SD.Next Release 2026-04-01 #4723
Locked
vladmandic
announced in
Announcements
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
-
SD.Next Release 2026-04-01
This release brings massive code refactoring to modernize codebase and removal of some obsolete features. Leaner & Faster!
And since its a bit quieter period when it comes to new models, notable additions would be : FireRed-Image-Edit, SkyWorks-UniPic-3 and new versions of Anima-Preview, Flux-Klein-KV image models and LTX 2.3 video model
If you're on Windows platform, we have a brand new All-in-one Installer & Launcher: simply download exe or zip and done!
And we have a new (optional) React-based UI Enso!
What else? Really a lot!
New color grading module, updated localization with new languages and improved translations, new CivitAI integration module, new finetunes loader, several new upscalers, improvements to LLM/VLM in captioning and prompt enhance, a lot of new control preprocessors, new realtime server info panel, some new UI themes
And major work on API hardening: security, rate limits, secrets handling, new endpoints, etc.
But also many smaller quality-of-life improvements - for full details, see ChangeLog
Note: Purely due to size of changes, clean install is recommended!
Just how big? Some stats: ~530 commits over 880 files
ReadMe | ChangeLog | Docs | WiKi | Discord | Sponsor
Details for 2026-04-01
Note: FireRed is a fine-tune of Qwen-Image-Edit regardless of its claim as a new base-model
Note: UniPic-3 is a fine-tune of Qwen-Image-Edit with new distillation regardless of its claim of major changes
note ltx-2.3 is a massive 22B parameters and full model is very large (72GB) so use of pre-quantized variant (32GB) is highly recommended
apply basic corrections to your images: brightness,contrast,saturation,shadows,highlights
move to professional photo corrections: hue,gamma,sharpness,temperature
correct tone: shadows,midtones,highlights
add effects: vignette,grain
apply professional lut-table using .cube file
hint color grading is available as step during generate or as processing item for already existing images
add support for spandrel engine with suport for new upscaling model families
add two new ai upscalers: RealPLKSR NomosWebPhoto and RealPLKSR AnimeSharpV2
add two new interpolation methods: HQX and ICB
use high-quality sharpfin accelerated library
extend
chainnersupport for additional modelsexpand allowed models
3.1/3.0/2.5 pro/flash/flash-lite
secrets.jsonand special handling for tokens/keys/passwordsused to be treated like any other
config.jsonparam which can cause security issuesanyline, depth_anything v2, dsine, lotus, marigold normals, oneformer, rtmlib pose, sam2, stablenormal, teed, vitpose
browse/discover mode with sort, period, type/base dropdowns; URL paste; subfolder sorting; auto-browse; dynamic dropdowns
instead of being used implicitly via quantization
available in main interface -> scripts
see main interface -> scripts -> rocm advanced config
gfx103X, thanks @crashingalexsantorch=2.11withcuda=13.0torch==2.11torch==2.11withrocm==7.2torch==2.11andopenvino==2026.0torch==2.10removed support forrtx1000series and older GPUsuse following before first startup to force installation of
torch==2.9.1withcuda==12.6:cuda==13.0requires newer nVidia driversuse following before first startup to force installation of
torch==2.11.0withcuda==12.68:nunchaku==1.2.1with WYSIWYG infinite canvas workspace, command palette, and numerous quality of life improvements across the board
enable using
--ensoflag and access using/ensoendpoint (e.g. http://localhost:7860/enso)see Enso Docs and Enso Home for details
note Enso is work-in-progress and alpha-ready
you can re-enable them in settings -> ui -> hide legacy tabs
en, en1, en2, en3, en4, hr, es, it, fr, de, pt, ru, zh, ja, ko, hi, ar, bn, ur, id, vi, tr, sr, po, he, xx, yy, qq, tlh
yes, this now includes stuff like latin, esperanto, arabic, hebrew, klingon and a lot more!
and also introduce some pseudo-locales such as: techno-babbel, for-n00bs
hint: click on locale icon in bottom-left corner to cycle through available locales, or set default in settings -> ui
/sdapi/v2/)job-based generation with queue, per-job WebSocket progress, file uploads with TTL, model/network enumeration
and a plethora of other improvements (work-in-progress)
for the time being ships with Enso, which must be enabled wih
--ensoflag on startup for v2 API to be availableconfigurable in settings -> server settings
/sdapi/v1/uploadendpoint with support for both POST with form-data or PUT using raw-bytes/sdapi/v1/torchendpoint for torch info (backend, version, etc.)/sdapi/v1/gpuendpoint for GPU info/sdapi/v1/rembgendpoint for background removal/sdapi/v1/unetendpoint to list available unets/ditsnote: SDNQ is quantization engine of choice for SD.Next
flux_enhancescriptpython==3.13full supportpython==3.14initial supportsee docs for details
clip, numba, skimage, torchsde, omegaconf, antlr, patch-ng, patch-ng, astunparse, addict, inflection, jsonmerge, kornia,resize-right, voluptuous, yapf, sqlalchemy, invisible-watermark, pi-heif, ftfy, blendmodes, PyWavelets, impthese are now installed on-demand when needed
huggingface_hub==1.5.0transformers==5.3.0constraints.txtcliscripts/test/modules/image/schedulers, todo, tome, etc.
torch.compilerebmgto core instead of extensionsnunchakuwith--reinstallflaggallery, thanks @awsrpkg_resourcestoimportlibpydantic==2.xcompatibilitymodules/loggertimestampbased startup checks with state cachingsharedmodule and introduceui_definitionsthreadingfor deferable operatiosthreadingfor io-independent parallel operationsclip,open-clipinsightface, thanks @hameerabbasivenvnormalbaepre-processordwposepre-processorhdmmodel supportxadapterscriptcodeformerandgfpganface restorerspyproject.tomlfor tool configslintrules, thanks @awsrtyto optional lint toolingpyrightto optional lint toolinggallerycache recursive cleanup, thanks @awsrclipinstaller doing unwantedsetuptoolsupdateuvinstaller fallbackmetadatarestore to always-on scriptswildcardweights parsing, thanks @Tillerzanimaloraunwanted unloadpreviewerror handlergalleryover remote/unsecure connectionsltx2-i2vpreviewimage--reinstallstricter enforcementapistate resetrembgdependenciesscriptsmore gracefullyerror-limiter, thanks @awsrlorasupport for flux2-kleinlorachange when used withsdnqsdnqfixestaesdinit errorsBeta Was this translation helpful? Give feedback.
All reactions