Skip to main content

Hidiffusion

A training-free method that increases the resolution and speed of pretrained diffusion models.

ImagePipeline-Get
Tip
  • Supports sdxl and sd1.5 text2image, image2image controlnet and inpainting
  • Use with DDIMScheduler

SDXL example usage and results. To use with any other sd1.5 or sdxl endpoint just add the param "hidiffusion": true

ImagePipeline-Post
https://api.imagepipeline.io/sdxl/text2image/v1
ImagePipeline-Auth
API-Key


{
"seed": 5688,
"width": 1024,
"height": 1024,
"prompt": "a fish next to dendrobium flowers, ultrarealistic, highly detailed, 8k",
"samples": 1,
"model_id": "d961a274-658c-4889-8c1a-bf85416cb1c1",
"scheduler": "DDIMScheduler", //Use DDIMScheduler for best results
"guidance_scale": 5, //Guidance scale value needs to be between 3-5
"negative_prompt": "deformed anatomy",
"num_inference_steps": 25,
"hidiffusion": true
}

Hidiffusion sdxl model image pipeline

JSON Parameters

ParameterPermissible valuesNotes
model_idstrmodel_id can be found in models page. Filter by SD-1.5 or SDXL models
promptstr, 75 tokensCheck our Prompt Guide for tips
negative_promptstr, 75 tokensCheck our Prompt Guide for tips
num_inference_stepsint, [1-100]Noise is removed with each step, resulting in a higher-quality image over time. Ideal value 20-30
samplesint, [1-4]Generates a maximum of 4 samples per API call
guidance_scalefloat, [1-20]Higher guidance scale prioritizes text prompt relevance but sacrifices image quality. Ideal value 7.5-12.5
widthintWidth in pixels. Lower than or equal to 1024 for best results
heightintHeight in pixels. Lower than or equal to 1024 for best results
hidiffusionboolSet to true for better image quality and faster inference