lllyasviel/Fooocus · error · ValueError

Wrong params!

Error message

Wrong params!

What it means

In codeformer's VQAutoEncoder.load_state_dict path, the vqgan checkpoint loaded with weights_only=True must contain either a 'params_ema' or 'params' top-level key. If neither key exists, the file is not a CodeFormer VQGAN weights file in the expected format and ValueError('Wrong params!') is raised.

Source

Thrown at ldm_patched/pfn/architecture/face/codeformer.py:392

            )
        self.generator = Generator(
            nf, ch_mult, res_blocks, img_size, attn_resolutions, emb_dim
        )

        if model_path is not None:
            chkpt = torch.load(model_path, map_location="cpu", weights_only=True)
            if "params_ema" in chkpt:
                self.load_state_dict(
                    torch.load(model_path, map_location="cpu", weights_only=True)["params_ema"]
                )
                logger.info(f"vqgan is loaded from: {model_path} [params_ema]")
            elif "params" in chkpt:
                self.load_state_dict(
                    torch.load(model_path, map_location="cpu", weights_only=True)["params"]
                )
                logger.info(f"vqgan is loaded from: {model_path} [params]")
            else:
                raise ValueError("Wrong params!")

    def forward(self, x):
        x = self.encoder(x)
        quant, codebook_loss, quant_stats = self.quantize(x)
        x = self.generator(quant)
        return x, codebook_loss, quant_stats


def calc_mean_std(feat, eps=1e-5):
    """Calculate mean and std for adaptive_instance_normalization.
    Args:
        feat (Tensor): 4D tensor.
        eps (float): A small value added to the variance to avoid
            divide-by-zero. Default: 1e-5.
    """
    size = feat.size()
    assert len(size) == 4, "The input feature should be 4D tensor."
    b, c = size[:2]

View on GitHub (pinned to ae05379cc9)

Solutions

1. Download the official CodeFormer VQGAN weights (vqgan_codeformer.pth) which contain 'params_ema'
2. If your file is a raw state dict, re-wrap it: torch.save({'params': sd}, path)
3. Verify with torch.load(..., weights_only=True) which top-level key exists before pointing CodeFormer at the file

Example fix

import torch

ckpt = torch.load(model_path, map_location='cpu', weights_only=True)
# before: neither 'params_ema' nor 'params' -> ValueError: Wrong params!
# after:
if 'params_ema' not in ckpt and 'params' not in ckpt:
    torch.save({'params': ckpt}, model_path)  # re-wrap raw state dict
    ckpt = torch.load(model_path, map_location='cpu', weights_only=True)

import torch

def is_codeformer_vqgan(path) -> bool:
    try:
        ckpt = torch.load(path, map_location='cpu', weights_only=True)
    except Exception:
        return False
    return isinstance(ckpt, dict) and ('params_ema' in ckpt or 'params' in ckpt)

if not is_codeformer_vqgan(vqgan_path):
    raise SystemExit(f'{vqgan_path} is not a CodeFormer VQGAN file (needs params/params_ema key)')

def is_vqgan_ckpt(obj) -> bool:
    return isinstance(obj, dict) and ('params_ema' in obj or 'params' in obj)

try:
    restorer = CodeFormer(vqgan_path, codeformer_path, ...)
except ValueError as e:
    if str(e) == 'Wrong params!':
        raise ModelFileError(f'{vqgan_path}: expected keys params/params_ema missing - wrong or corrupt CodeFormer VQGAN file') from e
    raise

When it happens

Trigger: Calling the CodeFormer face-restoration constructor with fidelity_ckpt/vqgan model_path pointing at the wrong artifact: the codeformer prompting/decoder weights instead of the VQGAN, a raw state dict, or a truncated download. torch.load succeeds, but the dict lacks both keys.

Common situations: Mixing up codeformer.pth and vqgan_codeformer.pth in the models folder; using an officially reformatted or community-quantized file that dropped the 'params'/'params_ema' wrapper; interrupted downloads.

Related errors

• checkpoint url or path is invalid
• checkpoint url or path is invalid
• CORRUPTED MODEL: one of the q-k-v values for the text encode
• ERROR: Could not detect model type of: {}
• normalize should be True if scale is passed

AI-assisted analysis of lllyasviel/Fooocus@ae05379cc9 (2026-08-15). Data as JSON: /api/errors/083a9507de7ac05d. Report an issue: GitHub.