KielForge-HD

KielForge-HD is a custom-trained Stable Diffusion XL (SDXL) LoRA model designed for high-definition, cinematic character generation. It supports both Text-to-Image and Image-to-Image workflows, delivering rich skin textures, dramatic lighting, and sharp stylistic consistency.

Model Details

• Base Model: stabilityai/stable-diffusion-xl-base-1.0
• Model Type: Diffusion-based text-to-image and image-to-image LoRA / Checkpoint
• Training Resolution: SDXL Standard (1024x1024)
• Creator: KielTech

---

Files Provided in this Repository

• pytorch_lora_weights.safetensors: Standard PEFT LoRA weights (recommended for dynamic loading and blending).
• kielforge_sdxl_merged_fp16.safetensors: Full 16-bit standalone merged checkpoint.
• kielforge_sdxl_Q4_K_M.gguf: 4-bit quantized GGUF format for memory-efficient local inference (e.g., ComfyUI).

---

Quick Start (Python / Diffusers)

You can load and use the LoRA weights directly with the Diffusers library for both text-to-image and image-to-image tasks:

import torch
from diffusers import StableDiffusionXLPipeline

# Load base SDXL pipeline
pipeline = StableDiffusionXLPipeline.from_pretrained(
    "stabilityai/stable-diffusion-xl-base-1.0",
    torch_dtype=torch.float16,
    variant="fp16",
    use_safetensors=True
)
pipeline.enable_model_cpu_offload()

# Load KielForge-HD LoRA weights
pipeline.load_lora_weights("kiel2/KielForge-HD", weight_name="pytorch_lora_weights.safetensors")

# Text-to-Image Generation

prompt = "A cinematic close-up portrait of the man, detailed skin texture, natural lighting, masterpiece"
negative_prompt = "blurry, low quality, distorted, deformed face, plastic skin, waxy"

image = pipeline(
    prompt=prompt,
    negative_prompt=negative_prompt,
    num_inference_steps=35,
    guidance_scale=7.5,
    cross_attention_kwargs={"scale": 0.75}
).images[0]

image.save("output.png")

Image-to-Image Workflow Example

import torch
from diffusers import StableDiffusionXLImg2ImgPipeline
from diffusers.utils import load_image

# Load pipeline for Image-to-Image
pipeline = StableDiffusionXLImg2ImgPipeline.from_pretrained(
    "stabilityai/stable-diffusion-xl-base-1.0",
    torch_dtype=torch.float16,
    variant="fp16",
    use_safetensors=True
)
pipeline.enable_model_cpu_offload()

# Load LoRA weights
pipeline.load_lora_weights("kiel2/KielForge-HD", weight_name="pytorch_lora_weights.safetensors")

# Load input source image
init_image = load_image("[https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/input_file_sd.png](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/input_file_sd.png)").resize((1024, 1024))

prompt = "A cinematic portrait of the man in an outdoor canyon setting"
image = pipeline(
    prompt=prompt,
    image=init_image,
    strength=0.75,
    guidance_scale=7.5,
    cross_attention_kwargs={"scale": 0.75}
).images[0]

image.save("img2img_output.png")