")

steg.reveal(stego_image)   # → 'hi there'

Images are resized to 128x128 internally; longer messages are silently truncated to 8 characters.

Robustness under severe distortion

Clean recovery averages 99.9% bit-accuracy; under distortion the average drops to 93.7%, worst-case 62.5%. No error-correction coding is applied — applications needing near-100% reliability should add redundancy (e.g. a repetition or Hamming code) on top of the raw bit channel.

Applications

  • ✅ Photo restoration — upscale and denoise legacy or low-quality archives
  • ✅ OCR preprocessing — clean scans before text extraction
  • ✅ Thumbnails to full-size — arbitrary-resolution upscaling with max
  • ✅ Watermarking research — recoverable invisible tags via steganography

API Reference

Full class reference: Vision →