jackli888's picture
Duplicate from user238921933/stable-diffusion-webui
d0181bb
import html
import gradio as gr
import modules.textual_inversion.textual_inversion
import modules.textual_inversion.preprocess
from modules import sd_hijack, shared
def create_embedding(name, initialization_text, nvpt, overwrite_old):
filename = modules.textual_inversion.textual_inversion.create_embedding(name, nvpt, overwrite_old, init_text=initialization_text)
sd_hijack.model_hijack.embedding_db.load_textual_inversion_embeddings()
return gr.Dropdown.update(choices=sorted(sd_hijack.model_hijack.embedding_db.word_embeddings.keys())), f"Created: {filename}", ""
def preprocess(*args):
modules.textual_inversion.preprocess.preprocess(*args)
return f"Preprocessing {'interrupted' if shared.state.interrupted else 'finished'}.", ""
def train_embedding(*args):
assert not shared.cmd_opts.lowvram, 'Training models with lowvram not possible'
apply_optimizations = shared.opts.training_xattention_optimizations
try:
if not apply_optimizations:
sd_hijack.undo_optimizations()
embedding, filename = modules.textual_inversion.textual_inversion.train_embedding(*args)
res = f"""
Training {'interrupted' if shared.state.interrupted else 'finished'} at {embedding.step} steps.
Embedding saved to {html.escape(filename)}
"""
return res, ""
except Exception:
raise
finally:
if not apply_optimizations:
sd_hijack.apply_optimizations()