Link: https://www.instagram.com/ich.tina/

My jobs: https://promptai.altervista.org

If you want you can support me by buying my CK, the cost is a Tantum, all updates are free: https://promptai.altervista.org/zerokool-checkpoint-sd-1-5/

These days I'm trying to figure out what to do with the prompt, I've noticed, that using overly complex prompts doesn't help, slows down the creation process, and often affects the model itself. I think a lot depends on both the quality of the LORA/TI, and the quality of the CK you use. I realized after doing thousands of LORAs, that what influences besides the method of training is how you choose the images, not so much the quality which must always be very good, but just the pose, the background, the light. For completeness I train LORAs with 110 steps per images, if I exceed 15 photos, if I stay below, I give 2100 total steps dividing by the images I have. I always prefer to use 15 photos, I often clone images in the same process. Later of I will do a full article, on how I choose, and managed the training process.

Thank you all.

Every day I try to bring new interesting and diverse content, also based on your requests. I would be very grateful for any donations, as they allow me to finance my creations. A donation of at least 8 Euros gives you the password to the Premium section of my site and my Discord channel for the month. Be sure to use an email where you can receive messages.
PayPal: [email protected]

Base model is SD 1.5 - be sure to use the trigger word and an appropriate weight, between 0.65 and 0.95 works. A basic prompt and placing the trigger first is best. Example prompt: "<model> upper body, absurdres, illustration, smile, realistic, nature, outdoor, intricate details"
I also offer custom template creation services - simply provide me with photos of yourself, a friend, a relative or a pet and I will create a model tailored to their appearance. The file will remain your exclusive property.
Sampler: DPM++ 2M Karras / DPM++ SDE Karras