piratediffusion
patreon
What's new: Weighted Prompts, LLM, big Web UI redesign, two new commands, 919(!) models, and June 15th Maintenance
🕑 Added 2023-06-05 08:09:11 +0000 UTCPirate Diffusion is growing! We have so many new things so let's dive right in
Various Web UI improvements
Concepts Browser update - There are now 919 models on-demand (that's 99 base + many LoRas), with improved sorting by Popularity, New, Rating, and Alphabetically. Remember to hit that adult toggle to see the full list, and within the descriptions many LoRAs contain obscure optional triggers you can now easily copy into your prompts to surface their full effects. Infinite mix/match possibilities.
Tiny quality of life fix: We've also lowered the default weight to 0.7 when a weight isn't specified, as we've found this is the sweet spot for many of the LoRAs.
Weighted Prompts are now supported - Advanced prompters, rejoice! You can now write prompts like (dog:2.0, cat:0.5) to make relatively weighted keywords, and our updated Web UI has sliders for keywords as well as LoRAs to make this easier to do on your mobile phone. This feature is also available in Telegram when adding the /parser:new command. Click that guide to see how it all works.
Of course, you can still use split Positive/Negative boxes style, too and avoid weights all together. There is an advanced and simple mode you can toggle.
To use it in our improved Web UI, click Login (top right) and proceed to your private area and then click 'advanced' and then 'weights' (top right).
We've also added visual recipes (prompt templates) that let you achieve stunning results just by selecting one. Literally a hundred you can choose from, all created by our community! They really put work into them, we were super impressed by the quality as we tested each one and added a thumbnail to make them easier to use.
Two New Commands for Telegram
You can now Delete and Refine - two new Telegram commands to help you save time. Refine lets you jump back into a prompt that you already ran, and Delete removes the image from both chat and your web archive. Don't forget you can also cancel if you make a typo or have a change of mind before the image arrives, if you're fast enough. Today's render averages were under 30 seconds!
We've lost count of how many commands are now possible with Telegram, but there are 77 pages of tutorials and documentation at your disposal on our website!
Anime lovers, how about some prompt inspiration?
Don't miss our prompt inspiration channel this week for prompts on Sushi Mix as well as six full guides on the AnimeMaker.ai website, which are two Telegram groups also included as part of our Pirate Diffusion / Graydient service.
Over on 18+ Dalle, you've gotta try the Chainsaw Man LoRAs using a realistic model, wow! This red-hot Reddit is quite bonkers so come kill some brain cells
Pictured: "Allergies" by Trako. Did you even notice cat number 2?
Hey, eyes up here - Important June 15th Groups Spam Cleanup
Unfortunately, some spam bots snuck into our private groups, and though we tried hard to ban them they seem to be coming back from crawled links. We hate to inconvenience you, but we changed all invite links and hid our groups map as we make this important security update to stop future spammers from sneaking in.
On June 15th, any unidentified accounts in our groups without an active Patreon account will get removed from the community groups, and all private bots not owned by a Pro user will expire.
We will not delete accounts, so there is no danger of losing data, but to avoid a temporary interruption of service please:
1. Know the Patreon email associated with your account by clicking Settings in Patreon and copying that email, and then:
2. Login to our website and update your email (or) type the /email command in Telegram to update your email
3. Please make sure your account is in good standing. If you have any doubts or questions please contact us at the NEW vip link:
https://t.me/+jUibfB_aUE4xZWFh
What's coming up next?
The next planned upgrade is full support for Textual Inversions and Lycoris, as well as new Control Net models. We have some housekeeping first, and this is next. We have also started exploring bringing non-AI tools into the service to improve your quality of life, such as being able to make convenient edits to your images.
Next-generation features are also planned: our own LLM
A large language model is coming to Pirate Diffusion. POLLY is a drunken parrot that knows a lot about Stable Diffusion, LoRAs, and writing good prompts. We have our LLM running queries for 10 alpha testers right now and it will offer a Chat GPT style open-ended conversation ability, but much more drunk and silly, and all about rendering great images without having to type long prompts. We look forward to rolling out in the coming weeks. Naturally, Patrons will be the first to try all of the features of POLLY. It's kind of crazy that we now have our own LLM. Bonkers.
And truly, thanks for your support:
We couldn't have acquired the insane server running POLLY without your patronage, it's got 48 cores and 250GB of RAM! And we hope you've noticed some speed improvements to our service overall as we bring on more 4090s. Thanks again for supporting us, we're fully committed to bringing you the best service possible.
That's all for this week, Render-Ahoy!
