lllyasviel/Fooocus · error · Exception

invalid style model {}

Error message

invalid style model {}

What it means

load_style_model loads a file with safe_load=True and decides it is a style model ONLY if the key 'style_embedding' is present; any other content raises 'invalid style model'. Style models here are T2I-Adapter StyleAdapter checkpoints (the Fooocus/ComfyUI style-model format), and the key check is the entire format validation.

Source

Thrown at ldm_patched/modules/sd.py:296

    def get_sd(self):
        return self.first_stage_model.state_dict()

class StyleModel:
    def __init__(self, model, device="cpu"):
        self.model = model

    def get_cond(self, input):
        return self.model(input.last_hidden_state)


def load_style_model(ckpt_path):
    model_data = ldm_patched.modules.utils.load_torch_file(ckpt_path, safe_load=True)
    keys = model_data.keys()
    if "style_embedding" in keys:
        model = ldm_patched.t2ia.adapter.StyleAdapter(width=1024, context_dim=768, num_head=8, n_layes=3, num_token=8)
    else:
        raise Exception("invalid style model {}".format(ckpt_path))
    model.load_state_dict(model_data)
    return StyleModel(model)


def load_clip(ckpt_paths, embedding_directory=None):
    clip_data = []
    for p in ckpt_paths:
        clip_data.append(ldm_patched.modules.utils.load_torch_file(p, safe_load=True))

    class EmptyClass:
        pass

    for i in range(len(clip_data)):
        if "transformer.resblocks.0.ln_1.weight" in clip_data[i]:
            clip_data[i] = ldm_patched.modules.utils.transformers_convert(clip_data[i], "", "text_model.", 32)

    clip_target = EmptyClass()
    clip_target.params = {}

View on GitHub (pinned to ae05379cc9)

Solutions

1. Verify the file is an actual T2I-Adapter style model (e.g. the official 'style' T2I adapters used by Fooocus) and not a LoRA/CLIP/controlnet
2. Load the file and check for the 'style_embedding' key: if absent, it is not this format
3. Re-download the style model from a trusted source; a truncated download can also lose keys

Example fix

from ldm_patched.modules.utils import load_torch_file

sd = load_torch_file(path, safe_load=True)
# before: any file without 'style_embedding' -> Exception: invalid style model
# after: gate the call
if 'style_embedding' not in sd:
    raise SystemExit(f'{path} is not a T2I style model')
style_model = ldm_patched.modules.sd.load_style_model(path)

from ldm_patched.modules.utils import load_torch_file

def is_style_model(path):
    try:
        sd = load_torch_file(path, safe_load=True)
    except Exception:
        return False
    return 'style_embedding' in sd

if not is_style_model(user_path):
    skip_style_model(user_path)

def is_style_model_state(sd: dict) -> bool:
    return isinstance(sd, dict) and 'style_embedding' in sd

try:
    style = ldm_patched.modules.sd.load_style_model(path)
except Exception as e:
    if 'invalid style model' in str(e):
        show_hint(f'{path} is not a T2I style model (missing style_embedding key)')
    else:
        raise

When it happens

Trigger: Calling load_style_model(ckpt_path) with a path that is not a T2I style adapter: a LoRA, a CLIP vision model, a full checkpoint, or a random safetensors file. Also fires if the file is a style adapter saved under a different key layout by a newer/older exporter.

Common situations: User drops the wrong file into the style models folder (e.g. mistakes an ip-adapter or controlnet for a style model); file extension is .safetensors but content is a LoRA; drag-and-drop confusion in the Fooocus style model input.

Related errors

• provide num_res_blocks either as an int (globally constant)
• sigma_min and sigma_max must not be 0
• ERROR: Could not detect model type of: {}
• sam model {sam_model} does not exist.
• Folder path is not a valid directory.

AI-assisted analysis of lllyasviel/Fooocus@ae05379cc9 (2026-08-15). Data as JSON: /api/errors/3d3171a7d2cb45b4. Report an issue: GitHub.