diff --git a/diffusers_helper/lora_utils.py b/diffusers_helper/lora_utils.py
index e3b14332..9d821c78 100644
--- a/diffusers_helper/lora_utils.py
+++ b/diffusers_helper/lora_utils.py
@@ -1,4 +1,4 @@
-from pathlib import Path
+from pathlib import Path, PurePath
from typing import Dict, List, Optional, Union
from diffusers.loaders.lora_pipeline import _fetch_state_dict
from diffusers.loaders.lora_conversion_utils import _convert_hunyuan_video_lora_to_diffusers
@@ -33,7 +33,17 @@ def load_lora(transformer, lora_path: Path, weight_name: Optional[str] = "pytorc
state_dict = _convert_hunyuan_video_lora_to_diffusers(state_dict)
- adapter_name = weight_name.split(".")[0]
+ # should weight_name even be Optional[str] or just str?
+ # For now, we assume it is never None
+ # The module name in the state_dict must not include a . in the name
+ # See https://github.com/pytorch/pytorch/pull/6639/files#diff-4be56271f7bfe650e3521c81fd363da58f109cd23ee80d243156d2d6ccda6263R133-R134
+ adapter_name = PurePath(str(weight_name).replace('_DOT_', '.')).stem.replace('.', '_DOT_')
+ if '_DOT_' in adapter_name:
+ print(
+ f"LoRA file '{weight_name}' contains a '.' in the name. " +
+ 'This may cause issues. Consider renaming the file.' +
+ f" Using '{adapter_name}' as the adapter name to be safe."
+ )
# Check if adapter already exists and delete it if it does
if hasattr(transformer, 'peft_config') and adapter_name in transformer.peft_config:
diff --git a/modules/interface.py b/modules/interface.py
index bfa240dc..0513187f 100644
--- a/modules/interface.py
+++ b/modules/interface.py
@@ -2,6 +2,7 @@
import time
import datetime
import random
+import json
import os
from typing import List, Dict, Any, Optional
from PIL import Image
@@ -125,7 +126,8 @@ def create_interface(
height=420,
elem_classes="contain-image"
)
-
+
+
with gr.Accordion("Latent Image Options", open=False):
latent_type = gr.Dropdown(
["Black", "White", "Noise", "Green Screen"], label="Latent Image", value="Black", info="Used as a starting point if no image is provided"
@@ -142,11 +144,21 @@ def create_interface(
with gr.Row():
steps = gr.Slider(label="Steps", minimum=1, maximum=100, value=25, step=1)
total_second_length = gr.Slider(label="Video Length (Seconds)", minimum=1, maximum=120, value=6, step=0.1)
- with gr.Row():
- resolution = gr.Slider(
- label="Output Resolution (Width)", minimum=128, maximum=768, value=640,
- step=32, info="Nearest valid bucket size will be used. Height will be adjusted automatically."
+ with gr.Row("Resolution"):
+ resolutionW = gr.Slider(
+ label="Width", minimum=128, maximum=768, value=640, step=32,
+ info="Nearest valid width will be used."
+ )
+ resolutionH = gr.Slider(
+ label="Height", minimum=128, maximum=768, value=640, step=32,
+ info="Nearest valid height will be used."
)
+ def on_input_image_change(img):
+ if img is not None:
+ return gr.update(info="Nearest valid bucket size will be used. Height will be adjusted automatically."), gr.update(visible=False)
+ else:
+ return gr.update(info="Nearest valid width will be used."), gr.update(visible=True)
+ input_image.change(fn=on_input_image_change, inputs=[input_image], outputs=[resolutionW, resolutionH])
with gr.Row("LoRAs"):
lora_selector = gr.Dropdown(
choices=lora_names,
@@ -155,7 +167,7 @@ def create_interface(
value=[],
info="Select one or more LoRAs to use for this job"
)
-
+ lora_names_states = gr.State(lora_names)
lora_sliders = {}
for lora in lora_names:
lora_sliders[lora] = gr.Slider(
@@ -173,20 +185,18 @@ def create_interface(
save_metadata = gr.Checkbox(label="Save Metadata", value=True, info="Save to JSON file")
with gr.Row("TeaCache"):
use_teacache = gr.Checkbox(label='Use TeaCache', value=True, info='Faster speed, but often makes hands and fingers slightly worse.')
