TE
TechEcho
Home24h TopNewestBestAskShowJobs
GitHubTwitter
Home

TechEcho

A tech news platform built with Next.js, providing global tech news and discussions.

GitHubTwitter

Home

HomeNewestBestAskShowJobs

Resources

HackerNews APIOriginal HackerNewsNext.js

© 2025 TechEcho. All rights reserved.

Show HN: InvokeAI, an open source Stable Diffusion toolkit and WebUI

414 pointsby sophrocyneover 2 years ago
Hey everyone!<p>Excited to be able to share the release of `InvokeAI 2.0 - A Stable Diffusion Toolkit`, an open source project that aims to provide both enthusiasts and professionals a suite of robust image creation tools. Optimized for efficiency, InvokeAI needs only ~3.5GB of VRAM to generate a 512x768 image (and less for smaller images), and is compatible with Windows&#x2F;Linux&#x2F;Mac (M1 &amp; M2).<p>InvokeAI was one of the earliest forks off of the core CompVis repo (formerly lstein&#x2F;stable-diffusion), and recently evolved into a full-fledged community driven and open source stable diffusion toolkit titled InvokeAI. The new version of the tool introduces an entirely new WebUI Front-end with a Desktop mode, and an optimized back-end server that can be interacted with via CLI or extended with your own fork.<p>This version of the app improves in-app workflows leveraging GFPGAN and Codeformer for face restoration, and RealESRGAN upscaling - Additionally, the CLI also supports a large variety of features: - Inpainting - Outpainting - Prompt Unconditioning - Textual Inversion - Improved Quality for Hi-Resolution Images (Embiggen, Hi-res Fixes, etc.) - And more...<p>Future updates planned included UI driven outpainting&#x2F;inpainting, robust Cross Attention support, and an advanced node workflow for automating and sharing your workflows with the community.<p>We&#x27;re excited by the release, and about the future of democratizing the ability to create. Check out the repo (<a href="https:&#x2F;&#x2F;github.com&#x2F;invoke-ai&#x2F;InvokeAI" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;invoke-ai&#x2F;InvokeAI</a>) to get started, and join us on Discord (<a href="https:&#x2F;&#x2F;discord.gg&#x2F;ZmtBAhwWhy" rel="nofollow">https:&#x2F;&#x2F;discord.gg&#x2F;ZmtBAhwWhy</a>)!

