stdxl
v1.0.4
Published
This is a Node.js module that allows you to generate images using the Replicate API. This module is based on a Python version originally created by KoushikNavuluri (https://github.com/KoushikNavuluri/stable-diffusion-xl-api) and adapted to Node.js by Rivs
Downloads
12
Readme
Stable Diffusion XL ( API )
This is a Node.js module that allows you to generate images using the Replicate API. This module is based on a Python version originally created by KoushikNavuluri and adapted to Node.js by Rivs.
Credits to the Original Creator
This module is an adaptation of an original Python version created by KoushikNavuluri. I would like to thank the original creator for providing the original Python version.
Description
This module provides a simple way to generate images using the Replicate API. It includes a genImage
function that accepts parameters such as prompt, width, height, among others. Additionally, the negativePrompt
parameter has been added to allow the inclusion of negative prompts in image generation.
Installation
To use this module in your Node.js project, you can install it via npm. Run the following command:
npm install stdxl
Usage
Here's an example of how to use the genImage function:
const imageGenerator = require('stdxl');
async function generateImage() {
const prompt = "Super hero Cat";
const negativePrompt = "Super hero cape";
const image = await imageGenerator.genImage(prompt, negativePrompt);
console.log(image);
}
generateImage();
Output
List of parameters
* prompt = Input text prompt
* negative_prompt = Input text negative prompt
* width = Width of output image(max:1024)
* height = height of output image(max:1024)
* count = Number of images to output. (minimum: 1; maximum: 4)
* refine = Which refine style to use ( no_refiner or expert_ensemble_refiner or base_image_refiner )
* scheduler = scheduler (valid_schedulers = ["DDIM" or "DPMSolverMultistep" or "HeunDiscrete" or "KarrasDPM" or "K_EULER_ANCESTRAL" or "K_EULER" or "PNDM"])
* guidance_scale = Scale for classifier-free guidance (minimum: 1; maximum: 50)
* prompt_strength = Prompt strength in image (maximum: 1)
* num_inference_steps = Number of denoising steps (minimum: 1; maximum: 500)
* high_noise_frac = for expert_ensemble_refiner, the fraction of noise to use (maximum: 1)