-
n_prompt = gr.Textbox(label="Negative Prompt", value="", visible=False) # Not used
with gr.Row():
seed = gr.Number(label="Seed", value=31337, precision=0)
randomize_seed = gr.Checkbox(label="Randomize", value=False, info="Generate a new random seed for each job")
-
with gr.Accordion("Advanced Parameters", open=False):
latent_window_size = gr.Slider(label="Latent Window Size", minimum=1, maximum=33, value=9, step=1, visible=True, info='Change at your own risk, very experimental') # Should not change
cfg = gr.Slider(label="CFG Scale", minimum=1.0, maximum=32.0, value=1.0, step=0.01, visible=False) # Should not change
gs = gr.Slider(label="Distilled CFG Scale", minimum=1.0, maximum=32.0, value=10.0, step=0.01)
rs = gr.Slider(label="CFG Re-Scale", minimum=0.0, maximum=1.0, value=0.0, step=0.01, visible=False) # Should not change
- gpu_memory_preservation = gr.Slider(label="GPU Inference Preserved Memory (GB) (larger means slower)", minimum=6, maximum=128, value=6, step=0.1, info="Set this number to a larger value if you encounter OOM. Larger value causes slower speed.")
+ gpu_memory_preservation = gr.Slider(label="GPU Inference Preserved Memory (GB) (larger means slower)", minimum=1, maximum=128, value=6, step=0.1, info="Set this number to a larger value if you encounter OOM. Larger value causes slower speed.")
with gr.Accordion("Output Parameters", open=False):
mp4_crf = gr.Slider(label="MP4 Compression", minimum=0, maximum=100, value=16, step=1, info="Lower means better quality. 0 is uncompressed. Change to 16 if you get black outputs. ")
clean_up_videos = gr.Checkbox(
@@ -217,6 +227,7 @@ def create_interface(
elem_classes="contain-image"
)
+
with gr.Accordion("Latent Image Options", open=False):
f1_latent_type = gr.Dropdown(
["Black", "White", "Noise", "Green Screen"], label="Latent Image", value="Black", info="Used as a starting point if no image is provided"
@@ -233,11 +244,21 @@ def create_interface(
with gr.Row():
f1_steps = gr.Slider(label="Steps", minimum=1, maximum=100, value=25, step=1)
f1_total_second_length = gr.Slider(label="Video Length (Seconds)", minimum=1, maximum=120, value=5, step=0.1)
- with gr.Row():
- f1_resolution = gr.Slider(
- label="Output Resolution (Width)", minimum=128, maximum=768, value=640,
- step=32, info="Nearest valid bucket size will be used. Height will be adjusted automatically."
+ with gr.Row("Resolution"):
+ f1_resolutionW = gr.Slider(
+ label="Width", minimum=128, maximum=768, value=640, step=32,
+ info="Nearest valid width will be used."
+ )
+ f1_resolutionH = gr.Slider(
+ label="Height", minimum=128, maximum=768, value=640, step=32,
+ info="Nearest valid height will be used."
)
+ def f1_on_input_image_change(img):
+ if img is not None:
+ return gr.update(info="Nearest valid bucket size will be used. Height will be adjusted automatically."), gr.update(visible=False)
+ else:
+ return gr.update(info="Nearest valid width will be used."), gr.update(visible=True)
+ f1_input_image.change(fn=f1_on_input_image_change, inputs=[f1_input_image], outputs=[f1_resolutionW, f1_resolutionH])
with gr.Row("LoRAs"):
f1_lora_selector = gr.Dropdown(
choices=lora_names,
@@ -246,7 +267,7 @@ def create_interface(
value=[],
info="Select one or more LoRAs to use for this job"
)
-
+ f1_lora_names_states = gr.State(lora_names)
f1_lora_sliders = {}
for lora in lora_names:
f1_lora_sliders[lora] = gr.Slider(
@@ -275,7 +296,7 @@ def create_interface(
f1_cfg = gr.Slider(label="CFG Scale", minimum=1.0, maximum=32.0, value=1.0, step=0.01, visible=False)
f1_gs = gr.Slider(label="Distilled CFG Scale", minimum=1.0, maximum=32.0, value=10.0, step=0.01)
f1_rs = gr.Slider(label="CFG Re-Scale", minimum=0.0, maximum=1.0, value=0.0, step=0.01, visible=False)
- f1_gpu_memory_preservation = gr.Slider(label="GPU Inference Preserved Memory (GB) (larger means slower)", minimum=6, maximum=128, value=6, step=0.1, info="Set this number to a larger value if you encounter OOM. Larger value causes slower speed.")
