Skip to content

Changelog

This document contains an history of all changes to the API.

  • 2023-May-24: new model dreamshaper_6
  • 2023-May-18: new models deliberate_2 and dreamshaper_5
  • 2023-Apr-27: new endpoint /edit-image providing access to the model Instruct-Pix2Pix, enabling general-purpose editing of images with text instructions
  • 2023-Apr-24: new prompt weighting syntax parser: adds support for the (word word:weight) syntax, and fixes bugs and slowdowns that would rarely occur in some edge cases
  • 2023-Apr-15: public release of the api.dezgo.com API, with documentation at dev.dezgo.com
  • 2023-Apr-04: added models Anything 5 and CyberRealistic 1.3
  • 2023-Apr-01: added /text-inpainting endpoint
  • 2023-Mar-20: several backend improvements to maximize infrastructure reliability
  • 2023-Mar-08: added prompt weighting
  • 2023-Mar-02: revamped the model list and associated documentation
  • 2023-Feb-26: the total number of supported diffusion models is now 57 + added a new "models" field in /info to be able to retrieve metadata about each model
  • 2023-Feb-10: support for request parameters serialized as JSON (in addition to FormData / Multipart)
  • 2023-Jan-07: added 7 new finetuned models:
    • synthwavepunk_v2
    • redshift_diffusion
    • analog_diffusion
    • disco_diffusion_style
    • iconsmi_appiconsmodelforsd
    • lowpoly_world
    • vintedois_diffusion_v0_1
  • 2022-Dec-21: added Real-ESRGAN upscaling, as a parameter on existing methods and as a new standalone /upscale method.
  • 2022-Dec-13:
    • multi-model selection for text2image/image2image (SD 1.4/1.5/SD 2.0 512px/SD 2.0 768px/SD 2.1 512px/SD 2.1 768px/WaifuDiffusion 1.3/Anything3.0/Trinart2.0)
    • specialized inpainting models (SD-inpaint-1/SD-inpaint-2)
    • negative_prompt
    • new samplers: DPM, Euler, Euler-Ancestral
    • default settings: DPM sampler with 30 steps
    • image resolutions can now be a multiple of 8px (instead of 64px previously)
  • 2022-Oct-22: upgrade the model to Stable Diffusion v1.5 + new finetuned autoencoder f8-ft-MSE. This provides incremental improvements to fine details like eyes, teeth, hands but also overall composition/proportions.
  • 2022-Oct-13: reduced minimum resolution to 320x320
  • 2022-Sep-18: added new methods: "image2image" and "inpainting"
  • 2022-Sep-17: added new "sampler" field with possible values: "k_lms", "ddim", "pndm"
  • 2022-Sep-17: bump the range of possible seed values to 4294967295 (unsigned 32-bit integer)
  • 2022-Sep-07: added response header x-input-seed, containing the seed used to generate the image (which can either be randomly generated by the API or chosen by the user)