Skip to content

ControlNet

ControlNet models add extra conditions (like edge maps, depth maps, human poses) to Stable Diffusion models, allowing for more precise control over the generated image structure. RBLN NPUs can accelerate ControlNet model inference using RBLN Optimum.

Supported NPU

Diffusers pipelines run on ATOM™ only. Rebel100™ does not support them yet.

Usage within Pipelines

Typically, you do not interact with RBLNControlNetModel directly. Instead, it is automatically loaded and managed as part of an RBLN ControlNet pipeline, such as RBLNStableDiffusionControlNetPipeline or RBLNStableDiffusionXLControlNetPipeline. The Cosmos variant, RBLNCosmosControlNetModel, is loaded the same way by RBLNCosmos2_5_TransferPipeline.

When configuring an RBLN ControlNet pipeline, you can pass specific settings for the ControlNet model via the controlnet argument in the pipeline's configuration object:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
from optimum.rbln import RBLNStableDiffusionControlNetPipelineConfig

# Example: Configure pipeline with specific batch size for the ControlNet
config = RBLNStableDiffusionControlNetPipelineConfig(
    batch_size=1,  # Pipeline inference batch size
    # ... other SD pipeline settings ...
    controlnet=dict(
        batch_size=2  # ControlNet batch size (e.g., doubled for CFG)
    ),
)

# ... load pipeline using this config ...

Refer to the ControlNet Pipeline documentation (Standard, SDXL) for details on pipeline usage and configuration, including handling guidance scale effects on the ControlNet's batch size.

API Reference

RBLNControlNetModel

Bases: RBLNModel

RBLN implementation of ControlNetModel for diffusion models.

This model is used to accelerate ControlNetModel models from diffusers library on RBLN NPUs.

This class inherits from [RBLNModel]. Check the superclass documentation for the generic methods the library implements for all its models.

forward(sample, timestep, encoder_hidden_states, controlnet_cond, conditioning_scale=1.0, added_cond_kwargs=None, return_dict=True, **kwargs)

Forward pass for the RBLN-optimized ControlNetModel.

Parameters:

Name Type Description Default
sample FloatTensor

The noisy input tensor.

 required
timestep Tensor | float | int

The number of timesteps to denoise an input.

 required
encoder_hidden_states Tensor

The encoder hidden states.

 required
controlnet_cond FloatTensor

The conditional input tensor of shape (batch_size, max_seq_len, hidden_size).

 required
conditioning_scale float

The scale factor for ControlNet outputs.

 1.0
added_cond_kwargs dict[str, Tensor]

Additional conditions for the Stable Diffusion XL UNet.

 None
return_dict bool

Whether or not to return a [~diffusers.models.controlnets.controlnet.ControlNetOutput] instead of a plain tuple

 True

Returns:

Type Description
ControlNetOutput | tuple

(~diffusers.models.controlnets.controlnet.ControlNetOutput, Tuple)

from_pretrained(model_id, export=None, rbln_config=None, **kwargs) classmethod

The from_pretrained() function is utilized in its standard form as in the HuggingFace transformers library. User can use this function to load a pre-trained model from the HuggingFace library and convert it to a RBLN model to be run on RBLN NPUs.

Parameters:

Name Type Description Default
model_id str | Path

The model id of the pre-trained model to be loaded. It can be downloaded from the HuggingFace model hub or a local path, or a model id of a compiled model using the RBLN Compiler.

 required
export bool | None

A boolean flag to indicate whether the model should be compiled. If None, it will be determined based on the existence of the compiled model files in the model_id.

 None
rbln_config dict | RBLNModelConfig | None

Configuration for RBLN model compilation and runtime. This can be provided as a dictionary or an instance of the model's configuration class (e.g., RBLNLlamaForCausalLMConfig for Llama models). For detailed configuration options, see the specific model's configuration class documentation.

 None
kwargs Any

Additional keyword arguments. Arguments with the prefix rbln_ are passed to rbln_config, while the remaining arguments are passed to the HuggingFace library.