+ f1_gpu_memory_preservation = gr.Slider(label="GPU Inference Preserved Memory (GB) (larger means slower)", minimum=1, maximum=128, value=6, step=0.1, info="Set this number to a larger value if you encounter OOM. Larger value causes slower speed.")
with gr.Accordion("Output Parameters", open=False):
f1_mp4_crf = gr.Slider(label="MP4 Compression", minimum=0, maximum=100, value=16, step=1, info="Lower means better quality. 0 is uncompressed. Change to 16 if you get black outputs. ")
f1_clean_up_videos = gr.Checkbox(
@@ -339,6 +360,67 @@ def create_interface(
object-fit: cover;
}
"""
+ # with gr.TabItem("Outputs"):
+ # outputDirectory = settings.get("output_dir", settings.default_settings['output_dir'])
+ # def get_gallery_items():
+ # items = []
+ # for f in os.listdir(outputDirectory):
+ # if f.endswith(".png"):
+ # prefix = os.path.splitext(f)[0]
+ # latest_video = get_latest_video_version(prefix)
+ # if latest_video:
+ # video_path = os.path.join(outputDirectory, latest_video)
+ # mtime = os.path.getmtime(video_path)
+ # preview_path = os.path.join(outputDirectory, f)
+ # items.append((preview_path, prefix, mtime))
+ # items.sort(key=lambda x: x[2], reverse=True)
+ # return [(i[0], i[1]) for i in items]
+ # def get_latest_video_version(prefix):
+ # max_number = -1
+ # selected_file = None
+ # for f in os.listdir(outputDirectory):
+ # if f.startswith(prefix + "_") and f.endswith(".mp4"):
+ # num = int(f.replace(prefix + "_", '').replace(".mp4", ''))
+ # if num > max_number:
+ # max_number = num
+ # selected_file = f
+ # return selected_file
+ # def load_video_and_info_from_prefix(prefix):
+ # video_file = get_latest_video_version(prefix)
+ # if not video_file:
+ # return None, "JSON not found."
+ # video_path = os.path.join(outputDirectory, video_file)
+ # json_path = os.path.join(outputDirectory, prefix) + ".json"
+ # info = {"description": "no info"}
+ # if os.path.exists(json_path):
+ # with open(json_path, "r", encoding="utf-8") as f:
+ # info = json.load(f)
+ # return video_path, json.dumps(info, indent=2, ensure_ascii=False)
+ # gallery_items_state = gr.State(get_gallery_items())
+ # with gr.Row():
+ # with gr.Column(scale=2):
+ # thumbs = gr.Gallery(
+ # # value=[i[0] for i in get_gallery_items()],
+ # columns=[4],
+ # allow_preview=False,
+ # object_fit="cover",
+ # height="auto"
+ # )
+ # refresh_button = gr.Button("Update")
+ # with gr.Column(scale=5):
+ # video_out = gr.Video(sources=[], autoplay=True, loop=True, visible=False)
+ # with gr.Column(scale=1):
+ # info_out = gr.Textbox(label="Generation info", visible=False)
+ # def refresh_gallery():
+ # new_items = get_gallery_items()
+ # return gr.update(value=[i[0] for i in new_items]), new_items
+ # refresh_button.click(fn=refresh_gallery, outputs=[thumbs, gallery_items_state])
+
+ # def on_select(evt: gr.SelectData, gallery_items):
+ # prefix = gallery_items[evt.index][1]
+ # video, info = load_video_and_info_from_prefix(prefix)
+ # return gr.update(value=video, visible=True), gr.update(value=info, visible=True)
+ # thumbs.select(fn=on_select, inputs=[gallery_items_state], outputs=[video_out, info_out])
with gr.Tab("Settings"):
with gr.Row():
@@ -430,7 +512,7 @@ def cleanup_temp_files():
# Connect the main process function (wrapper for adding to queue)
def process_with_queue_update(model_type, *args):
# Extract all arguments (ensure order matches inputs lists)
- input_image, prompt_text, n_prompt, seed_value, total_second_length, latent_window_size, steps, cfg, gs, rs, gpu_memory_preservation, use_teacache, mp4_crf, randomize_seed_checked, save_metadata_checked, blend_sections, latent_type, clean_up_videos, selected_loras, resolution, *lora_args = args
+ input_image, prompt_text, n_prompt, seed_value, total_second_length, latent_window_size, steps, cfg, gs, rs, gpu_memory_preservation, use_teacache, mp4_crf, randomize_seed_checked, save_metadata_checked, blend_sections, latent_type, clean_up_videos, selected_loras, resolutionW, resolutionH, *lora_args = args
