invoke-ai/InvokeAI · error · ValueError

Incompatible 'noise' and 'latents' shapes

Error message

Incompatible 'noise' and 'latents' shapes: ${latents.shape=} ${noise.shape=}

What it means

After resolving latents, the function checks that `noise` and `latents` have matching trailing dimensions (channel/height/width). A mismatch means the noise schedule and the initial sample describe different latent sizes, which would break the diffusion loop, so it raises with both shapes in the message.

Solutions

Resize the source image / re-encode latents so their HxW matches the noise dimensions (multiples of the latent scale factor)
Regenerate noise with width/height matching the latents
Confirm both come from the same model family (same latent channel count)

Example fix

// before
noise = NoiseInvocation(width=1024, height=1024)
latents = vae_encode(image_resized_to_768x512)
// after
noise = NoiseInvocation(width=1024, height=1024)
latents = vae_encode(image_resized_to_1024x1024)  # dims now match noise

Defensive patterns

Strategy: validation

Validate before calling

if noise is not None and latents is not None and noise.shape[1:] != latents.shape[1:]:
    raise ValueError(f"noise {noise.shape} and latents {latents.shape} trailing dims must match")

Type guard

def shapes_compatible(noise, latents) -> bool:
    return noise is None or latents is None or noise.shape[1:] == latents.shape[1:]

Try / catch

try:
    out = invocation.invoke(context)
except ValueError as e:
    if "Incompatible 'noise' and 'latents' shapes" in str(e):
        latents = resize_latents_to(latents, noise.shape[2:])
        out = invocation.invoke(context)
    else:
        raise

Prevention

Resize source images so VAE latents match the noise width/height
Keep latent dims multiples of 8 pixels (latent scale factor)
Never mix latents across models with different latent shapes

When it happens

Trigger: Providing noise generated for one resolution (e.g. 1024x1024 latents) together with latents from an image of a different size (e.g. 768x512 VAE output), or latents from a different model whose channel count differs.

Common situations: Img2img where the uploaded image wasn't resized to the declared width/height; mixing latents across SD1 (4ch) and SDXL (4ch different H/W) or other architectures; manually constructed tensors with wrong dimensions.

Understand the failure class

Background: Tensor shape mismatch errors ("must have shape", "expected shape ... got ..."): when tensor dimensions disagree with what an op or layer was told to expect — this error's family across 6 libraries.

Related errors

AI-assisted analysis of invoke-ai/InvokeAI@0b6a024f2f (2026-08-29). Data as JSON: /api/errors/6b871ba5a05d8973. Report an issue: GitHub.

Appendix: source

Thrown at invokeai/app/invocations/denoise_latents.py:848

        - Text-to-Image SDXL Refiner Denoising: `latents` is provided, `noise` is not.
        - Image-to-Image SDXL Refiner Denoising: `latents` is provided, `noise` is not.

        NOTE(ryand): I wrote this docstring, but I am not the original author of this code. There may be other workflows
        I haven't considered.
        """
        noise = None
        if noise_field is not None:
            noise = context.tensors.load(noise_field.latents_name)

        if latents_field is not None:
            latents = context.tensors.load(latents_field.latents_name)
        elif noise is not None:
            latents = torch.zeros_like(noise)
        else:
            raise ValueError("'latents' or 'noise' must be provided!")

        if noise is not None and noise.shape[1:] != latents.shape[1:]:
            raise ValueError(f"Incompatible 'noise' and 'latents' shapes: {latents.shape=} {noise.shape=}")

        # The seed comes from (in order of priority): the noise field, the latents field, or 0.
        seed = 0
        if noise_field is not None and noise_field.seed is not None:
            seed = noise_field.seed
        elif latents_field is not None and latents_field.seed is not None:
            seed = latents_field.seed
        else:
            seed = 0

        return seed, noise, latents

    def invoke(self, context: InvocationContext) -> LatentsOutput:
        if os.environ.get("USE_MODULAR_DENOISE", False):
            return self._new_invoke(context)
        else:
            return self._old_invoke(context)

View on GitHub (pinned to 0b6a024f2f)