PrunaAI's P-Image-Upscale is an AI model for fast image upscaling up to 8 megapixels, offering target and factor modes with optional realism and detail enhancements.
# PrunaAI P-Image-Upscale: fast AI image upscaling up to 8 MP
Source: [https://replicate.com/prunaai/p-image-upscale](https://replicate.com/prunaai/p-image-upscale)
P\-Image\-Upscale is Pruna’s AI\-powered image upscaling model\. Give it an image and a target resolution, and it increases the resolution while preserving detail\. It can upscale images to 4 megapixels in under a second, with support for outputs up to 8 MP\.
## How it works
You can upscale in two modes:
- **Target mode**\(default\): set a target resolution in megapixels \(1–8 MP\), and the model scales your image to match\.
- **Factor mode**: multiply each side of the image by a scaling factor \(1–8x\)\. Output is capped at 8 MP\.
## Enhancement options
Two optional toggles let you improve the quality of the upscaled image:
- **Enhance realism**\(on by default\): improves the overall realism of the output\. Works especially well on AI\-generated images\. May deviate slightly from the original\.
- **Enhance details**: sharpens fine textures and small details\. Can increase contrast and may introduce minor differences from the original\.
You can use either or both depending on the result you’re after\.
## Output formats
Supports JPEG, PNG, and WebP\. For JPEG and WebP, you can set the output quality from 0–100 \(default is 80\)\.
## Pricing
Output resolutionPrice1–4 MP$0\.005 / image5–8 MP$0\.01 / image## Links
- [Pruna API documentation](https://docs.api.pruna.ai/guides/models/p-image-upscale)
Model created4 months ago
Model updated4 weeks ago
Pruna's p-image-edit is a premium AI model on Replicate offering fast state-of-the-art image editing under one second, combining speed, affordability, and high visual quality with precise prompt adherence and text rendering capabilities.
P-Image is Pruna's text-to-image generation model that produces state-of-the-art images in less than a second, offering a combination of speed, affordability, and quality.
Alibaba’s 6B-parameter Z-Image-Turbo text-to-image model, further compressed by PrunaAI, generates 1024×1024 photorealistic images with bilingual text in <1s on 8 diffusion steps.
HiDream-ai has released HiDream-O1-Image-Dev, an 8B parameter open-source image generation model that uses a pixel-level unified transformer without external VAEs. It ranks #8 in the Artificial Analysis Text to Image Arena and supports high-resolution generation up to 2,048x2,048.
Parallel Rollout Approximation (PRA) improves pixel-space autoregressive image generation by using low-dimensional intermediate states and parallel training, achieving new state-of-the-art results on ImageNet-1K generation.