Sagging Breasts by Deedeemegadoodo

A lot of time and effort went into making this LoRA. I did a lot of test training runs that did not give good results and its not cheap. If you like what you see and want to support, please use the "Generate Codes" feature in Runpod and share the code with me. It is the only way I can keep working on new LoRA's for Z-Image. Thanks

Recommended Settings (v1 - Z-Image-Turbo):

• Workflow is in the showcase images. My suggestion is that you give it a try. The stars of the show are latent upscale and EasyCache. Do not complain if the images you generated are not the same quality as I am showcasing here if you use the "Default" comfyui workflow or use Civitai onsite generation. Its just how all Z-Image LoRA's are (They massively degrade the turbo model quality)

• I use the text encoder from here - https://huggingface.co/Lockout/qwen3-4b-heretic-zimage

• Flowmatch Scheduler from here - https://github.com/erosDiffusion/ComfyUI-EulerDiscreteScheduler

• Trigger words:

  • Deedeemegadoodo photo, 严重下垂的巨大乳房，自然重力形态，皮肤褶皱与纹理，乳晕深色，乳头粗大凸起，真实解剖学呈现，无美化.

  • Deedeemegadoodo photo, topless woman, extreme breast sagging on huge breasts, large erect nipples, 硕大凸起的乳头, unidealized.

  • Deedeemegadoodo photo, capturing the essence of 巨大下垂的乳房 in a state of pure, unposed being. The image is about the downward arc, the full, heavy bags of tissue in their authentic drape, and the striking contrast of the large, stiffly erect nipples and their prominent areolae against the profound sag.

  • Or anything that works. Its all experimentation here. Include "Deedeemegadoodo photo" at the start and add describe your breasts

• For prompting, look at the prompts I used in the showcased images. You can skip describing the clothing of the subject you want to see topless and describe the clothing of other subjects. It works pretty well (sometimes the male subject gets breasts if your prompt is not good). Prompting is very important with Z-Image. I have a ChatGPT custom instruction that you can use - https://pastebin.com/qmeTgwt9

• You have to experiment the sampler/scheduler/lora weight/steps/denoise yourself. I am giving you the workflow on how I tested it. Its all up to you

Enjoy Kings.