pixel_portrait_lora_v1-lora

pixel_portrait_lora_v1-lora

Model trained with AI Toolkit by Ostris

Trigger words

No trigger words defined, but it's recommended to start with "A portrait of [subject] with [appearance/traits]".

Sample outputs

Images from left-to-right were generated at 512x512 with:

No LoRA, 500 steps, 1000 steps, 1500 steps, 2000 steps (final checkpoint)

Medieval Knight

Prompt: A portrait of a medieval knight with iron armor, wearing a full iron helmet with a red plume on the back side of the helm. The knight is facing front right at a 45-degree angle with a neutral pose. The iron armor is clean and reflective, showing a clean polish.

Witch

Prompt: A portrait of a witch with long red hair and blue eyes, wearing a dark purple witch hat and robes. The witch is facing left with a neutral expression. The robes are trimmed with white and light purple colors.

Werewolf

Prompt: A portrait of a werewolf with light gray fur and yellow eyes, wearing a red scarf and a sword on his back. The werewolf is facing forward with a confident expression.

How to get pixel-perfect images

To get pixel-perfect images, downscale by a factor of 8. So 512x512 images should downscale to 64x64, 1024x1024 to 128x128, and so on.

You can generate at higher resolutions like 1024x1024, but you may get worse pixel quality because the LoRA was trained on 512x512 images only.

Download model and use it with ComfyUI, AUTOMATIC1111, SD.Next, Invoke AI, etc.

Weights for this model are available in Safetensors format.

Download them in the Files & versions tab.

Use it with the 🧨 diffusers library

from diffusers import AutoPipelineForText2Image
import torch

pipeline = AutoPipelineForText2Image.from_pretrained('Qwen/Qwen-Image', torch_dtype=torch.bfloat16).to('cuda')
pipeline.load_lora_weights('svntax-dev/pixel_portrait_lora_v1-lora', weight_name='pixel_portrait_lora_v1_000001500.safetensors')
image = pipeline('a beautiful landscape').images[0]
image.save("my_image.png")

For more details, including weighting, merging and fusing LoRAs, check the documentation on loading LoRAs in diffusers