Skip to content
Python

added ditto support

deepbeepmeepa31cb5d98302
diff --git a/README.md b/README.md
index 34f10de..bcb802a 100644
--- a/README.md
+++ b/README.md
@@ -21,9 +21,7 @@ WanGP supports the Wan (and derived models), Hunyuan Video and LTV Video models
**Follow DeepBeepMeep on Twitter/X to get the Latest News**: https://x.com/deepbeepmeep
## 🔥 Latest Updates :
-### October 20 2025: WanGP v9.01 - 0.001 Later
-
-What else will you ever need after this one ?
+### October 24 2025: WanGP v9.10, What else will you ever need after this one ?
With WanGP v9 you will have enough features to go to a desert island with no internet connection and comes back with a full Hollywood movie.
@@ -31,6 +29,7 @@ First here are the new models supported:
- **Wan 2.1 Alpha** : a very requested model that can generate videos with *semi transparent background* (as it is very lora picky it supports only the *Self Forcing / lightning* loras accelerators)
- **Chatterbox Multilingual**: the first *Voice Generator* in WanGP. Let's say you have a flu and lost your voice (somehow I can't think of another usecase), the world will still be able to hear you as *Chatterbox* can generate up to 15s clips of your voice using a recorded voice sample. Chatterbox works with numerous languages out the box.
- **Flux DreamOmni2** : another wannabe *Nano Banana* image Editor / image composer. The *Edit Mode* ("Conditional Image is first Main Subject ...") seems to work better than the *Gen Mode* (Conditional Images are People / Objects ..."). If you have at least 16 GB of VRAM it is recommended to force profile 3 for this model (it uses an autoregressive model for the prompt encoding and the start may be slow).
+- **Ditto** (new with *WanGP 9.1* !): a powerful Video 2 Video model, can change for instance the style or the material visible in the video. Be aware it is an instruct based model, so the prompt should contain intructions.
Upgraded Features:
- A new **Audio Gallery** to store your Chatterbox generations and import your audio assets. *Metadata support* (stored gen settings) for *Wav files* generated with WanGP available from day one.
diff --git a/defaults/vace_ditto_14B.json b/defaults/vace_ditto_14B.json
new file mode 100644
index 0000000..93a68c3
--- /dev/null
+++ b/defaults/vace_ditto_14B.json
@@ -0,0 +1,18 @@
+{
+ "model": {
+ "name": "Ditto 14B",
+ "architecture": "vace_ditto_14B",
+ "modules": [
+ "vace_14B"
+ ],
+ "description": "Ditto demonstrates superior instruction-following ability and establish a new SOTA in instruction-based video editing (style, transfer, change of material, etc...). This model expects the Prompt to contain instructions to apply to the whole frames. The Masked Area option is experimental.",
+ "URLs": "t2v",
+ "preload_URLs": [
+ "https://huggingface.co/DeepBeepMeep/Wan2.1/resolve/main/ditto_global.safetensors",
+ "https://huggingface.co/DeepBeepMeep/Wan2.1/resolve/main/ditto_global_style.safetensors",
+ "https://huggingface.co/DeepBeepMeep/Wan2.1/resolve/main/ditto_sim2real.safetensors"
+ ]
+
+ },
+ "prompt": "Render the subjects as classical sculptures carved from single blocks of pristine white marble"
+}
\ No newline at end of file
diff --git a/models/wan/any2video.py b/models/wan/any2video.py
index 92c73e9..6484f41 100644
--- a/models/wan/any2video.py
+++ b/models/wan/any2video.py
@@ -1171,5 +1171,10 @@ class WanAny2V:
preloadURLs = get_model_recursive_prop(model_type, "preload_URLs")
if len(preloadURLs) > 0:
return [fl.locate_file(os.path.basename(preloadURLs[0]))] , [1]
+ elif base_model_type == "vace_ditto_14B":
+ preloadURLs = get_model_recursive_prop(model_type, "preload_URLs")
+ model_mode = int(model_mode)
+ if len(preloadURLs) > model_mode:
+ return [fl.locate_file(os.path.basename(preloadURLs[model_mode]))] , [1]
return [], []
diff --git a/models/wan/configs/vace_ditto_14B.json b/models/wan/configs/vace_ditto_14B.json
new file mode 100644
index 0000000..e48a816
--- /dev/null
+++ b/models/wan/configs/vace_ditto_14B.json
@@ -0,0 +1,16 @@
+{
+ "_class_name": "VaceWanModel",
+ "_diffusers_version": "0.30.0",
+ "dim": 5120,
+ "eps": 1e-06,
+ "ffn_dim": 13824,
+ "freq_dim": 256,
+ "in_dim": 16,
+ "model_type": "t2v",
+ "num_heads": 40,
+ "num_layers": 40,
+ "out_dim": 16,
+ "text_len": 512,
+ "vace_layers": [0, 5, 10, 15, 20, 25, 30, 35],
+ "vace_in_dim": 96
+}
diff --git a/models/wan/modules/model.py b/models/wan/modules/model.py
index c90540f..7485c01 100644
--- a/models/wan/modules/model.py
+++ b/models/wan/modules/model.py
@@ -900,8 +900,8 @@ class WanModel(ModelMixin, ConfigMixin):
if first == None:
return sd
- new_sd = {}
+ # new_sd = {}
# for k,v in sd.items():
# if k.endswith("modulation.diff"):
# pass
@@ -914,6 +914,25 @@ class WanModel(ModelMixin, ConfigMixin):
# for k,v in sd.items():
# new_sd["diffusion_model." + k] = v
# sd = new_sd
+ if ".default." in first:
+ new_sd = {}
+ for k,v in sd.items():
+ k = k.replace(".default.", ".")