 {}

Returns:

Type Description
RBLNModel

A RBLN model instance ready for inference on RBLN NPU devices.

get_hf_class() classmethod

Lazily load and cache the HuggingFace class this model wraps.

The class is looked up in hf_library_name under the name without the RBLN prefix (e.g. RBLNLlamaForCausalLM -> LlamaForCausalLM).

save_pretrained(save_directory, push_to_hub=False, **kwargs)

Saves a model and its configuration file to a directory, so that it can be re-loaded using the [~optimum.rbln.modeling_base.RBLNBaseModel.from_pretrained] class method.

Parameters:

Name Type Description Default
save_directory str | Path

Directory where to save the model file.

 required
push_to_hub bool

Whether or not to push your model to the HuggingFace model hub after saving it.

 False

from_model(model, config=None, rbln_config=None, model_save_dir=None, subfolder='', **kwargs) classmethod

Converts and compiles a pre-trained HuggingFace library model into a RBLN model. This method performs the actual model conversion and compilation process.

Parameters:

Name Type Description Default
model PreTrainedModel | ModelMixin

The PyTorch model to be compiled. The object must be a HuggingFace transformers PreTrainedModel or a Diffusers component based on diffusers.ModelMixin, such as a VAE, UNet or ControlNet.

 required
config PretrainedConfig | None

The configuration object associated with the model.

 None
rbln_config RBLNModelConfig | dict | None

Configuration for RBLN model compilation and runtime. This can be provided as a dictionary or an instance of the model's configuration class (e.g., RBLNLlamaForCausalLMConfig for Llama models). For detailed configuration options, see the specific model's configuration class documentation.

 None
kwargs Any

Additional keyword arguments. Arguments with the prefix rbln_ are passed to rbln_config, while the remaining arguments are passed to the HuggingFace library.

 {}

The method performs the following steps:

  1. Compiles the PyTorch model into an optimized RBLN graph
  2. Configures the model for the specified NPU device
  3. Creates the necessary runtime objects if requested
  4. Saves the compiled model and configurations

Returns:

Type Description
RBLNModel

A RBLN model instance ready for inference on RBLN NPU devices.

RBLNControlNetModelConfig

Bases: RBLNModelConfig

Configuration class for RBLN ControlNet models.

__init__(batch_size=None, max_seq_len=None, unet_sample_size=None, vae_sample_size=None, text_model_hidden_size=None, **kwargs)

Parameters:

Name Type Description Default
batch_size int | None

The batch size for inference. Defaults to 1.

 None
max_seq_len int | None

Maximum sequence length for text inputs when used with cross-attention.

 None
unet_sample_size tuple[int, int] | None

The spatial dimensions (height, width) of the UNet output samples.

 None
vae_sample_size tuple[int, int] | None

The spatial dimensions (height, width) of the VAE input/output images.

 None
text_model_hidden_size int | None

Hidden size of the text encoder model used for conditioning.

 None
kwargs Any

Additional arguments passed to the parent RBLNModelConfig.

 {}

Raises:

Type Description
ValueError

If batch_size is not a positive integer.

from_pretrained(path, rbln_config=None, return_unused_kwargs=False, **kwargs) classmethod

Load a RBLNModelConfig from a path.

Parameters:

Name Type Description Default
path str

Path to the RBLNModelConfig file or directory containing the config file.

 required
rbln_config dict[str, Any] | None

Additional configuration to override.

 None
return_unused_kwargs bool

Whether to return unused kwargs.

 False
kwargs Any

Additional keyword arguments to override configuration values. Keys starting with 'rbln_' will have the prefix removed and be used to update the configuration.

 {}

Returns:

Name Type Description
RBLNModelConfig Self | tuple[Self, dict[str, Any]]

The loaded configuration instance.
Note

This method loads the configuration from the specified path and applies any provided overrides. If the loaded configuration class doesn't match the expected class, a warning will be logged.