# DO NOT parse the prompt here. Parsing happens once in the worker.
@@ -439,7 +521,7 @@ def process_with_queue_update(model_type, *args):
# Pass the model_type and the ORIGINAL prompt_text string to the backend process function
result = process_fn(model_type, input_image, prompt_text, n_prompt, seed_value, total_second_length, # Pass original prompt_text string
latent_window_size, steps, cfg, gs, rs, gpu_memory_preservation,
- use_teacache, mp4_crf, save_metadata_checked, blend_sections, latent_type, clean_up_videos, selected_loras, resolution, *lora_args)
+ use_teacache, mp4_crf, save_metadata_checked, blend_sections, latent_type, clean_up_videos, selected_loras, resolutionW, resolutionH, *lora_args)
# If randomize_seed is checked, generate a new random seed for the next job
new_seed_value = None
@@ -492,7 +574,9 @@ def end_process_with_update():
latent_type,
clean_up_videos,
lora_selector,
- resolution
+ resolutionW,
+ resolutionH,
+ lora_names_states
]
# Add LoRA sliders to the input list
ips.extend([lora_sliders[lora] for lora in lora_names])
@@ -518,7 +602,9 @@ def end_process_with_update():
f1_latent_type,
f1_clean_up_videos,
f1_lora_selector,
- f1_resolution
+ f1_resolutionW,
+ f1_resolutionH,
+ f1_lora_names_states
]
# Add F1 LoRA sliders to the input list
f1_ips.extend([f1_lora_sliders[lora] for lora in lora_names])
@@ -626,8 +712,6 @@ def load_metadata_from_json(json_path):
return [gr.update()] * (2 + num_orig_sliders)
try:
- import json
-
with open(json_path, 'r') as f:
metadata = json.load(f)
diff --git a/modules/interface.py.new b/modules/interface.py.new
deleted file mode 100644
index 3c45e843..00000000
--- a/modules/interface.py.new
+++ /dev/null
@@ -1,332 +0,0 @@
-import gradio as gr
-import time
-import datetime
-import random
-import os
-from typing import List, Dict, Any, Optional
-
-from modules.video_queue import JobStatus
-from modules.prompt_handler import get_section_boundaries, get_quick_prompts
-from diffusers_helper.gradio.progress_bar import make_progress_bar_css, make_progress_bar_html
-
-
-def create_interface(
- process_fn,
- monitor_fn,
- end_process_fn,
- update_queue_status_fn,
- load_lora_file_fn,
- job_queue,
- settings,
- default_prompt: str = '"[1s: The person waves hello] [3s: The person jumps up and down] [5s: The person does a dance]',
- lora_names: list = [],
- lora_values: list = []
-):
- """
- Create the Gradio interface for the video generation application
-
- Args:
- process_fn: Function to process a new job
- monitor_fn: Function to monitor an existing job
- end_process_fn: Function to cancel the current job
- update_queue_status_fn: Function to update the queue status display
- default_prompt: Default prompt text
- lora_names: List of loaded LoRA names
-
- Returns:
- Gradio Blocks interface
- """
- # Get section boundaries and quick prompts
- section_boundaries = get_section_boundaries()
- quick_prompts = get_quick_prompts()
-
- # Create the interface
- css = make_progress_bar_css()
- css += """
- .contain-image img {
- object-fit: contain !important;
- width: 100% !important;
- height: 100% !important;
- background: #222;
- }
- """
-
- css += """
- #fixed-toolbar {
- position: fixed;
- top: 0;
- left: 0;
- width: 100vw;
- z-index: 1000;
- background: rgb(11, 15, 25);
- color: #fff;
- padding: 10px 20px;
- display: flex;
- align-items: center;
- gap: 16px;
- box-shadow: 0 2px 8px rgba(0,0,0,0.1);
- border-bottom: 1px solid #4f46e5;
- }
- #toolbar-add-to-queue-btn button {
- font-size: 14px !important;
