invoke-ai/InvokeAI · error · TypeError

Expected AutoencoderKL or FluxAutoEncoder for Z-Image VAE, g

Error message

Expected AutoencoderKL or FluxAutoEncoder for Z-Image VAE, got {type(vae_info.model).__name__}. Ensure you are using a compatible VAE model.

What it means

In the Z-Image latents-to-image (decode) invocation, the loaded VAE is type-checked before decoding: it must be diffusers AutoencoderKL or FluxAutoEncoder. Any other VAE class means the selected model cannot decode Z-Image latents, so a TypeError is raised instead of producing garbage images. The check also determines the is_flux_vae branch used for decoding context.

Source

Thrown at invokeai/app/invocations/z_image_latents_to_image.py:50

    title="Latents to Image - Z-Image",
    tags=["latents", "image", "vae", "l2i", "z-image"],
    category="latents",
    version="1.1.0",
    classification=Classification.Prototype,
)
class ZImageLatentsToImageInvocation(BaseInvocation, WithMetadata, WithBoard):
    """Generates an image from latents using Z-Image VAE (supports both Diffusers and FLUX VAE)."""

    latents: LatentsField = InputField(description=FieldDescriptions.latents, input=Input.Connection)
    vae: VAEField = InputField(description=FieldDescriptions.vae, input=Input.Connection)

    @torch.no_grad()
    def invoke(self, context: InvocationContext) -> ImageOutput:
        latents = context.tensors.load(self.latents.latents_name)

        vae_info = context.models.load(self.vae.vae)
        if not isinstance(vae_info.model, (AutoencoderKL, FluxAutoEncoder)):
            raise TypeError(
                f"Expected AutoencoderKL or FluxAutoEncoder for Z-Image VAE, got {type(vae_info.model).__name__}. "
                "Ensure you are using a compatible VAE model."
            )

        is_flux_vae = isinstance(vae_info.model, FluxAutoEncoder)

        # Estimate working memory needed for VAE decode
        estimated_working_memory = estimate_vae_working_memory_flux(
            operation="decode",
            image_tensor=latents,
            vae=vae_info.model,
        )

        # FLUX VAE doesn't support seamless, so only apply for AutoencoderKL
        seamless_context = (
            nullcontext() if is_flux_vae else SeamlessExt.static_patch_model(vae_info.model, self.vae.seamless_axes)
        )

View on GitHub (pinned to 0b6a024f2f)

Solutions

1. Point the VAE field at the Z-Image VAE (AutoencoderKL) or a FLUX VAE (FluxAutoEncoder).
2. Use ZImageModelLoader with a Diffusers Z-Image source so the correct VAE submodel is resolved automatically.
3. Correct the model's base/type metadata in the Model Manager if the right file is misregistered.
4. Update InvokeAI if you believe this VAE should be supported.

Example fix

// before
vae = ModelField(id='sd3-vae')
// after
vae = ModelField(id='flux-vae')  # or z-image AutoencoderKL VAE

cfg = context.models.get_config(latents_to_image.vae.vae)
assert cfg.base in (BaseModelType.ZImage, BaseModelType.Flux), f"VAE {cfg.name} is {cfg.base}, not usable for Z-Image decode"

from diffusers import AutoencoderKL
from invokeai.backend.flux.vae import FluxAutoEncoder

def can_decode_zimage_latents(vae) -> bool:
    return isinstance(vae, (AutoencoderKL, FluxAutoEncoder))

try:
    out = z_image_l2i.invoke(context)
except TypeError as e:
    if "Expected AutoencoderKL or FluxAutoEncoder" in str(e):
        context.logger.error("Decode VAE is incompatible with Z-Image latents; select the Flux/Z-Image VAE.")
    else:
        raise

When it happens

Trigger: Running ZImageLatentsToImage with self.vae referencing a model whose class is neither AutoencoderKL nor FluxAutoEncoder, checked immediately after context.models.load(self.vae.vae).

Common situations: Using an SD/SDXL/SD3 VAE in a Z-Image workflow; graph templates reused across model families; VAE model misconfigured in the Model Manager; user manually pointing a decode node at a checkpoint's embedded wrong VAE.

Related errors

• Expected AutoencoderKL or FluxAutoEncoder, got {type(vae).__
• Expected AutoencoderKLWan or FluxAutoEncoder for Anima VAE,
• No VAE source provided. Either set 'VAE' to a FLUX VAE model
• Expected Qwen3Encoder_Checkpoint_Config, got {type(config)._
• Expected Qwen3Encoder_GGUF_Config, got {type(config).__name_

AI-assisted analysis of invoke-ai/InvokeAI@0b6a024f2f (2026-08-29). Data as JSON: /api/errors/4f4b8cfa76d66f21. Report an issue: GitHub.