sanjaim899/sd15-sks-video-lora

SD 1.5 Video Scene LoRA (sks)

A LoRA fine-tuned on Stable Diffusion 1.5 using a large dataset of images extracted from a custom video. This LoRA captures the visual style, lighting, and composition of the source footage.

Model Details

• Base Model: Stable Diffusion 1.5 (stable-diffusion-v1-5/stable-diffusion-v1-5)
• Training Data: ~10,000 images extracted from a 2-hour video
• Resolution: 512x512
• LoRA Rank: 4
• LoRA Alpha: 2
• Optimizer: AdamW8bit
• Learning Rate: 1e-4 (cosine scheduler)
• Epochs: 2
• Trigger Word: sks
• Framework: kohya-ss/sd-scripts
• Training Hardware: Consumer GPU (4GB VRAM)

Usage

Diffusers (Python)

import torch
from diffusers import StableDiffusionPipeline

pipe = StableDiffusionPipeline.from_pretrained(
    "stable-diffusion-v1-5/stable-diffusion-v1-5",
    torch_dtype=torch.float16,
).to("cuda")

pipe.load_lora_weights("sanjaim899/sd15-sks-video-lora", adapter_name="trained")
pipe.set_adapters(["trained"], adapter_weights=[0.7])

image = pipe(
    prompt="sks, cinematic scene, highly detailed",
    negative_prompt="blurry, low quality, watermark",
    num_inference_steps=30,
    guidance_scale=7.5,
).images[0]

image.save("output.png")