- padding: 4px 16px !important;
- height: 32px !important;
- min-width: 80px !important;
- }
-
- .gr-button-primary{
- color:white;
- }
- body, .gradio-container {
- padding-top: 60px !important; /* Adjust if your toolbar is taller */
- }
- """
-
- block = gr.Blocks(css=css, title="FramePack Studio", theme="soft").queue()
-
- with block:
- with gr.Row(elem_id="fixed-toolbar"):
- gr.Markdown("FramePack Studio
")
- queue_stats_display = gr.Markdown("Queue: 0 | Completed: 0
")
- refresh_stats_btn = gr.Button("Refresh", elem_id="refresh-stats-btn") # Add a refresh icon/button
- start_button = gr.Button(value="Add to Queue", elem_id="toolbar-add-to-queue-btn")
-
- with gr.Tabs():
- with gr.TabItem("Generate"):
- with gr.Row():
- with gr.Column():
- input_image = gr.Image(
- sources='upload',
- type="numpy",
- label="Image (optional)",
- height=420,
- elem_classes="contain-image"
- )
-
- with gr.Accordion("Latent Image Options", open=False):
- latent_type = gr.Dropdown(
- ["Black", "White", "Noise", "Green Screen"],
- label="Latent Image",
- value="Black",
- info="Used as a starting point if no image is provided"
- )
-
- prompt = gr.Textbox(label="Prompt", value=default_prompt)
-
- with gr.Accordion("Prompt Parameters", open=False):
- blend_sections = gr.Slider(
- minimum=0, maximum=10, value=4, step=1,
- label="Number of sections to blend between prompts"
- )
- with gr.Accordion("Generation Parameters", open=True):
- with gr.Row():
- steps = gr.Slider(label="Steps", minimum=1, maximum=100, value=25, step=1)
- total_second_length = gr.Slider(label="Video Length (Seconds)", minimum=1, maximum=120, value=5, step=0.1)
- with gr.Row("LoRAs"):
- lora_selector = gr.Dropdown(
- choices=lora_names,
- label="Select LoRAs to Load",
- multiselect=True,
- value=[],
- info="Select one or more LoRAs to use for this job"
- )
-
- lora_sliders = {}
- for lora in lora_names:
- lora_sliders[lora] = gr.Slider(
- minimum=0.0, maximum=2.0, value=1.0, step=0.01,
- label=f"{lora} Weight", visible=False, interactive=True
- )
-
- with gr.Row("Metadata"):
- json_upload = gr.File(
- label="Upload Metadata JSON (optional)",
- file_types=[".json"],
- type="filepath",
- height=100,
- )
- save_metadata = gr.Checkbox(label="Save Metadata", value=True, info="Save to JSON file")
- with gr.Row("TeaCache"):
- use_teacache = gr.Checkbox(label='Use TeaCache', value=True, info='Faster speed, but often makes hands and fingers slightly worse.')
- n_prompt = gr.Textbox(label="Negative Prompt", value="", visible=False) # Not used
-
- with gr.Row():
- seed = gr.Number(label="Seed", value=31337, precision=0)
- randomize_seed = gr.Checkbox(label="Randomize", value=False, info="Generate a new random seed for each job")
-
- with gr.Accordion("Advanced Parameters", open=False):
- latent_window_size = gr.Slider(label="Latent Window Size", minimum=1, maximum=33, value=9, step=1, visible=True, info='Change at your own risk, very experimental') # Should not change
- cfg = gr.Slider(label="CFG Scale", minimum=1.0, maximum=32.0, value=1.0, step=0.01, visible=False) # Should not change
- gs = gr.Slider(label="Distilled CFG Scale", minimum=1.0, maximum=32.0, value=10.0, step=0.01)
- rs = gr.Slider(label="CFG Re-Scale", minimum=0.0, maximum=1.0, value=0.0, step=0.01, visible=False) # Should not change
- gpu_memory_preservation = gr.Slider(label="GPU Inference Preserved Memory (GB) (larger means slower)", minimum=6, maximum=128, value=6, step=0.1, info="Set this number to a larger value if you encounter OOM. Larger value causes slower speed.")
