tscd_juggernaut_final

TSCD-Juggernaut_final

<Gallery />

Model description

Fine-tune a distill LoRA version of digiplay/Juggernaut_final using Trajectory Segmented Consistency Model (TSCD) introduce in Hyper-SD: Trajectory Segmented Consistency Model for Efficient Image Synthesis

This LoRA is fine-tuned on laion/conceptual-captions-12m-webdataset for 16.000 iterations using unofficial training implementation at https://github.com/rootonchair/consistency_models_distill

Usage

import os
import torch
from diffusers import StableDiffusionPipeline, TCDScheduler

pipeline = StableDiffusionPipeline.from_pretrained("digiplay/Juggernaut_final", torch_dtype=torch.float16, safety_checker = None).to("cuda")
pipeline.load_lora_weights("rootonchair/tscd_juggernaut_final")
pipeline.scheduler = TCDScheduler.from_config(pipeline.scheduler.config)
test_prompts = [
  "cgmech, white mecha robot, cape, science fiction, torn clothes, glowing, standing, robot joints, mecha, armor, cowboy shot, intense sunlight, silver dragonborn, outdoors, landscape, nature, volumetrics dtx",
  "Portrait photo of muscular bearded guy in a worn mech suit, elegant, sharp focus, photo by greg rutkowski, soft lighting, vibrant colors",
  "photo of a supercar, 8k uhd, high quality, road, sunset, motion blur, depth blur, cinematic, filmic image 4k",
  "a portrait of a white cat wearing glasses, highly detailed",
  "a beautiful runrise at the beach, cinematic, masterpiece",
  "A photo of beautiful mountain with realistic sunset and blue lake, highly detailed, masterpiece",
  "Self-portrait oil painting, a beautiful cyborg with golden hair, 8k",
  "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k";
]
seed = 1234
images = pipeline(test_prompts, width=512, height=512, num_inference_steps=4, guidance_scale=0, generator=torch.Generator(device="cuda").manual_seed(seed)).images

Generation Result

Original model

2-steps inference

TSCD

LCM-Lora

4-steps inference

TSCD

LCM-Lora

8-steps inference

TSCD

LCM-Lora

Download model

Weights for this model are available in Safetensors format.

Download them in the Files & versions tab.