ComfyUI Node: Wan SCAIL-2 Phase Sampler

Wan SCAIL-2 Phase Sampler:

The WanSCAIL2PhaseSampler is a sophisticated node designed for two-phase sampling, replicating the core logic of the SCAIL2LoopSampler. It is particularly useful in scenarios where you need to process video frames, as it uses pixel frames as anchor inputs and manages VAE encoding, anchor frame rewriting, and noise mask management internally. This node is intended to be used in conjunction with the WanSCAILToVideoMultiRef node, replacing the KSampler in the conditioning node's downstream. It is especially beneficial for handling previous frames from the tail end of a decoded chunk, allowing for a seamless transition between video segments. The two-phase sampling logic is based on the ComfyUI-Scail2-Sampler-Helper project, ensuring a robust and reliable performance.

Wan SCAIL-2 Phase Sampler Input Parameters:

model1

This parameter specifies the primary model used for sampling. It is crucial for defining the initial phase of the sampling process. The model should be compatible with the node's requirements to ensure accurate results.

positive

This parameter represents the positive conditioning input, which influences the sampling process by providing a positive bias to the model's output.

negative

The negative parameter serves as the negative conditioning input, counterbalancing the positive input to refine the sampling results.

vae

The VAE (Variational Autoencoder) parameter is used for encoding and decoding the latent space, playing a critical role in managing the anchor frames and ensuring the integrity of the sampling process.

latent

This input represents the latent space data, which is essential for the node's operation. It must originate from a compatible conditioning node like WanSCAILToVideoMultiRef.

width

Specifies the width of the input frames, impacting the resolution and aspect ratio of the processed output.

height

Defines the height of the input frames, similar to the width parameter, affecting the resolution and aspect ratio.

seed

The seed parameter is used for random number generation, ensuring reproducibility of the sampling process. It has a default value of 0 and can range from 0 to a very large number.

steps

This parameter determines the number of sampling steps, influencing the quality and detail of the output. It ranges from 1 to 10,000, with a default of 6.

cfg

The CFG (Classifier-Free Guidance) parameter adjusts the strength of the guidance during sampling, with a default value of 1.0 and a range from 0.0 to 100.0.

sampler_name

Specifies the name of the sampler to be used, with options provided by the ComfyUI framework. The default is "euler."

scheduler

Defines the scheduler type for the sampling process, with options available within the ComfyUI framework. The default is "normal."

denoise

This parameter controls the denoising strength, ranging from 0.0 to 1.0, with a default of 1.00, affecting the clarity of the output.

split_step

Determines the step at which the sampling process is split into two phases. A value of 0 indicates a single-phase process, while values greater than 0 enable two-phase sampling if previous frames are provided.

phase1_noise

Specifies the noise mask value for Phase 1, ranging from 0.0 (frozen) to 1.0 (free), with a default of 0.5, affecting the flexibility of the anchor frames.

phase2_noise

Defines the noise mask value for Phase 2, similar to phase1_noise, with a default of 0.1, influencing the anchor frames' behavior in the second phase.

previous_frame_count

Indicates how many frames from the tail of the previous frames should be used as anchors, with a default of 5 and a range from 1 to 4096.

model2

An optional parameter specifying a secondary model for Phase 2. If not provided, model1 is used.

previous_frames

An optional input representing the tail end of the decoded previous frames. If not connected, the node defaults to a single-phase process.

Wan SCAIL-2 Phase Sampler Output Parameters:

latent

The output latent parameter represents the processed latent space data after the two-phase sampling process. It is crucial for further processing or rendering, providing a refined and coherent representation of the input data.

Wan SCAIL-2 Phase Sampler Usage Tips:

• Ensure that the latent input is sourced from a compatible conditioning node like WanSCAILToVideoMultiRef to maintain the integrity of the sampling process.
• Adjust the split_step parameter to fine-tune the transition between the two phases, optimizing the node's performance for specific tasks or video segments.

Wan SCAIL-2 Phase Sampler Common Errors and Solutions:

"Invalid latent input"

• Explanation: This error occurs when the latent input is not sourced from a compatible conditioning node.
• Solution: Ensure that the latent input is connected to a node like WanSCAILToVideoMultiRef.

"Model not compatible"

• Explanation: The specified model is not compatible with the node's requirements.
• Solution: Verify that the model used in model1 (and optionally model2) is compatible with the node's specifications.

"Previous frames not provided for two-phase sampling"

• Explanation: The node is set to two-phase sampling mode, but previous_frames are not connected.
• Solution: Connect the previous_frames input or set split_step to 0 for single-phase sampling.