New 8Buff_Real Stable Diffusion Released!


New model focused on realistic images, based on 8Buff_Gen model, our new 8Buff_Real SD 1.5 based model is available here: https://civitai.com/models/118761?modelVersionId=128937

Full and pruned versions available!

Notes and details:

  • This model does require a good deal of prompt crafting, the trade off is you have a lot of control on the images you create. I would recommend finding a prompt that generates the style and quality to your liking and saving it as a style.
  • Use 512x512 size for most consistent results. Does Well with 512x768, 768x512, and 768x768.
  • Recommended sampling methods: DPM++ 2S a Karras, Euler a, DDIM, DPM2 a Karras, DPM++2S a,
Improvements:
  • Requires less sampling steps to get quality images. Start at 30 and adjust from there.
  • Age consistency better with wider age range.

  • Fine tuning of prompts for nationality to get certain physical features.
  • General purpose realistic model with a fine level of control. Merged with a lot of different models, links below to those that were remembered or provided inspiration for direct training.
  • Great with deform and Automatic1111 Animator extension for GIFs or videos. See here for results: https://youtube.com/shorts/KwgcCbbQ35k?feature=share


Known issues:
  • Face distortion is still an issue but better than previous versions. Restore face and positive prompts address.
  • Tends to draw topless women about 10% of the time. This can be addressed by adding negative prompts nude and NSFw for image generation. 
  • People tend to lean towards asian traits for eyes and hair. This can be addressed by adding nationality to the positive prompt. Nationalities that work with high consistency are: American, Indian, Nigerian, Japanese, and Ethiopian. Others such as European, Swiss, Mexican, work but not as consistently.
  • When using the Indian nationality prompt, it tends to give everyone a bindi. This can be addressed with adding (((bindi))) to your negative prompt.
  • Ages can be inconsistent sometimes with very young and very old people. Try positive prompts to like toddler, teen, adult, and elderly or provide a number (21 year-old).
Check out our How-tos here:
https://eightbuff.blogspot.com/p/my-text-to-img-journey-list-of-how-tos.html

Buy us a cup of coffee:
https://www.buymeacoffee.com/eightbuffalo

Follow up on Facebook:
https://facebook.com/EightBuffaloMediaGroup


Comments

Popular Posts