mirror of
https://github.com/vladmandic/automatic
synced 2026-08-27 23:51:00 +02:00
CalamitousFelicitousness
48fad8524e
Krea 2 is a 12.9B single-stream flow-matching DiT trained from scratch, using a Qwen3-VL-4B text encoder and the Qwen-Image VAE. The transformer is vendored as a diffusers ModelMixin whose module tree mirrors the checkpoint, so weights load with no key conversion; the pipeline ports the reference encode, flow-matching denoise, and VAE decode. The text encoder is shared at runtime via the existing dedup registry, so Base and Turbo reuse one Qwen3-VL-4B copy. Covers text-to-image, image-to-image, native LoRA, and the single-file UNET override. Also completes SD.Next's partial Qwen-Image VAE support (5D decode input and TAESD preview mapping) that K2 shares.
245 lines
12 KiB
Python
245 lines
12 KiB
Python
"""
|
|
Tiny AutoEncoder for Stable Diffusion
|
|
(DNN for encoding / decoding SD's latent space)
|
|
|
|
https://github.com/madebyollin/taesd
|
|
"""
|
|
import os
|
|
import time
|
|
import threading
|
|
from PIL import Image
|
|
import torch
|
|
from modules import devices, paths, shared
|
|
from modules.logger import log
|
|
|
|
|
|
debug = os.environ.get('SD_PREVIEW_DEBUG', None) is not None
|
|
|
|
|
|
TAESD_MODELS = {
|
|
'TAESD 1.3 Mocha Croissant': { 'fn': 'taesd_13_', 'uri': 'https://github.com/madebyollin/taesd/raw/7f572ca629c9b0d3c9f71140e5f501e09f9ea280', 'model': None },
|
|
'TAESD 1.2 Chocolate-Dipped Shortbread': { 'fn': 'taesd_12_', 'uri': 'https://github.com/madebyollin/taesd/raw/8909b44e3befaa0efa79c5791e4fe1c4d4f7884e', 'model': None },
|
|
'TAESD 1.1 Fruit Loops': { 'fn': 'taesd_11_', 'uri': 'https://github.com/madebyollin/taesd/raw/3e8a8a2ab4ad4079db60c1c7dc1379b4cc0c6b31', 'model': None },
|
|
'TAESD 1.0': { 'fn': 'taesd_10_', 'uri': 'https://github.com/madebyollin/taesd/raw/88012e67cf0454e6d90f98911fe9d4aef62add86', 'model': None },
|
|
'TAE FLUX.1': { 'fn': 'taef1.pth', 'uri': 'https://github.com/madebyollin/taesd/raw/main/taef1_decoder.pth', 'model': None },
|
|
'TAE FLUX.2': { 'fn': 'taef2.pth', 'uri': 'https://github.com/madebyollin/taesd/raw/main/taef2_decoder.pth', 'model': None },
|
|
'TAE SD3': { 'fn': 'taesd3.pth', 'uri': 'https://github.com/madebyollin/taesd/raw/main/taesd3_decoder.pth', 'model': None },
|
|
'TAE HunyuanVideo': { 'fn': 'taehv.pth', 'uri': 'https://github.com/madebyollin/taehv/raw/refs/heads/main/taehv.pth', 'model': None },
|
|
'TAE WanVideo': { 'fn': 'taew1.pth', 'uri': 'https://github.com/madebyollin/taehv/raw/refs/heads/main/taew2_1.pth', 'model': None },
|
|
'TAE MochiVideo': { 'fn': 'taem1.pth', 'uri': 'https://github.com/madebyollin/taem1/raw/refs/heads/main/taem1.pth', 'model': None },
|
|
}
|
|
CQYAN_MODELS = {
|
|
'Hybrid-Tiny SD': {
|
|
'sd': { 'repo': 'cqyan/hybrid-sd-tinyvae', 'model': None },
|
|
'sdxl': { 'repo': 'cqyan/hybrid-sd-tinyvae-xl', 'model': None },
|
|
},
|
|
'Hybrid-Small SD': {
