Tiled KSampler

Documentation

• Class name: Tiled KSampler
• Category: Sampling/Tiled
• Output node: False

The Tiled KSampler node is designed for advanced sampling in generative models, specifically tailored for handling tiled inputs. It incorporates circular padding and customizable sampling strategies to generate high-quality latent representations.

Input types

Required

• model
  • The generative model used for sampling. It's crucial for defining the architecture and parameters that will be used during the sampling process.
  • Comfy dtype: MODEL
  • Python dtype: torch.nn.Module
• seed
  • A seed value to ensure reproducibility of the sampling process. It influences the randomness of the generated samples.
  • Comfy dtype: INT
  • Python dtype: int
• tiling
  • A flag indicating whether circular padding should be applied, enhancing the quality of tiled samples.
  • Comfy dtype: INT
  • Python dtype: int
• steps
  • The number of steps to perform during the sampling process, affecting the detail and quality of the generated samples.
  • Comfy dtype: INT
  • Python dtype: int
• cfg
  • A configuration parameter influencing the sampling process, allowing for fine-tuning of the generative model's behavior.
  • Comfy dtype: FLOAT
  • Python dtype: float
• sampler_name
  • The name of the sampler to use, determining the specific sampling strategy employed.
  • Comfy dtype: COMBO[STRING]
  • Python dtype: str
• scheduler
  • The scheduler to use, controlling the progression of sampling steps and potentially improving sample quality.
  • Comfy dtype: COMBO[STRING]
  • Python dtype: str
• positive
  • Positive conditioning information to guide the sampling towards desired characteristics.
  • Comfy dtype: CONDITIONING
  • Python dtype: Dict[str, torch.Tensor]
• negative
  • Negative conditioning information to steer the sampling away from undesired characteristics.
  • Comfy dtype: CONDITIONING
  • Python dtype: Dict[str, torch.Tensor]
• latent_image
  • An initial latent image to start the sampling from, allowing for modifications or enhancements of existing samples.
  • Comfy dtype: LATENT
  • Python dtype: Dict[str, torch.Tensor]
• denoise
  • A parameter to control the denoising strength during sampling, affecting the clarity and sharpness of the generated samples.
  • Comfy dtype: FLOAT
  • Python dtype: float

Output types

• latent
  • Comfy dtype: LATENT
  • The generated latent representation, which can be further processed or converted into a final image.
  • Python dtype: Dict[str, torch.Tensor]

Usage tips

• Infra type: GPU
• Common nodes: unknown

Source code

class Tiled_KSampler:
    @classmethod
    def INPUT_TYPES(cls):
        return {"required":
                {"model": ("MODEL", ),
                 "seed": ("INT", {"default": 0, "min": 0, "max": 0xffffffffffffffff}),
                 "tiling": ("INT", {"default": 1, "min": 0, "max": 1}),
                 "steps": ("INT", {"default": 20, "min": 1, "max": 10000}),
                 "cfg": ("FLOAT", {"default": 8.0, "min": 0.0, "max": 100.0}),
                 "sampler_name": (comfy.samplers.KSampler.SAMPLERS, ),
                 "scheduler": (comfy.samplers.KSampler.SCHEDULERS, ),
                 "positive": ("CONDITIONING", ),
                 "negative": ("CONDITIONING", ),
                 "latent_image": ("LATENT", ),
                 "denoise": ("FLOAT", {"default": 1.0, "min": 0.0, "max": 1.0, "step": 0.01}),
                 }}

    RETURN_TYPES = ("LATENT",)
    FUNCTION = "sample"

    CATEGORY = "Sampling/Tiled"
    def apply_circular(self, model, enable):
        for layer in [layer for layer in model.modules() if isinstance(layer, torch.nn.Conv2d)]:
            layer.padding_mode = 'circular' if enable else 'zeros'

    def sample(self, model, seed, tiling, steps, cfg, sampler_name, scheduler, positive, negative, latent_image, denoise=1.0):
        self.apply_circular(model.model, tiling == 1)
        return nodes.common_ksampler(model, seed, steps, cfg, sampler_name, scheduler, positive, negative, latent_image, denoise=denoise)