Upload composable_lora_script.py
Browse files- composable_lora_script.py +57 -0
composable_lora_script.py
ADDED
@@ -0,0 +1,57 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
#
|
2 |
+
# Composable-Diffusion with Lora
|
3 |
+
#
|
4 |
+
import torch
|
5 |
+
import gradio as gr
|
6 |
+
|
7 |
+
import composable_lora
|
8 |
+
import modules.scripts as scripts
|
9 |
+
from modules import script_callbacks
|
10 |
+
from modules.processing import StableDiffusionProcessing
|
11 |
+
|
12 |
+
|
13 |
+
def unload():
|
14 |
+
torch.nn.Linear.forward = torch.nn.Linear_forward_before_lora
|
15 |
+
torch.nn.Conv2d.forward = torch.nn.Conv2d_forward_before_lora
|
16 |
+
|
17 |
+
|
18 |
+
if not hasattr(torch.nn, 'Linear_forward_before_lora'):
|
19 |
+
torch.nn.Linear_forward_before_lora = torch.nn.Linear.forward
|
20 |
+
|
21 |
+
if not hasattr(torch.nn, 'Conv2d_forward_before_lora'):
|
22 |
+
torch.nn.Conv2d_forward_before_lora = torch.nn.Conv2d.forward
|
23 |
+
|
24 |
+
torch.nn.Linear.forward = composable_lora.lora_Linear_forward
|
25 |
+
torch.nn.Conv2d.forward = composable_lora.lora_Conv2d_forward
|
26 |
+
|
27 |
+
script_callbacks.on_script_unloaded(unload)
|
28 |
+
|
29 |
+
|
30 |
+
class ComposableLoraScript(scripts.Script):
|
31 |
+
def title(self):
|
32 |
+
return "Composable Lora"
|
33 |
+
|
34 |
+
def show(self, is_img2img):
|
35 |
+
return scripts.AlwaysVisible
|
36 |
+
|
37 |
+
def ui(self, is_img2img):
|
38 |
+
with gr.Group():
|
39 |
+
with gr.Accordion("Composable Lora", open=False):
|
40 |
+
enabled = gr.Checkbox(value=False, label="Enabled")
|
41 |
+
opt_uc_text_model_encoder = gr.Checkbox(value=False, label="Use Lora in uc text model encoder")
|
42 |
+
opt_uc_diffusion_model = gr.Checkbox(value=False, label="Use Lora in uc diffusion model")
|
43 |
+
|
44 |
+
return [enabled, opt_uc_text_model_encoder, opt_uc_diffusion_model]
|
45 |
+
|
46 |
+
def process(self, p: StableDiffusionProcessing, enabled: bool, opt_uc_text_model_encoder: bool, opt_uc_diffusion_model: bool):
|
47 |
+
composable_lora.enabled = enabled
|
48 |
+
composable_lora.opt_uc_text_model_encoder = opt_uc_text_model_encoder
|
49 |
+
composable_lora.opt_uc_diffusion_model = opt_uc_diffusion_model
|
50 |
+
|
51 |
+
composable_lora.num_batches = p.batch_size
|
52 |
+
|
53 |
+
prompt = p.all_prompts[0]
|
54 |
+
composable_lora.load_prompt_loras(prompt)
|
55 |
+
|
56 |
+
def process_batch(self, p: StableDiffusionProcessing, *args, **kwargs):
|
57 |
+
composable_lora.reset_counters()
|