|
|
'sd': { 'repo': 'cqyan/hybrid-sd-small-vae', 'model': None },
|
|
'sdxl': { 'repo': 'cqyan/hybrid-sd-small-vae-xl', 'model': None },
|
|
},
|
|
}
|
|
|
|
prev_warnings = False
|
|
first_run = True
|
|
prev_cls = ''
|
|
prev_type = ''
|
|
prev_model = ''
|
|
lock = threading.Lock()
|
|
|
|
|
|
def warn_once(msg, variant=None):
|
|
variant = variant or shared.opts.taesd_variant
|
|
global prev_warnings # pylint: disable=global-statement
|
|
if not prev_warnings:
|
|
prev_warnings = True
|
|
log.warning(f'Decode: type="taesd" variant="{variant}": {msg}')
|
|
return Image.new('RGB', (8, 8), color = (0, 0, 0))
|
|
|
|
|
|
def get_model(model_cls, variant=None):
|
|
if variant is not None:
|
|
pass
|
|
if model_cls in {'sd'}:
|
|
model_cls = 'sd'
|
|
variant = shared.opts.taesd_variant
|
|
elif model_cls in {'sdxl', 'ldm', 'pixartalpha'}:
|
|
model_cls = 'sdxl'
|
|
variant = shared.opts.taesd_variant
|
|
elif model_cls in {'pixartsigma', 'hunyuandit', 'omnigen', 'auraflow'}:
|
|
model_cls = 'sdxl'
|
|
variant = shared.opts.taesd_variant
|
|
elif model_cls in {'f1', 'h1', 'zimage', 'lumina2', 'chroma', 'longcat', 'omnigen2', 'flite', 'ovis', 'kandinsky5', 'glmimage', 'cogview3', 'cogview4', 'ultraflux'}:
|
|
model_cls = 'f1'
|
|
variant = 'TAE FLUX.1'
|
|
elif model_cls in {'f2', 'ernieimage', 'lens', 'ideogram4'}:
|
|
model_cls = 'f2'
|
|
variant = 'TAE FLUX.2'
|
|
elif model_cls in {'sd3'}:
|
|
variant = 'TAE SD3'
|
|
elif model_cls in {'wanai', 'qwen', 'chrono', 'cosmos', 'anima', 'fibo', 'joy', 'krea2'}:
|
|
variant = 'TAE WanVideo'
|
|
else:
|
|
warn_once(f'cls={shared.sd_model.__class__.__name__} type={shared.sd_model_type} unsuppported', variant=variant)
|
|
return model_cls, None
|
|
if debug:
|
|
log.debug(f'TAESD detect: cls={model_cls} variant={variant}')
|
|
return model_cls, variant
|
|
|
|
|
|
def load_model(model_type = 'decoder', variant = None, vae_file: str | None = None):
|
|
global prev_cls, prev_type, prev_model, prev_warnings # pylint: disable=global-statement
|
|
model_cls = shared.sd_model_type if shared.sd_loaded else None
|
|
if vae_file is not None and os.path.exists(vae_file):
|
|
model_cls = 'sdxl'
|
|
if model_cls is None or model_cls == 'none':
|
|
return None, variant
|
|
model_cls, variant = get_model(model_cls, variant)
|
|
if model_cls is None or variant is None:
|
|
return None, variant
|
|
folder = os.path.join(paths.models_path, "TAESD")
|
|
dtype = devices.dtype_vae if devices.dtype_vae != torch.bfloat16 else torch.float16 # taesd does not support bf16
|
|
os.makedirs(folder, exist_ok=True)
|
|
if variant.startswith('TAE'):
|
|
cfg = TAESD_MODELS[variant]
|
|
if (model_cls == prev_cls) and (model_type == prev_type) and (variant == prev_model) and (cfg['model'] is not None):
|
|
return cfg['model'], variant
|
|
fn = os.path.join(folder, cfg['fn'] + model_type + '_' + model_cls + '.pth')
|
|
if not os.path.exists(fn):
|
|
uri = cfg['uri']
