nerijs commited on
Commit
e597801
1 Parent(s): 8460168

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +36 -0
README.md CHANGED
@@ -20,6 +20,42 @@ widget:
20
  Downscale 8 times to get pixel perfect images (use Nearest Neighbors)
21
  Use a fixed VAE to avoid artifacts (0.9 or fp16 fix)
22
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
23
  ### Tips:
24
  Don't use refiner
25
 
 
20
  Downscale 8 times to get pixel perfect images (use Nearest Neighbors)
21
  Use a fixed VAE to avoid artifacts (0.9 or fp16 fix)
22
 
23
+ ### Need more performance?
24
+ Use it with a LCM Lora!
25
+
26
+ Use 8 steps and guidance scale of 1.5
27
+
28
+ ```python
29
+ from diffusers import DiffusionPipeline, LCMScheduler
30
+ import torch
31
+
32
+ model_id = "stabilityai/stable-diffusion-xl-base-1.0"
33
+ lcm_lora_id = "latent-consistency/lcm-lora-sdxl"
34
+ pipe = DiffusionPipeline.from_pretrained(model_id, variant="fp16")
35
+ pipe.scheduler = LCMScheduler.from_config(pipe.scheduler.config)
36
+
37
+ pipe.load_lora_weights(lcm_lora_id, adapter_name="lora")
38
+ pipe.load_lora_weights("./pixel-art-xl.safetensors", adapter_name="pixel")
39
+
40
+ pipe.set_adapters(["lora", "pixel"], adapter_weights=[1.0, 1.2])
41
+ pipe.to(device="cuda", dtype=torch.float16)
42
+
43
+ prompt = "pixel, a cute corgi"
44
+ negative_prompt = "3d render, realistic"
45
+
46
+ num_images = 9
47
+
48
+ for i in range(num_images):
49
+ img = pipe(
50
+ prompt=prompt,
51
+ negative_prompt=negative_prompt,
52
+ num_inference_steps=8,
53
+ guidance_scale=1.5,
54
+ ).images[0]
55
+
56
+ img.save(f"lcm_lora_{i}.png")
57
+ ```
58
+
59
  ### Tips:
60
  Don't use refiner
61