An unstable LoRA made specially for fans of hyper gigantic breasts. Hundreds of high-quality images are used to train this model in order to deliver the best result possible.
TLDR: Use it with prompts in natural language. Adjust the boob size by: very large breasts, huge breasts, gigantic breasts, hyper gigantic breasts ✅️
Recommended parameters:
Step: 30~50, the larger the better
CFG scale: 4.0
Sampler: Euler
Scheduler: Simple
Base model: PornMaster-色情大師 Flux2 Klein - V4_base_fp8 | Flux.2 Klein Checkpoint | Civitai
Resolution: 1024 x 1024
Not recommended to combine with the turbo LoRA, as the experiments confirmed degraded performance.
For those who wants to know more:
Prompts
The captions in the training dataset are carefully crafted, with their formats loosely based on the following structure: A young woman + with adj. breasts (as above) + is doing something (standing, walking, kneeling, etc.) + in some place. Description of clothes, appearance, expression, breasts. Description of the background. Description of the camera use, lighting, quality, etc.
Example 1: A completely nude woman is kneeling upright in a weathered barn. Her hyper gigantic breasts rest heavily on the ground, spreading across the straw beneath them. Her freckled face is framed by tousled, reddish-brown hair. Late afternoon sunlight slants into the barn, illuminating motes of dust and hay. Near-profile view, from side, shallow depth of field, photorealistic.
Example 2: A young blonde queen is standing in a grand palace hall, her hyper gigantic breasts resting heavily on the floor. A woman in maid outfits is kneeling beside her, struggling to lift her breasts off the ground. The background features a vast palace hall with gold ornate decorations. Side view, long shot, photorealistic.
Additionally, it is recommended to add "giantess" to the negative prompt, since the model sometimes tends to produce giantess women.
Workflow
As suggested, this LoRA does not work well together with the turbo LoRA. However, one possible way to achieve fast generation is to combine with a refinement network. For example, PornMaster [Flux.2 Klein 9B] - Anime to Realism Workflow - Realism_V3 | Flux.2 Klein Workflows | Civitai.
Specifically, lower the generation step of this LoRA to 10~20 steps to produce a rough image. Then feed the image to the refiner to improve details and so on.