|
|
if not uri.endswith('.pth'):
|
|
uri += '/tae' + model_cls + '_' + model_type + '.pth'
|
|
try:
|
|
torch.hub.download_url_to_file(uri, fn)
|
|
log.print() # new line
|
|
log.info(f'Decode: type="taesd" variant="{variant}": uri="{uri}" fn="{fn}" download')
|
|
except Exception as e:
|
|
warn_once(f'download uri={uri} {e}', variant=variant)
|
|
if os.path.exists(fn):
|
|
prev_cls = model_cls
|
|
prev_type = model_type
|
|
prev_model = variant
|
|
log.print() # new line
|
|
log.debug(f'Decode: type="taesd" variant="{variant}" fn="{fn}" layers={shared.opts.taesd_layers} load')
|
|
vae = None
|
|
if 'TAE HunyuanVideo' in variant:
|
|
from modules.taesd.taehv import TAEHV
|
|
vae = TAEHV(checkpoint_path=fn)
|
|
elif 'TAE WanVideo' in variant:
|
|
from modules.taesd.taehv import TAEHV
|
|
vae = TAEHV(checkpoint_path=fn)
|
|
elif 'TAE MochiVideo' in variant:
|
|
from modules.taesd.taem1 import TAEM1
|
|
vae = TAEM1(checkpoint_path=fn)
|
|
else:
|
|
from modules.taesd.taesd import TAESD
|
|
vae = TAESD(decoder_path=fn if model_type=='decoder' else None, encoder_path=fn if model_type=='encoder' else None)
|
|
"""
|
|
_vae = diffusers.AutoencoderKL()
|
|
from installer import Dot
|
|
_config = diffusers.AutoencoderKL().config.copy()
|
|
vae.config = Dot(_config) # set config for compatibility with standard vae
|
|
"""
|
|
if vae is not None:
|
|
prev_warnings = False # reset warnings for new model
|
|
vae = vae.to(devices.device, dtype=dtype)
|
|
TAESD_MODELS[variant]['model'] = vae
|
|
vae.config = {}
|
|
return vae, variant
|
|
elif variant.startswith('Hybrid'):
|
|
cfg = CQYAN_MODELS[variant].get(model_cls, None)
|
|
if (model_cls == prev_cls) and (model_type == prev_type) and (variant == prev_model) and (cfg['model'] is not None):
|
|
return cfg['model'], variant
|
|
if cfg is None:
|
|
warn_once(f'cls={shared.sd_model.__class__.__name__} type={model_cls} unsuppported', variant=variant)
|
|
return None, variant
|
|
repo = cfg['repo']
|
|
prev_cls = model_cls
|
|
prev_type = model_type
|
|
prev_model = variant
|
|
log.debug(f'Decode: type="taesd" variant="{variant}" id="{repo}" load')
|
|
if 'tiny' in repo:
|
|
from diffusers.models import AutoencoderTiny
|
|
vae = AutoencoderTiny.from_pretrained(repo, cache_dir=shared.opts.hfcache_dir, torch_dtype=dtype)
|
|
else:
|
|
from modules.taesd.hybrid_small import AutoencoderSmall
|
|
vae = AutoencoderSmall.from_pretrained(repo, cache_dir=shared.opts.hfcache_dir, torch_dtype=dtype)
|
|
vae = vae.to(devices.device, dtype=dtype)
|
|
CQYAN_MODELS[variant][model_cls]['model'] = vae
|
|
return vae, variant
|
|
elif variant is None:
|
|
warn_once(f'cls={shared.sd_model.__class__.__name__} type={model_cls} variant is none', variant=variant)
|
|
else:
|
|
warn_once(f'cls={shared.sd_model.__class__.__name__} type={model_cls} unsuppported', variant=variant)
|
|
return None, variant
|
|
|
|
|
|
def restore_preview_size(image, vae):