+ new_sd[k] = v
+ sd = new_sd
+
+ if first.startswith("vace_blocks."):
+ new_sd = {}
+ for k,v in sd.items():
+ if k.startswith("vace_blocks."):
+ l = k.split(".")
+ block_no = self.vace_layers[int(l[1])]
+ l[0] = "blocks." + str(block_no)
+ l[1] = "vace"
+ k = ".".join(l)
+ print(k)
+ new_sd[k] = v
+ sd = new_sd
if first.startswith("lora_unet_"):
new_sd = {}
diff --git a/models/wan/wan_handler.py b/models/wan/wan_handler.py
index e4cd602..853ac1c 100644
--- a/models/wan/wan_handler.py
+++ b/models/wan/wan_handler.py
@@ -4,7 +4,7 @@ import gradio as gr
from shared.utils import files_locator as fl
def test_vace(base_model_type):
- return base_model_type in ["vace_14B", "vace_14B_2_2", "vace_1.3B", "vace_multitalk_14B", "vace_standin_14B", "vace_lynx_14B"]
+ return base_model_type in ["vace_14B", "vace_14B_2_2", "vace_1.3B", "vace_multitalk_14B", "vace_standin_14B", "vace_lynx_14B", "vace_ditto_14B"]
def test_class_i2v(base_model_type):
return base_model_type in ["i2v", "i2v_2_2", "fun_inp_1.3B", "fun_inp", "flf2v_720p", "fantasy", "multitalk", "infinitetalk", "i2v_2_2_multitalk", "animate" ]
@@ -37,7 +37,7 @@ class family_handler():
@staticmethod
def query_supported_types():
return ["multitalk", "infinitetalk", "fantasy", "vace_14B", "vace_14B_2_2", "vace_multitalk_14B", "vace_standin_14B", "vace_lynx_14B",
- "t2v_1.3B", "standin", "lynx_lite", "lynx", "t2v", "t2v_2_2", "vace_1.3B", "phantom_1.3B", "phantom_14B",
+ "t2v_1.3B", "standin", "lynx_lite", "lynx", "t2v", "t2v_2_2", "vace_1.3B", "vace_ditto_14B", "phantom_1.3B", "phantom_14B",
"recam_1.3B", "animate", "alpha", "alpha_lynx",
"i2v", "i2v_2_2", "i2v_2_2_multitalk", "ti2v_2_2", "lucy_edit", "flf2v_720p", "fun_inp_1.3B", "fun_inp"]
@@ -332,7 +332,17 @@ class family_handler():
("Positioned Frames followed by People / Objects (if any) then a Face", "FI")]
extra_model_def["background_removal_label"]= "Remove Backgrounds behind People / Objects, keep it for Landscape, Lynx Face or Positioned Frames"
extra_model_def["no_processing_on_last_images_refs"] = 1
-
+ if base_model_type in ["vace_ditto_14B"]:
+ del extra_model_def["guide_preprocessing"], extra_model_def["image_ref_choices"], extra_model_def["video_guide_outpainting"]
+ extra_model_def["mask_preprocessing"] = { "selection": ["", "A"], }
+ extra_model_def["model_modes"] = {
+ "choices": [
+ ("Global", 0),
+ ("Global Style", 1),
+ ("Sim 2 Real", 2)],
+ "default": 0,
+ "label" : "Ditto Process"
+ }
if (not vace_class) and standin:
@@ -655,6 +665,10 @@ class family_handler():
"mask_expand": 20,
"audio_prompt_type": "R",
})
+ elif base_model_type in ["vace_ditto_14B"]:
+ ui_defaults.update({
+ "video_prompt_type": "V",
+ })
if text_oneframe_overlap(base_model_type):
ui_defaults["sliding_window_overlap"] = 1
diff --git a/requirements.txt b/requirements.txt
index 22bac3b..db9f5f0 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -52,7 +52,7 @@ pydantic==2.10.6
# Math & modeling
torchdiffeq>=0.2.5
tensordict>=0.6.1
-mmgp==3.6.6
+mmgp==3.6.7
peft==0.15.0
matplotlib
diff --git a/wgp.py b/wgp.py
index 049242e..5a951f4 100644
--- a/wgp.py
+++ b/wgp.py
@@ -73,8 +73,8 @@ from collections import defaultdict
global_queue_ref = []
AUTOSAVE_FILENAME = "queue.zip"
PROMPT_VARS_MAX = 10
-target_mmgp_version = "3.6.6"
-WanGP_version = "9.01"
+target_mmgp_version = "3.6.7"
+WanGP_version = "9.10"
settings_version = 2.39
max_source_video_frames = 3000
prompt_enhancer_image_caption_model, prompt_enhancer_image_caption_processor, prompt_enhancer_llm_model, prompt_enhancer_llm_tokenizer = None, None, None, None
@@ -3469,7 +3469,7 @@ def refresh_gallery(state): #, msg
queue = gen.get("queue", [])
