mirror of
https://github.com/vladmandic/automatic
synced 2026-09-01 18:46:53 +02:00
CalamitousFelicitousness
10647c7b67
Wan 2.2 A14B ships a per-model boundary_ratio (0.9 I2V, 0.875 T2V) that selects the high- or low-noise expert per step. The video and base-model image loaders both load the shipped value; the slider override is applied at generation time in set_pipeline_args, the one point both paths pass through before invoking the pipeline. The denoising loop reads config.boundary_ratio each call, so tuning takes effect with no reload for video and base-model images alike. The slider defaults to -1, meaning use the model's value; 0 to 1 set the boundary explicitly. Single-expert stages stay load-time because they drop a transformer to free VRAM.
99 lines
5.0 KiB
Python
99 lines
5.0 KiB
Python
import transformers
|
|
import diffusers
|
|
from modules import shared, devices, sd_models, model_quant, sd_hijack_te, sd_hijack_vae
|
|
from modules.logger import log
|
|
from pipelines import generic
|
|
|
|
|
|
def init_text_encoder(repo_id, diffusers_load_config=None):
|
|
if diffusers_load_config is None:
|
|
diffusers_load_config = {}
|
|
load_args, quant_args = model_quant.get_dit_args(diffusers_load_config, module='TE', device_map=True)
|
|
repo_id = 'Wan-AI/Wan2.1-T2V-1.3B-Diffusers' if 'Wan2.' in repo_id else repo_id # always use shared umt5
|
|
log.debug(f'Load model: type=WanAI te="{repo_id}" quant="{model_quant.get_quant_type(quant_args)}" args={load_args}')
|
|
text_encoder = transformers.UMT5EncoderModel.from_pretrained(
|
|
repo_id,
|
|
subfolder="text_encoder",
|
|
cache_dir=shared.opts.hfcache_dir,
|
|
**load_args,
|
|
**quant_args,
|
|
)
|
|
if shared.opts.diffusers_offload_mode != 'none' and text_encoder is not None:
|
|
sd_models.move_model(text_encoder, devices.cpu)
|
|
return text_encoder
|
|
|
|
|
|
def load_wan(checkpoint_info, diffusers_load_config=None):
|
|
if diffusers_load_config is None:
|
|
diffusers_load_config = {}
|
|
repo_id = sd_models.path_to_repo(checkpoint_info)
|
|
sd_models.hf_auth_check(checkpoint_info)
|
|
transformer_cls = diffusers.WanVACETransformer3DModel if 'VACE' in repo_id else diffusers.WanTransformer3DModel
|
|
|
|
boundary_ratio = None
|
|
if 'a14b' in repo_id.lower() or 'fun-14b' in repo_id.lower():
|
|
if shared.opts.model_wan_stage == 'high noise' or shared.opts.model_wan_stage == 'first':
|
|
transformer = generic.load_transformer(repo_id, cls_name=transformer_cls, load_config=diffusers_load_config, subfolder='transformer')
|
|
transformer_2 = None
|
|
boundary_ratio = 0.0
|
|
elif shared.opts.model_wan_stage == 'low noise' or shared.opts.model_wan_stage == 'second':
|
|
transformer = None
|
|
transformer_2 = generic.load_transformer(repo_id, cls_name=transformer_cls, load_config=diffusers_load_config, subfolder='transformer_2')
|
|
boundary_ratio = 1000.0
|
|
elif shared.opts.model_wan_stage == 'combined' or shared.opts.model_wan_stage == 'both':
|
|
transformer = generic.load_transformer(repo_id, cls_name=transformer_cls, load_config=diffusers_load_config, subfolder='transformer')
|
|
transformer_2 = generic.load_transformer(repo_id, cls_name=transformer_cls, load_config=diffusers_load_config, subfolder='transformer_2')
|
|
# load with the checkpoint's boundary; the slider override is applied at runtime in set_pipeline_args
|
|
boundary_ratio = None
|
|
else:
|
|
log.error(f'Load model: type=WanAI stage="{shared.opts.model_wan_stage}" unsupported')
|
|
return None
|
|
else:
|
|
transformer = generic.load_transformer(repo_id, cls_name=transformer_cls, load_config=diffusers_load_config, subfolder='transformer')
|
|
transformer_2 = None
|
|
if repo_id is None or repo_id.lower() == 'none':
|
|
return None
|
|
|
|
text_encoder = init_text_encoder(repo_id, diffusers_load_config)
|
|
|
|
load_args, _quant_args = model_quant.get_dit_args(diffusers_load_config, module='Model')
|
|
|
|
if 'Wan2.2-I2V' in repo_id:
|
|
pipe_cls = diffusers.WanImageToVideoPipeline
|
|
diffusers.pipelines.auto_pipeline.AUTO_IMAGE2IMAGE_PIPELINES_MAPPING["wanai"] = diffusers.WanImageToVideoPipeline
|
|
elif 'Wan2.2-VACE' in repo_id:
|
|
pipe_cls = diffusers.WanVACEPipeline
|
|
diffusers.pipelines.auto_pipeline.AUTO_TEXT2IMAGE_PIPELINES_MAPPING["wanai"] = diffusers.WanVACEPipeline
|
|
diffusers.pipelines.auto_pipeline.AUTO_IMAGE2IMAGE_PIPELINES_MAPPING["wanai"] = diffusers.WanVACEPipeline
|
|
diffusers.pipelines.auto_pipeline.AUTO_INPAINT_PIPELINES_MAPPING["wanai"] = diffusers.WanVACEPipeline
|
|
else:
|
|
from pipelines.wan.wan_image import WanImagePipeline
|
|
pipe_cls = diffusers.WanPipeline
|
|
diffusers.pipelines.auto_pipeline.AUTO_TEXT2IMAGE_PIPELINES_MAPPING["wanai"] = diffusers.WanPipeline
|
|
diffusers.pipelines.auto_pipeline.AUTO_IMAGE2IMAGE_PIPELINES_MAPPING["wanai"] = WanImagePipeline
|
|
log.debug(f'Load model: type=WanAI model="{checkpoint_info.name}" repo="{repo_id}" cls={pipe_cls.__name__} offload={shared.opts.diffusers_offload_mode} dtype={devices.dtype} args={load_args} stage="{shared.opts.model_wan_stage}" boundary={boundary_ratio}')
|
|
wan_args = {
|
|
'transformer': transformer,
|
|
'transformer_2': transformer_2,
|
|
'text_encoder': text_encoder,
|
|
'cache_dir': shared.opts.diffusers_dir,
|
|
**load_args,
|
|
}
|
|
if boundary_ratio is not None: # omit so from_pretrained keeps the checkpoint's shipped boundary_ratio
|
|
wan_args['boundary_ratio'] = boundary_ratio
|
|
pipe = pipe_cls.from_pretrained(repo_id, **wan_args)
|
|
pipe.task_args = {
|
|
'num_frames': 1,
|
|
'output_type': 'np',
|
|
}
|
|
|
|
del text_encoder
|
|
del transformer
|
|
del transformer_2
|
|
|
|
sd_hijack_te.init_hijack(pipe)
|
|
sd_hijack_vae.init_hijack(pipe)
|
|
|
|
devices.torch_gc()
|
|
return pipe
|