19 comments

cercatrovaover 2 years ago
Speaking of SD, I wonder if 1.4 will be the last truly open release as Emad said 1.5 would release a while ago but it&#x27;s been held up for &quot;compliance&quot; reasons. Maybe they got legal threats due to using artists&#x27; works and stock images. If so, that would be sad to see it.<p>In a way it reminds me of people who make unofficial remakes of games but get cease and desists if they show gameplay while in development. The correct move is to fully develop the game and release it, then if you get C&amp;Ds, too late, the game is already available to download.
评论 #33160477 未加载
评论 #33156330 未加载
评论 #33155690 未加载
评论 #33155835 未加载
swyxover 2 years ago
[OT] its been hard for me to trace the universe of stable diffusion forks so ive been maintaining a list here: <a href="https:&#x2F;&#x2F;github.com&#x2F;sw-yx&#x2F;prompt-eng#sd-major-forks" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;sw-yx&#x2F;prompt-eng#sd-major-forks</a><p>please let me know&#x2F;send PRs if i missed anything, its been a couple months so i&#x27;m overdue for a round of cleanup&#x2F;reorganizing
评论 #33157303 未加载
评论 #33156107 未加载
评论 #33157310 未加载
评论 #33156607 未加载
评论 #33157565 未加载
评论 #33156037 未加载
评论 #33157721 未加载
评论 #33159667 未加载
评论 #33155669 未加载
lawikover 2 years ago
Oh, I used the dreeam.py script to back a Telegram bot. It later ended up in my demo for my talk Chat Bots as User Interfaces (with Elixir): <a href="https:&#x2F;&#x2F;www.youtube.com&#x2F;watch?v=DFGHaER6_j4" rel="nofollow">https:&#x2F;&#x2F;www.youtube.com&#x2F;watch?v=DFGHaER6_j4</a><p>I primarily used the InvokeAI release because I found it was easy to get going with on Linux and then it was simple enough to hack around with.<p>Also the first tool I&#x27;ve ever used where I&#x27;ve rode on the ragged edge of what my 3070 is okay with. I&#x27;ve had graphical glitches due to occupying all the video memory (KDE doesn&#x27;t like it). I&#x27;ve had to quit apps to make it work.<p>Thanks for making a useful thing of all this Stable Diffusion stuff. I&#x27;ve enjoyed it.
tehsauceover 2 years ago
A Shameless plug, if anyone is interested in building apps using stable diffusion and wants to keep things as cheap as possible, I built a very user-friendly API that is 1&#x2F;4 the cost of the official stable diffusion API. There is also a free demo.<p>You can try it out:<p><a href="https:&#x2F;&#x2F;computerender.com" rel="nofollow">https:&#x2F;&#x2F;computerender.com</a>.
评论 #33157283 未加载
KaoruAoiShihoover 2 years ago
Is there anything new here that might interest an existing user of auti&#x27;s gui to switch?
评论 #33155448 未加载
评论 #33156489 未加载
评论 #33156475 未加载
评论 #33159672 未加载
nohatover 2 years ago
I&#x27;ve been using a modified version of lsteins fork since almost the beginning. Recommended! It does lack some of the features of eg automatic1111, but it has good cli, and actually has a license, which is pretty important (as novelai has learned).
Timwiover 2 years ago
Sounds awesome! Unfortunately, it says that it requires a GPU. Please consider making it accessible to people without a GPU, for example using OpenVino like this (command line only) project does:<p><a href="https:&#x2F;&#x2F;github.com&#x2F;bes-dev&#x2F;stable_diffusion.openvino" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;bes-dev&#x2F;stable_diffusion.openvino</a><p>Thanks!
评论 #33157091 未加载
评论 #33157137 未加载
cmxchover 2 years ago
How hard of a requirement is the NVidia graphics chip? Polaris era AMD chips do work decently at the 4gb level (although a bit finicky) and Navi&#x2F;Big Navi AMD cards work reasonably well with modern ROCm.
评论 #33155864 未加载
评论 #33156564 未加载
iFireover 2 years ago
Can you make the ui InvokeAI as easy to install as running a Windows 11 command line script?<p>I couldn&#x27;t get it to work following <a href="https:&#x2F;&#x2F;invoke-ai.github.io&#x2F;InvokeAI&#x2F;installation&#x2F;INSTALL_WINDOWS&#x2F;" rel="nofollow">https:&#x2F;&#x2F;invoke-ai.github.io&#x2F;InvokeAI&#x2F;installation&#x2F;INSTALL_WI...</a><p>Similar to <a href="https:&#x2F;&#x2F;github.com&#x2F;cmdr2&#x2F;stable-diffusion-ui&#x2F;releases&#x2F;tag&#x2F;v2.16" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;cmdr2&#x2F;stable-diffusion-ui&#x2F;releases&#x2F;tag&#x2F;v2...</a>
评论 #33156219 未加载
neilvover 2 years ago
Nice! lstein is the SD fork that I ended up using, and I&#x27;m delighted to see it evolve into InvokeAI and keep getting better.
Ukeover 2 years ago
How good are solutions like stable diffusion at inpainting nowadays? What about the watermarks of getty et at that have been part of some of dall-e 2.0 images. Could one feasably remove such watermarks or stuff like a white grid array with these solutions?<p>So how convincing are these solutions in the worst case is what i am asking.
lucasfcostaover 2 years ago
