A beginner's guide to the Stable-Diffusion model by Zedge on Replicate

#ai #coding #programming #machinelearning

This is a simplified guide to an AI model called Stable-Diffusion maintained by Zedge. If you like these kinds of analysis, you should join AImodels.fyi or follow us on Twitter.

Model overview

stable-diffusion is a private instance maintained by zedge. This implementation generates images from text descriptions, functioning as a latent text-to-image diffusion model. The model produces photo-realistic images based on any text input provided by the user. Similar implementations include stable-diffusion by stability-ai, which offers comparable core functionality, as well as more advanced variants like stable-diffusion-3, which delivers improved image quality, typography, and complex prompt understanding.

Model inputs and outputs

This model accepts a detailed set of parameters to control image generation and post-processing. The core input is a text prompt describing the desired image. Additional parameters control dimensions, quantity, safety features, and background handling. The output consists of the generated image or images in a format ready for use.

Inputs

Prompt: Text description of the image to generate (default: "astronaut in a 90s college party, vhs photo")
Width: Width of the output image in pixels (default: 1024)
Height: Height of the output image in pixels (default: 1024)
Seed: Random seed for reproducibility; negative values produce random results (default: -1)
Safety Prompt: Custom prompt for content safety analysis using InternVL
Num Outputs: Number of images to generate, up to 4 (default: 1)
Warm Delay: Model warming parameter; returns empty response after specified seconds if set
Disable NSFW Checker: Option to skip safety checking for generated images
Verbose: Enables detailed timing information output
Remove Background: Automatically removes the background from generated images
Threshold: Transparency threshold for background removal (0-255; higher values increase transparency)
Trim Background: Crops transparent areas after background removal
Stray Removal: Removes small components smaller than the specified ratio of the largest component
Padding: Adds padding around the image after trimming