abort_interactive = not gen.get("abort", False)
if not in_progress or len(queue) == 0:
- return gr.Gallery(value = file_list) if last_was_audio else gr.Gallery(selected_index=choice, value = file_list), *pack_audio_gallery_state(audio_file_list, audio_choice), gr.HTML("", visible= False), gr.Button(visible=True), gr.Button(visible=False), gr.Row(visible=False), gr.Row(visible=False), update_queue_data(queue), gr.Button(interactive= abort_interactive), gr.Button(visible= False)
+ return gr.Gallery(value = file_list) if last_was_audio else gr.Gallery(selected_index=choice, value = file_list), gr.update() if last_was_audio else choice, *pack_audio_gallery_state(audio_file_list, audio_choice), gr.HTML("", visible= False), gr.Button(visible=True), gr.Button(visible=False), gr.Row(visible=False), gr.Row(visible=False), update_queue_data(queue), gr.Button(interactive= abort_interactive), gr.Button(visible= False)
else:
task = queue[0]
prompt = task["prompt"]
@@ -3527,7 +3527,11 @@ def refresh_gallery(state): #, msg
gen_buttons_visible = True
html = f"<TABLE WIDTH=100% ID=PINFO style='{table_style}'><TR style='height:140px'><TD width=100% style='{table_style}'>" + prompt + "</TD>" + thumbnails + "</TR></TABLE>"
html_output = gr.HTML(html, visible= True)
- return gr.Gallery(value = file_list) if last_was_audio else gr.Gallery(selected_index=choice, value = file_list), *pack_audio_gallery_state(audio_file_list, audio_choice), html_output, gr.Button(visible=False), gr.Button(visible=True), gr.Row(visible=True), gr.Row(visible= gen_buttons_visible), update_queue_data(queue), gr.Button(interactive= abort_interactive), gr.Button(visible= onemorewindow_visible)
+ if last_was_audio:
+ audio_choice = max(0, audio_choice)
+ else:
+ choice = max(0, choice)
+ return gr.Gallery(value = file_list) if last_was_audio else gr.Gallery(selected_index=choice, value = file_list), gr.update() if last_was_audio else choice, *pack_audio_gallery_state(audio_file_list, audio_choice), html_output, gr.Button(visible=False), gr.Button(visible=True), gr.Row(visible=True), gr.Row(visible= gen_buttons_visible), update_queue_data(queue), gr.Button(interactive= abort_interactive), gr.Button(visible= onemorewindow_visible)
@@ -6992,7 +6996,8 @@ def use_video_settings(state, input_file_list, choice, source):
if any_audio:
input_file_list = unpack_audio_list(input_file_list)
file_list, file_settings_list = get_file_list(state, input_file_list, audio_files=any_audio)
- if choice != None and choice >=0 and len(file_list)>0:
+ if choice != None and len(file_list)>0:
+ choice= max(0, choice)
configs = file_settings_list[choice]
file_name= file_list[choice]
if configs == None:
@@ -7027,7 +7032,7 @@ def use_video_settings(state, input_file_list, choice, source):
else:
return *generate_dropdown_model_list(model_type), gr.update()
else:
- gr.Info(f"No Medium is Selected")
+ gr.Info(f"Please Select a File")
return gr.update(), gr.update(), gr.update(), gr.update()
loras_url_cache = None
@@ -9096,7 +9101,7 @@ def generate_video_tab(update_form = False, state_dict = None, ui_defaults = Non
if tab_id == 'generate':
output_trigger.change(refresh_gallery,
inputs = [state],
- outputs = [output, audio_files_paths, audio_file_selected, audio_gallery_refresh_trigger, gen_info, generate_btn, add_to_queue_btn, current_gen_column, current_gen_buttons_row, queue_html, abort_btn, onemorewindow_btn],
+ outputs = [output, last_choice, audio_files_paths, audio_file_selected, audio_gallery_refresh_trigger, gen_info, generate_btn, add_to_queue_btn, current_gen_column, current_gen_buttons_row, queue_html, abort_btn, onemorewindow_btn],
show_progress="hidden"
)