This is much needed. Even for a software engineer like me, it was quite cumbersome to use Stable Diffusion locally without such an UI.<p>I feel like there&#x27;s just so much to improve though. Maybe SD is the definitive proof that one single feature can trickle down into many others just by adding good UI on top of it.
评论 #33162935 未加载
hda2over 2 years ago
What about safety filters? All the safety filters in the SD interfaces&#x2F;services I used so far are too false-positive happy. Can these filters be disabled or at least toned down in InvokeAI? If so, how easily?
评论 #33160382 未加载
cmsjover 2 years ago
Yay! I built an IRC bot for SD using lstein&#x27;s repo because it was the first one that I could get to work reliably on M1, so I&#x27;m really glad to see the process continue really well with InvokeAI!
paulirishover 2 years ago
PSA: You can email support@github to ask them to &quot;detach my repo as a fork&quot;, in case the repo has matured so much it shouldn&#x27;t have the &quot;forked from …&quot; treatment.
评论 #33156721 未加载
pdntspaover 2 years ago
Min requirements say 12gb, I take it this doesn&#x27;t have the optimizations that automatic1111 has for &lt;8gb cards?
评论 #33157767 未加载
评论 #33157331 未加载
ionwakeover 2 years ago
I was unable to get this to run on the Mac M1 over the last week - has anyone here had any success?
评论 #33160333 未加载
pdntspaover 2 years ago
I am super stoked to see all these Stable Diffusion forks floating around, and I don&#x27;t want to shit on the authors and their work that hard, but I swear the installation and packaging of these things is INSANE.<p>* Every single one of these seems to be a web UI, when this is desktop software that needs a desktop computer or workstation to run. Have we all collectively forgotten how to program PyGTK?<p>* Model files always go <i>in</i> the code repo. Have we forgotten how home folders work or what their purpose is? At the very least this one instructs you to make a shortcut&#x2F;symlink if you don&#x27;t want to copy the ckpt file yet <i>again</i><p>* On that note, everything is autodownloaded to wherever the hell the programmer wants (once again, usually in the code repo itself). I must have four or five different copies of ESRGAN, and I spent a bunch of time monkeying around with automatic1111&#x27;s fork trying to get it to correctly see everything when I ripped out the models folder and symlinked one in from a different place on my hard drive.<p>To the authors: can you all please get together and standardize some of this stuff? Models should go in user&#x27;s homefolders, or at a customizable location, and NOT within the scope of stuff that can be touched by git pull. (Doing so causes git to freak out in many circumstances)<p>The breakneck pace of innovation here is <i>awesome</i>, but it feels like all gas no brakes on the usability front.<p>In the Bad Old Days(tm) you ran an install script which generates a desktop icon and you click that to run it. Meanwhile with this, on Windows, one has to open an anaconda prompt, activate the anaconda venv (or whatever it is), then manually invoke the whole thing with &#x27;python scripts&#x2F;invoke.py --web&#x27;. And if there&#x27;s a one-click install script included (which invoke doesn&#x27;t, but I am not knocking it for this!), half the time they seem to try and pull down the entire world all over again (a la sd-webui).<p>Like I get this need to make it easy to use, but it&#x27;s like c&#x27;mon, there&#x27;s is existing convention for all these things. Folks, please follow it!<p>If I had a wishlist, or the wherewithal to fork my own version, it would have:<p>* an actual GUI made with an actual windowing toolkit. I don&#x27;t know why the hell everyone is so afraid of GTK, but I would use that. pyGTK is pretty simple IME, you can even read the C++ docs and it all maps over really nice to python. It doesn&#x27;t need to be pretty!<p>* configurable model locations, preferably in an agreed-upon standardized hierarchy<p>* a standardized way of embedding prompt data into the PNG, a la automatic1111<p>* an uncomplicated but not overly optimistic setup process. An install.py and run.py, both with sensible defaults so that you don&#x27;t need any command-line switches to run it except for special circumstances, and if it wants to autodownload updates then CHECK WITH ME FIRST! And preferably one that doesn&#x27;t try to move my entire world (heres looking at you, sd-webui). And it will load the venv&#x2F;conda environment for me.<p>And yes, for all the &quot;put your money where your mouth is&quot;, I&#x27;ve been thinking about forking. But I don&#x27;t know if I have the time or energy to keep up with all the developments in this space. But hey you never know...
评论 #33158599 未加载
评论 #33162967 未加载
评论 #33159003 未加载
评论 #33161352 未加载
gernbover 2 years ago
This is great but it requires lots of &quot;geek&quot; (installing dependencies, borking your system with brew, etc...)<p>Vs DiffusionBee which just works<p><a href="https:&#x2F;&#x2F;diffusionbee.com&#x2F;" rel="nofollow">https:&#x2F;&#x2F;diffusionbee.com&#x2F;</a><p>Maybe the two projects can merge?