Examples:

1
config = RBLNResNetForImageClassificationConfig.from_pretrained("/path/to/model")

initialize_submodule_config(submodule_config=None, force_kwargs=False, **kwargs)

Build the config of a submodule from the parent's runtime options and kwargs.

Returns the submodule config instance when its class is known (an instance was passed, or cls_name was given), otherwise the merged kwargs dict for the model class to resolve later.

get_load_overrides()

Extract the options that must cross the load boundary, for passing to from_pretrained.

Returns the explicitly-set runtime options (device, device_map, create_runtimes, activate_profiler, timeout) and the subclass-specific non-save attributes (load-behavior flags such as _load_visual_runtime, which are never serialized so they can only travel with the caller), recursively including submodules. Compile-time attributes are not included: when loading a compiled model they come from the artifact's rbln_config.json, and this dict is merged on top of it.

Returns:

Type Description
dict[str, Any]

Dictionary of load overrides, keyed like the rbln_config dict accepted by from_pretrained.

CosmosControlNetWrapper

Bases: CosmosPerFrameAdaLNMixin, Module

Compile graph: the ControlNet block stack. Embeddings are computed on the host.

RBLNCosmosControlNetModel

Bases: RBLNModel

RBLN implementation of CosmosControlNetModel (Cosmos-Transfer2.5).

The block stack runs on the NPU; patch/time/rope/projection embeddings run on the host, mirroring RBLNCosmosTransformer3DModel. Its forward keeps the upstream signature and returns the per-block control residuals that the transformer injects every controlnet_block_every_n blocks.

from_pretrained(model_id, export=None, rbln_config=None, **kwargs) classmethod

The from_pretrained() function is utilized in its standard form as in the HuggingFace transformers library. User can use this function to load a pre-trained model from the HuggingFace library and convert it to a RBLN model to be run on RBLN NPUs.

Parameters:

Name Type Description Default
model_id str | Path

The model id of the pre-trained model to be loaded. It can be downloaded from the HuggingFace model hub or a local path, or a model id of a compiled model using the RBLN Compiler.

 required
export bool | None

A boolean flag to indicate whether the model should be compiled. If None, it will be determined based on the existence of the compiled model files in the model_id.

 None
rbln_config dict | RBLNModelConfig | None

Configuration for RBLN model compilation and runtime. This can be provided as a dictionary or an instance of the model's configuration class (e.g., RBLNLlamaForCausalLMConfig for Llama models). For detailed configuration options, see the specific model's configuration class documentation.

 None
kwargs Any

Additional keyword arguments. Arguments with the prefix rbln_ are passed to rbln_config, while the remaining arguments are passed to the HuggingFace library.

 {}

Returns:

Type Description
RBLNModel

A RBLN model instance ready for inference on RBLN NPU devices.

get_hf_class() classmethod

Lazily load and cache the HuggingFace class this model wraps.

The class is looked up in hf_library_name under the name without the RBLN prefix (e.g. RBLNLlamaForCausalLM -> LlamaForCausalLM).

save_pretrained(save_directory, push_to_hub=False, **kwargs)

Saves a model and its configuration file to a directory, so that it can be re-loaded using the [~optimum.rbln.modeling_base.RBLNBaseModel.from_pretrained] class method.

Parameters:

Name Type Description Default
save_directory str | Path

Directory where to save the model file.

 required
push_to_hub bool

Whether or not to push your model to the HuggingFace model hub after saving it.

 False

from_model(model, config=None, rbln_config=None, model_save_dir=None, subfolder='', **kwargs) classmethod

Converts and compiles a pre-trained HuggingFace library model into a RBLN model. This method performs the actual model conversion and compilation process.

Parameters:

Name Type Description Default
model PreTrainedModel | ModelMixin

The PyTorch model to be compiled. The object must be a HuggingFace transformers PreTrainedModel or a Diffusers component based on diffusers.ModelMixin, such as a VAE, UNet or ControlNet.

 required
config PretrainedConfig | None

The configuration object associated with the model.