|
|
# TAESD (image) and TAEHV (video) drop spatial upsample blocks when taesd_layers < 3, shrinking output 2x/4x.
|
|
# Rescale spatial dims so preview size stays constant. Other taes (TAEM1, Hybrid) ignore taesd_layers, so skip them.
|
|
from modules.taesd.taesd import TAESD
|
|
from modules.taesd.taehv import TAEHV
|
|
layers = shared.opts.taesd_layers
|
|
if layers >= 3 or not isinstance(vae, (TAESD, TAEHV)) or not isinstance(image, torch.Tensor) or image.ndim < 3 or image.shape[-3] != 3:
|
|
return image
|
|
try:
|
|
frames = image.reshape(-1, *image.shape[-3:]) # flatten any leading dims to a batch of CHW frames
|
|
frames = torch.nn.functional.interpolate(frames, scale_factor=float(2 ** (3 - layers)), mode='bilinear', align_corners=False)
|
|
image = frames.reshape(*image.shape[:-2], frames.shape[-2], frames.shape[-1])
|
|
except Exception:
|
|
pass
|
|
return image
|
|
|
|
|
|
def decode(latents):
|
|
global first_run # pylint: disable=global-statement
|
|
with lock:
|
|
try:
|
|
vae, variant = load_model(model_type='decoder')
|
|
if vae is None or max(latents.shape) > 256: # safetey check of large tensors
|
|
return latents
|
|
except Exception as e:
|
|
# from modules import errors
|
|
# errors.display(e, 'taesd"')
|
|
return warn_once(f'load: {e}')
|
|
try:
|
|
with devices.inference_context():
|
|
t0 = time.time()
|
|
dtype = devices.dtype_vae if (devices.dtype_vae != torch.bfloat16) else torch.float16 # taesd does not support bf16
|
|
tensor = latents.unsqueeze(0) if len(latents.shape) == 3 else latents
|
|
tensor = tensor.detach().clone().to(devices.device, dtype=dtype)
|
|
if debug:
|
|
log.debug(f'Decode: type="taesd" variant="{variant}" input={latents.shape} tensor={tensor.shape}')
|
|
# Fallback: reshape packed 128-channel latents to 32 channels if not already unpacked
|
|
if (variant == 'TAE FLUX.2') and (len(tensor.shape) == 4) and (tensor.shape[1] == 128):
|
|
b, _c, h, w = tensor.shape
|
|
tensor = tensor.reshape(b, 32, h * 2, w * 2)
|
|
if variant.startswith('TAESD') or variant in {'TAE FLUX.1', 'TAE FLUX.2', 'TAE SD3'}:
|
|
image = vae.decoder(tensor).clamp(0, 1).detach()
|
|
image = image[0]
|
|
else:
|
|
image = vae.decode(tensor, return_dict=False)[0]
|
|
image = (image / 2.0 + 0.5).clamp(0, 1).detach()
|
|
image = restore_preview_size(image, vae)
|
|
t1 = time.time()
|
|
if (t1 - t0) > 3.0 and not first_run:
|
|
log.warning(f'Decode: type="taesd" variant="{variant}" long decode time={t1 - t0:.2f}')
|
|
first_run = False
|
|
return image
|
|
except Exception as e:
|
|
# from modules import errors
|
|
# errors.display(e, 'taesd"')
|
|
return warn_once(f'decode: {e}', variant=variant)
|
|
|
|
|
|
def encode(image):
|
|
with lock:
|
|
vae, variant = load_model(model_type='encoder')
|
|
if vae is None:
|
|
return image
|
|
try:
|
|
with devices.inference_context():
|
|
latents = vae.encoder(image)
|
|
return latents.detach()
|
|
except Exception as e:
|
|
return warn_once(f'encode: {e}', variant=variant)
|