- with gr.Accordion("Output Parameters", open=False):
- mp4_crf = gr.Slider(label="MP4 Compression", minimum=0, maximum=100, value=16, step=1, info="Lower means better quality. 0 is uncompressed. Change to 16 if you get black outputs. ")
- clean_up_videos = gr.Checkbox(
- label="Clean up video files",
- value=True,
- info="If checked, only the final video will be kept after generation."
- )
-
- with gr.Column():
- preview_image = gr.Image(label="Next Latents", height=150, visible=True, type="numpy")
- result_video = gr.Video(label="Finished Frames", autoplay=True, show_share_button=False, height=256, loop=True)
- progress_desc = gr.Markdown('', elem_classes='no-generating-animation')
- progress_bar = gr.HTML('', elem_classes='no-generating-animation')
-
- with gr.Row():
- current_job_id = gr.Textbox(label="Current Job ID", visible=True, interactive=True)
- end_button = gr.Button(value="Cancel Current Job", interactive=True)
- with gr.Row():
- queue_status = gr.DataFrame(
- headers=["Job ID", "Status", "Created", "Started", "Completed", "Elapsed"],
- datatype=["str", "str", "str", "str", "str", "str"],
- label="Job Queue"
- )
-
- with gr.TabItem("Settings"):
- with gr.Row():
- with gr.Column():
- output_dir = gr.Textbox(
- label="Output Directory",
- value=settings.get("output_dir"),
- placeholder="Path to save generated videos"
- )
- metadata_dir = gr.Textbox(
- label="Metadata Directory",
- value=settings.get("metadata_dir"),
- placeholder="Path to save metadata files"
- )
- lora_dir = gr.Textbox(
- label="LoRA Directory",
- value=settings.get("lora_dir"),
- placeholder="Path to LoRA models"
- )
- auto_save = gr.Checkbox(
- label="Auto-save settings",
- value=settings.get("auto_save_settings", True)
- )
- save_btn = gr.Button("Save Settings")
- status = gr.HTML("")
-
- # Add a refresh timer that updates the queue status every 2 seconds
- refresh_timer = gr.Number(value=0, visible=False)
-
- def refresh_timer_fn():
- """Updates the timer value periodically to trigger queue refresh"""
- return int(time.time())
-
- def get_queue_stats():
- jobs = job_queue.get_all_jobs()
- in_queue = sum(1 for job in jobs if job.status in [JobStatus.PENDING, JobStatus.RUNNING])
- completed = sum(1 for job in jobs if job.status == JobStatus.COMPLETED)
- return f"Queue: {in_queue} | Completed: {completed}
"
-
- def save_settings(output_dir, metadata_dir, lora_dir, auto_save):
- try:
- settings.update({
- "output_dir": output_dir,
- "metadata_dir": metadata_dir,
- "lora_dir": lora_dir,
- "auto_save_settings": auto_save
- })
- return "Settings saved successfully!
"
- except Exception as e:
- return f"Error saving settings: {str(e)}
"
-
- # Connect the buttons to their respective functions
- start_button.click(
- fn=process_fn,
- inputs=[
- input_image, prompt, n_prompt, seed, total_second_length,
- latent_window_size, steps, cfg, gs, rs, gpu_memory_preservation,
- use_teacache, mp4_crf, save_metadata, blend_sections, latent_type,
- clean_up_videos, lora_selector
- ] + [lora_sliders[lora] for lora in lora_names],
- outputs=[result_video, current_job_id, preview_image, progress_desc, progress_bar, start_button, end_button]
- )
-
- # Connect the end button to cancel the current job and update the queue
- end_button.click(
- fn=end_process_fn,
- outputs=[queue_status]
- )
-
- # Auto-monitor the current job when job_id changes
- current_job_id.change(
- fn=monitor_fn,
- inputs=[current_job_id],
- outputs=[result_video, current_job_id, preview_image, progress_desc, progress_bar, start_button, end_button]
- )
-
- refresh_stats_btn.click(
- fn=lambda: (get_queue_stats(), update_queue_status_fn()),
- inputs=None,
- outputs=[queue_stats_display, queue_status]
- )
-
- # Connect JSON metadata loader
- json_upload.change(
- fn=load_lora_file_fn,
- inputs=[json_upload],
- outputs=[prompt, seed] + lora_values
- )
-
- # Function to update slider visibility based on selection
- def update_lora_sliders(selected_loras):
- updates = []
- for lora in lora_names:
- updates.append(gr.update(visible=(lora in selected_loras)))
- return updates
-
- # Connect the dropdown to the sliders
- lora_selector.change(
