Comfy-Org/ComfyUI · critical · RuntimeError

Attempt to create ChromaRadiance object without setting oper

Error message

Attempt to create ChromaRadiance object without setting operations

What it means

ChromaRadiance subclasses Chroma but must be built through ComfyUI's operations layer (comfy.ops / quant backends) because its modules rely on an operations object for all Linear/Conv construction; a plain nn.Module instantiation with operations=None would crash later inside operations.Linear. The constructor therefore fails fast with a RuntimeError the moment operations is missing. Per repo policy, operations is never optional for this model.

Source

Thrown at comfy/ldm/chroma_radiance/model.py:51

    nerf_max_freqs: int
    # Setting nerf_tile_size to 0 disables tiling.
    nerf_tile_size: int
    # Currently one of linear (legacy) or conv.
    nerf_final_head_type: str
    # None means use the same dtype as the model.
    nerf_embedder_dtype: Optional[torch.dtype]
    use_x0: bool
    # Use sequential txt_ids instead of zeros
    use_sequential_txt_ids: bool

class ChromaRadiance(Chroma):
    """
    Transformer model for flow matching on sequences.
    """

    def __init__(self, image_model=None, final_layer=True, dtype=None, device=None, operations=None, **kwargs):
        if operations is None:
            raise RuntimeError("Attempt to create ChromaRadiance object without setting operations")
        nn.Module.__init__(self)
        self.dtype = dtype
        params = ChromaRadianceParams(**kwargs)
        self.params = params
        self.patch_size = params.patch_size
        self.in_channels = params.in_channels
        self.out_channels = params.out_channels
        if params.hidden_size % params.num_heads != 0:
            raise ValueError(
                f"Hidden size {params.hidden_size} must be divisible by num_heads {params.num_heads}"
            )
        pe_dim = params.hidden_size // params.num_heads
        if sum(params.axes_dim) != pe_dim:
            raise ValueError(f"Got {params.axes_dim} but expected positional dim {pe_dim}")
        self.hidden_size = params.hidden_size
        self.num_heads = params.num_heads
        self.in_dim = params.in_dim
        self.out_dim = params.out_dim

View on GitHub (pinned to 1c6d8d45b3)

Solutions

Pass an operations object, e.g. operations=comfy.ops.cast_to or the operations used by your model backend
Prefer loading the model through ComfyUI's checkpoint loading / model detection so operations is injected automatically
If writing a custom model wrapper, mirror how other ComfyUI models receive operations from comfy.model_management

Example fix

# before
model = ChromaRadiance(**config)  # RuntimeError

# after
import comfy.ops
model = ChromaRadiance(dtype=dtype, device=device, operations=comfy.ops.disable_weight_init, **config)

Defensive patterns

Strategy: validation

Validate before calling

import comfy.ops
if operations is None:
    operations = comfy.ops.disable_weight_init  # or the appropriate backend ops
model = ChromaRadiance(dtype=dtype, device=device, operations=operations, **config)

Type guard

def has_operations(operations) -> bool:
    return operations is not None and all(hasattr(operations, a) for a in ("Linear", "Conv2d"))

Prevention

Construct ComfyUI models only through checkpoint loading / model detection
Never call model __init__ without an operations object

When it happens

Trigger: Calling ChromaRadiance(...) directly with the default operations=None, e.g. ChromaRadiance(**config) from a test or custom loader. Normal creation goes through comfy.supported_models / model detection, which always supplies operations (e.g. comfy.ops.cast_to or a quant-aware operations class).

Common situations: Custom scripts or unit tests instantiating the model class bare; a custom node building ChromaRadiance itself instead of going through the model detection/loading path; copying example code that predates the operations requirement.

Related errors

AI-assisted analysis of Comfy-Org/ComfyUI@1c6d8d45b3 (2026-08-14). Data as JSON: /api/errors/31a3eb1a5dc85fc3. Report an issue: GitHub.