BDSM bondage rope lora for Flux.2 Klein 9B
Trained in Diffusion Pipe on 789 bondage themed images (and some not)
Trained at 512,1024 buckets
Trained with 9B Base model
You can adjust "pressure" by increasing strenght
My noodleficated workflow is embeded inside the images, drag 'n drop into comfyui
Example prompts
a blonde woman with fair skin, medium build, and large breasts, bound by ropes in a BDSM setting. She is nude, with her arms stretched above her head, and her breasts are tightly constricted by the ropes. Her skin glistens with oil, highlighting her curves. The background features a brick wall and a concrete pillar. The lighting is soft, casting gentle shadows. The camera angle is straight-on, capturing her from the waist up.A high-resolution photograph captures a nude, fair-skinned woman with blonde hair, bound in intricate rope bondage. She is gagged with a red ball gag, her large breasts exposed and nipples erect. Her expression is one of discomfort. A hand, presumably male, is seen inserting a black vibrator into her vagina. The camera angle is slightly low, emphasizing her bound state. The lighting is soft, highlighting her skin texture and the ropes' details. The background is filled with other images and text in Japanese. The lighting is bright, highlighting her body, and the camera angle is close-up.a woman with light brown skin and dark hair, bound with ropes around her neck and breasts. She is suspended, her mouth open in a moan, eyes closed in pleasure. Her breasts are large and pendulous, with nipples exposed. The background is dimly lit, with a wooden structure and concrete floor. The camera angle is slightly above and to the side, emphasizing her expression and the tension in her body. The lighting is soft, highlighting her skin's texture and the ropes' roughness.Have fun! 😜
***** This LoRA is shared without warranties and with the condition that it is used in a lawful and responsible way. I do not support or take responsibility for illegal, harmful, or harassing uses. By downloading or using it, you accept that you are solely responsible for how it is used. *****
Description
FAQ
Comments (6)
Awesome lora thanks. Do you mind sharing your training parameters ? Not succeeding with my Klein trainings at the moment.
i assume my loras looks good on your actual workflow? have you tried other one? im using iffusion-pipe training at multiple buckets..
Heres my .toml file (hunyuanlora is my output folder, i still have that one):
output_dir = '/root/hunyuanlora/output'
dataset = 'examples/dataset.toml'
epochs = 2000
micro_batch_size_per_gpu = 1
pipeline_stages = 1
gradient_accumulation_steps = 4
gradient_clipping = 0.5
warmup_steps = 50
eval_every_n_epochs = 1
eval_before_first_step = true
eval_micro_batch_size_per_gpu = 1
eval_gradient_accumulation_steps = 1
save_every_n_epochs = 1
checkpoint_every_n_minutes = 30
activation_checkpointing = true
partition_method = 'parameters'
save_dtype = 'bfloat16'
caching_batch_size = 1
steps_per_print = 1
video_clip_mode = 'single_beginning'
[model]
type = 'flux2'
diffusion_model = '/root/hunyuanlora/models/flux/flux-2-klein-base-9b.safetensors'
vae = '/mnt/d/ComfyUI_Windows_portable/ComfyUI/models/vae/flux2-vae.safetensors'
text_encoders = [
{ path = '/mnt/d/ComfyUI_Windows_portable/ComfyUI/models/text_encoders/qwen_3_8b.safetensors', type = 'flux2' }
]
dtype = 'bfloat16'
diffusion_model_dtype = 'float8'
timestep_sample_method = 'logit_normal'
shift = 3
[adapter]
type = 'lora'
rank = 128
dtype = 'bfloat16'
[optimizer]
type = 'automagic'
weight_decay = 0.01
@K3NKÂ Thanks a lot mate. Will run some tests.
Yes your lora does wonders on my workflow. Have you manually captionned your dataset ? Those bondage images are quite difficult to get right with Joy Caption or any LLM...
@chuckest 100% joycaption, simple sysprompt tho..
this is the sysprompt im using for other dataset:
Write a detailed formal caption describing a woman in various NSFW poses, she is playing with a dildo, including lighting, positioning, dildo interaction and placement, camera angle, close-ups of technique details and tattoo placement, while excluding any text, watermarks or logos from the description.
awesome stuff. i just payed buzz for early access - can you share your workflow? I'm not getting quite the same results.
nevermind, user error. holy shit great work. following you