- fn=update_lora_sliders,
- inputs=[lora_selector],
- outputs=[lora_sliders[lora] for lora in lora_names]
- )
-
- # Connect settings save button
- save_btn.click(
- fn=save_settings,
- inputs=[output_dir, metadata_dir, lora_dir, auto_save],
- outputs=[status]
- )
-
- return block
-
-
-def format_queue_status(jobs):
- """Format job data for display in the queue status table"""
- rows = []
- for job in jobs:
- created = time.strftime('%Y-%m-%d %H:%M:%S', time.localtime(job.created_at)) if job.created_at else ""
- started = time.strftime('%Y-%m-%d %H:%M:%S', time.localtime(job.started_at)) if job.started_at else ""
- completed = time.strftime('%Y-%m-%d %H:%M:%S', time.localtime(job.completed_at)) if job.completed_at else ""
-
- # Calculate elapsed time
- elapsed_time = ""
- if job.started_at:
- if job.completed_at:
- start_datetime = datetime.datetime.fromtimestamp(job.started_at)
- complete_datetime = datetime.datetime.fromtimestamp(job.completed_at)
- elapsed_seconds = (complete_datetime - start_datetime).total_seconds()
- elapsed_time = f"{elapsed_seconds:.2f}s"
- else:
- # For running jobs, calculate elapsed time from now
- start_datetime = datetime.datetime.fromtimestamp(job.started_at)
- current_datetime = datetime.datetime.now()
- elapsed_seconds = (current_datetime - start_datetime).total_seconds()
- elapsed_time = f"{elapsed_seconds:.2f}s (running)"
-
- position = job.queue_position if hasattr(job, 'queue_position') else ""
-
- rows.append([
- job.id[:6] + '...',
- job.status.value,
- created,
- started,
- completed,
- elapsed_time
- ])
- return rows
\ No newline at end of file
diff --git a/studio.py b/studio.py
index 1b4f781e..96b65f6d 100644
--- a/studio.py
+++ b/studio.py
@@ -2,6 +2,7 @@
import json
import os
+from pathlib import PurePath
import time
import argparse
import traceback
@@ -152,14 +153,15 @@ def verify_lora_state(transformer, label=""):
settings = Settings()
# --- Populate LoRA names AFTER settings are loaded ---
-lora_folder_from_settings = settings.get("lora_dir", default_lora_folder) # Use setting, fallback to default
+lora_folder_from_settings: str = settings.get("lora_dir", default_lora_folder) # Use setting, fallback to default
print(f"Scanning for LoRAs in: {lora_folder_from_settings}")
if os.path.isdir(lora_folder_from_settings):
try:
lora_files = [f for f in os.listdir(lora_folder_from_settings)
if f.endswith('.safetensors') or f.endswith('.pt')]
for lora_file in lora_files:
- lora_names.append(lora_file.split('.')[0]) # Get name without extension
+ lora_name = PurePath(lora_file).stem
+ lora_names.append(lora_name) # Get name without extension
print(f"Found LoRAs: {lora_names}")
except Exception as e:
print(f"Error scanning LoRA directory '{lora_folder_from_settings}': {e}")
@@ -222,16 +224,17 @@ def move_lora_adapters_to_device(model, target_device):
# Function to load a LoRA file
-def load_lora_file(lora_file):
+def load_lora_file(lora_file: str | PurePath):
if not lora_file:
return None, "No file selected"
try:
# Get the filename from the path
- _, lora_name = os.path.split(lora_file)
+ lora_path = PurePath(lora_file)
+ lora_name = lora_path.name
# Copy the file to the lora directory
- lora_dest = os.path.join(lora_dir, lora_name)
+ lora_dest = PurePath(lora_dir, lora_path)
import shutil
shutil.copy(lora_file, lora_dest)
@@ -246,7 +249,7 @@ def load_lora_file(lora_file):
current_transformer = lora_utils.load_lora(current_transformer, lora_dir, lora_name)
# Add to lora_names if not already there
- lora_base_name = lora_name.split('.')[0]
+ lora_base_name = lora_path.stem
if lora_base_name not in lora_names:
lora_names.append(lora_base_name)
@@ -291,7 +294,9 @@ def worker(
job_stream=None,
output_dir=None,
metadata_dir=None,
- resolution=640 # Add resolution parameter with default value
+ resolutionW=640, # Add resolution parameter with default value
+ resolutionH=640,
+ lora_loaded_names=[]
):
global transformer_original, transformer_f1, current_transformer, high_vram
@@ -425,7 +430,7 @@ def worker(
stream_to_use.output_queue.push(('progress', (None, '', make_progress_bar_html(0, 'Image processing ...'))))
