Comfy-Org/ComfyUI · error · ValueError

Krea2 expects conditioning with {self.txtlayers}x{self.txtdi

Error message

Krea2 expects conditioning with {self.txtlayers}x{self.txtdim}={self.txtlayers * self.txtdim} features (a {self.txtlayers}-layer Qwen3-VL stack) but got {fused}. Load the text encoder with CLIPLoader type 'krea2'.

What it means

Krea2's diffusion model consumes fused conditioning of shape (B, seq, txtlayers*txtdim) built from a stack of Qwen3-VL hidden layers. This error means the last conditioning dimension does not equal that product, i.e. the conditioning came from a different text encoder setup than the 'krea2' CLIPLoader type.

Source

Thrown at comfy/ldm/krea2/model.py:386

        for i, block in enumerate(self.blocks):
            transformer_options["block_index"] = i
            combined = block(combined, tvec, freqs, None, timestep_zero_index=timestep_zero_index, transformer_options=transformer_options)

        final = self.last(combined, t)
        del combined
        out = final[:, txtlen:txtlen + img_tokens, :]
        out = rearrange(out, "b (h w) (c ph pw) -> b c (h ph) (w pw)",
                        h=h_, w=w_, ph=patch, pw=patch, c=self.channels)
        out = out[:, :, :h_orig, :w_orig]  # crop padding back off
        if temporal:
            out = out.reshape(b5, t5, self.channels, h_orig, w_orig).movedim(1, 2)
        return out

    def _unpack_context(self, context):
        # context: (B, seq, txtlayers*txtdim) -> (B, seq, txtlayers, txtdim).
        b, seq, fused = context.shape
        if fused != self.txtlayers * self.txtdim:
            raise ValueError(
                f"Krea2 expects conditioning with {self.txtlayers}x{self.txtdim}={self.txtlayers * self.txtdim} "
                f"features (a {self.txtlayers}-layer Qwen3-VL stack) but got {fused}. "
                f"Load the text encoder with CLIPLoader type 'krea2'."
            )
        return context.reshape(b, seq, self.txtlayers, self.txtdim)

View on GitHub (pinned to 1c6d8d45b3)

Solutions

1. Load the text encoder with CLIPLoader type 'krea2' so it emits the multi-layer Qwen3-VL fused features
2. Re-encode all prompts with the krea2 encoder after switching; do not reuse conditioning cached from another encoder
3. Verify context.shape[-1] equals the model's txtlayers*txtdim before sampling

Example fix

# before
clip = CLIPLoader("qwen3-vl.safetensors", type="stable_diffusion")
# after
clip = CLIPLoader("qwen3-vl.safetensors", type="krea2")

expected = model.txtlayers * model.txtdim
assert conditioning.shape[-1] == expected, (conditioning.shape, expected)

def is_krea2_conditioning(cond: torch.Tensor, model) -> bool:
    return cond.dim() == 3 and cond.shape[-1] == model.txtlayers * model.txtdim

When it happens

Trigger: Feeding conditioning from a standard single-layer CLIP/T5 text encoder output (features = encoder dim, not txtlayers*txtdim), or connecting a Krea2 UNet with a generic CLIPLoader instead of type 'krea2'.

Common situations: Wiring the Krea2 model node to a default text encoding path, using cached conditioning from another model, or loading the text encoder with the wrong loader type.

Related errors

• JoyImageTEModel: encoded sequence length {out.shape[1]} is s
• y is None, did you try using a controlnet for SDXL on SD1?
• Need at least {require_count} hooks to combine, but only had
• Input img and txt tensors must have 3 dimensions.
• Input txt tensors must have 3 dimensions.

AI-assisted analysis of Comfy-Org/ComfyUI@1c6d8d45b3 (2026-08-14). Data as JSON: /api/errors/6cae5170e690f1d2. Report an issue: GitHub.