****TRANSMISSION START****
Greetings, one and all. This is MKULTRA, bringing our weekly article. Today, we will covering the switch to Biglust 1.7 and covering some genning tips to ensure that you get the best images. We will also be including the preview for Season 4 (THE LIST)- Starting Monday April 14, 2025. Stay awhile and listen.
As usual if you want to support the content that I provide on this CivitAI page, please feel free to check out my Kofi @ https://ko-fi.com/mkultra2025. In addition to accepting donations, I will also be accepting private commissions. Please DM me for prices and Terms of Service for any commissions.
Let's Get Started on Some of the Changes!
Release of Biglust 1.7
As you may or may not have noticed, all of Season 3's LORAs (any LORAs where the location is an office) have been trained on Biglust 1.7. A colleague and teacher (MachineMinded) https://civitai.com/user/MachineMinded - has create a new merge using BigAsp2 and LUSTIFY.
Link: https://civitai.com/models/1433766/biglust-17
My LORAs stated that there is some backward compatibility between Biglust 1.7 & Biglust 1.6. However, after some extensive testing, it was discovered some aspects of the faces may not be transferring as well as they should be. As a result, if you are using Biglust 1.6 for my Season 3 LORAs, the likeness of the celebrities may not be accurate. This doesn't happen to all LORAs, but it is frequent enough to be marked as an issue. After some discussions, it was no concrete reasons why this is occurring and it is likely a quirk of the new LUSTIFY that changed in their latest model. It is highly recommended that you use Biglust 1.7 for all models trained, instead of Biglust 1.6. In addition, going forward - ALL future LORAs will be trained on Biglust 1.7.
Generation Tips - Biglust 1.7
In light of the switch to Biglust 1.7, there are several things to help you generate the best results. Biglust 1.7 is more receptive to natural language than WD14 prompting. For those who don't know, WD14 prompting is using tags to describe the image, while natural language is using sentences to describe the image.
Examples:
Sample WD14 Prompt - 1girl, CELEBNAME, masterpiece, realistic, photo, office setting, blue blouse, pants, sitting on office desk, smiling, office setting, window
Sample Natural Language Prompt: A realistic masterpiece of photo of 1girl CELEBNAME. CELEBNAME is sitting on a desk in an office setting. CELEBNAME is wearing a blue blouse and wearing pants and smiling. The photo is taking place in an office setting with a window.
Even though the natural language is a bit more wordy, the images generated will be more accurate into what you want. WD14 is more condensed, but you are essentially leaving a random gen to determine what the image is like. Natural language is more precise, but wordy. My personal strategy is finding the right prompt by experimenting and saving them in a txt file. That way, you can copy and paste the names in CELEBNAME space to quickly generate images.
Release Schedule Changes
As you may have notice, some of the Season 3's LORAs were not able to be completed. Some of the dataset gathering, upscaling and resizing took a bit longer than expected. Season 3 was slated to have 30 LORAs by Friday and I managed to create 28 LORAs. To accommodate and provide myself some breathing room, I am reducing my out to a maximum of 24 LORAs week. This comes to around 5 LORAs per day, which is still pretty significant. The extra downtime will also allow me to improve upon Season 1 LORAs that were undertrained and to help upgrade them to biglust 1.7.
With that being said - REQUESTS ON CIVITAI are now CLOSED - 4 Slots. Same Rules Apply:
Check conditions (based on criteria above) if your celebrity already has LORAs on CIvitAI.
First come, first serve. The order the comments received will determine the order.
If you request is accepted/decline - I will personally comment on your post/state the reasons.
If you don't get your request - Don't be a dick and wait for future Seasons. As my personal list runs out, there will be more space for CivitAI requests.
With all matters out of the way - Here is the New List for Monday April 14, 2025!
THE LIST
Lucy Lawless (Spartacus Era)
Amy Lou Wood
Candice Patton
Arti Mann
Bella Ramsay
Elyse Willems
Liu Yifei
Naomi Scott
Rachel Weisz (90s)
Lauren Cohan
Freema Agyeman
Sarah Silverman
Billie Piper
Kaley Cuoco
Soupytime
Winona Ryder (Early)
DizzyKitten
Hailey Van Lith
Julia Kelly
Amy Zhu
Claire Saffitz
Margot Robbie
JadeyAnh
Taylor Swift
Vic Michaelis
And that is all. Hopefully the tips can help you get the most of the LORAs. Enjoy the gens and as I always say: "Don't chaff yourself"
**** TRANSMISSION END****
MKULTRA