H, W, C = input_image.shape
- height, width = find_nearest_bucket(H, W, resolution=resolution)
+ height, width = find_nearest_bucket(H, W, resolution=resolutionW)
input_image_np = resize_and_center_crop(input_image, target_width=width, target_height=height)
if save_metadata:
@@ -448,24 +453,36 @@ def worker(
"latent_window_size": latent_window_size,
"mp4_crf": mp4_crf,
"timestamp": time.time(),
- "resolution": resolution, # Add resolution to metadata
+ "resolutionW": resolutionW, # Add resolution to metadata
+ "resolutionH": resolutionH,
"model_type": model_type # Add model type to metadata
}
- # # Add LoRA information to metadata if LoRAs are used
- # if selected_loras and len(selected_loras) > 0:
- # lora_data = {}
- # for i, lora_name in enumerate(selected_loras):
- # # Get the corresponding weight if available
- # weight = lora_values[i] if lora_values and i < len(lora_values) else 1.0
- # # Handle case where weight might be a list
- # if isinstance(weight, list):
- # # If it's a list, use the first element or default to 1.0
- # weight_value = weight[0] if weight and len(weight) > 0 else 1.0
- # else:
- # weight_value = weight
- # lora_data[lora_name] = float(weight_value)
-
- # metadata_dict["loras"] = lora_data
+ # Add LoRA information to metadata if LoRAs are used
+ def ensure_list(x):
+ if isinstance(x, list):
+ return x
+ elif x is None:
+ return []
+ else:
+ return [x]
+
+ selected_loras = ensure_list(selected_loras)
+ lora_values = ensure_list(lora_values)
+
+ if selected_loras and len(selected_loras) > 0:
+ lora_data = {}
+ for lora_name in selected_loras:
+ try:
+ idx = lora_loaded_names.index(lora_name)
+ weight = lora_values[idx] if lora_values and idx < len(lora_values) else 1.0
+ if isinstance(weight, list):
+ weight_value = weight[0] if weight and len(weight) > 0 else 1.0
+ else:
+ weight_value = weight
+ lora_data[lora_name] = float(weight_value)
+ except ValueError:
+ lora_data[lora_name] = 1.0
+ metadata_dict["loras"] = lora_data
with open(os.path.join(metadata_dir, f'{job_id}.json'), 'w') as f:
json.dump(metadata_dict, f, indent=2)
@@ -539,7 +556,8 @@ def worker(
# --- LoRA loading and scaling ---
if selected_loras:
- for idx, lora_name in enumerate(selected_loras):
+ for lora_name in selected_loras:
+ idx = lora_loaded_names.index(lora_name)
lora_file = None
for ext in [".safetensors", ".pt"]:
# Find any file that starts with the lora_name and ends with the extension
@@ -910,14 +928,16 @@ def process(
latent_type,
clean_up_videos,
selected_loras,
- resolution,
- *lora_values
+ resolutionW,
+ resolutionH,
+ lora_loaded_names,
+ *lora_values
):
# Create a blank black image if no
# Create a default image based on the selected latent_type
if input_image is None:
- default_height, default_width = 640, 640
+ default_height, default_width = resolutionH, resolutionW
if latent_type == "White":
# Create a white image
input_image = np.ones((default_height, default_width, 3), dtype=np.uint8) * 255
@@ -965,7 +985,9 @@ def process(
'clean_up_videos': clean_up_videos,
'output_dir': settings.get("output_dir"),
'metadata_dir': settings.get("metadata_dir"),
- 'resolution': resolution # Add resolution parameter
+ 'resolutionW': resolutionW, # Add resolution parameter
+ 'resolutionH': resolutionH,
+ 'lora_loaded_names': lora_loaded_names
}
# Add LoRA values if provided - extract them from the tuple