 None
rbln_config RBLNModelConfig | dict | None

Configuration for RBLN model compilation and runtime. This can be provided as a dictionary or an instance of the model's configuration class (e.g., RBLNLlamaForCausalLMConfig for Llama models). For detailed configuration options, see the specific model's configuration class documentation.

 None
kwargs Any

Additional keyword arguments. Arguments with the prefix rbln_ are passed to rbln_config, while the remaining arguments are passed to the HuggingFace library.

 {}

The method performs the following steps:

  1. Compiles the PyTorch model into an optimized RBLN graph
  2. Configures the model for the specified NPU device
  3. Creates the necessary runtime objects if requested
  4. Saves the compiled model and configurations

Returns:

Type Description
RBLNModel

A RBLN model instance ready for inference on RBLN NPU devices.

RBLNCosmosControlNetModelConfig

Bases: RBLNModelConfig

Configuration class for the RBLN Cosmos ControlNet (Cosmos-Transfer2.5).

The ControlNet shares its latent geometry with the transformer it drives, so the size fields mirror RBLNCosmosTransformer3DModelConfig and are filled by the pipeline config.

__init__(batch_size=None, num_frames=None, height=None, width=None, max_seq_len=None, embedding_dim=None, num_latent_frames=None, latent_height=None, latent_width=None, img_context_num_tokens=None, **kwargs)

Parameters:

Name Type Description Default
batch_size int | None

The batch size for inference. Defaults to 1.

 None
num_frames int | None

The number of frames per generated chunk.

 None
height int | None

The height in pixels of the generated video.

 None
width int | None

The width in pixels of the generated video.

 None
max_seq_len int | None

Maximum sequence length of prompt embeds.

 None
embedding_dim int | None

Embedding vector dimension of prompt embeds (after the ControlNet's own cross-attention projection input).

 None
num_latent_frames int | None

The number of frames in latent space.

 None
latent_height int | None

The height in pixels in latent space.

 None
latent_width int | None

The width in pixels in latent space.

 None
img_context_num_tokens int | None

Number of image-context tokens the graph was compiled with; filled from the model config (Transfer2.5 feeds a zero image context of this size when none is given).

 None
kwargs Any

Additional arguments passed to the parent RBLNModelConfig.

 {}

Raises:

Type Description
ValueError

If batch_size is not a positive integer.

from_pretrained(path, rbln_config=None, return_unused_kwargs=False, **kwargs) classmethod

Load a RBLNModelConfig from a path.

Parameters:

Name Type Description Default
path str

Path to the RBLNModelConfig file or directory containing the config file.

 required
rbln_config dict[str, Any] | None

Additional configuration to override.

 None
return_unused_kwargs bool

Whether to return unused kwargs.

 False
kwargs Any

Additional keyword arguments to override configuration values. Keys starting with 'rbln_' will have the prefix removed and be used to update the configuration.

 {}

Returns:

Name Type Description
RBLNModelConfig Self | tuple[Self, dict[str, Any]]

The loaded configuration instance.
Note

This method loads the configuration from the specified path and applies any provided overrides. If the loaded configuration class doesn't match the expected class, a warning will be logged.

Examples:

1
config = RBLNResNetForImageClassificationConfig.from_pretrained("/path/to/model")

initialize_submodule_config(submodule_config=None, force_kwargs=False, **kwargs)

Build the config of a submodule from the parent's runtime options and kwargs.

Returns the submodule config instance when its class is known (an instance was passed, or cls_name was given), otherwise the merged kwargs dict for the model class to resolve later.

get_load_overrides()

Extract the options that must cross the load boundary, for passing to from_pretrained.

Returns the explicitly-set runtime options (device, device_map, create_runtimes, activate_profiler, timeout) and the subclass-specific non-save attributes (load-behavior flags such as _load_visual_runtime, which are never serialized so they can only travel with the caller), recursively including submodules. Compile-time attributes are not included: when loading a compiled model they come from the artifact's rbln_config.json, and this dict is merged on top of it.

Returns:

Type Description
dict[str, Any]

Dictionary of load overrides, keyed like the rbln_config dict accepted by from_pretrained.