Mikey Sampler Tiled Advanced Base Only¶
Documentation¶
- Class name:
MikeySamplerTiledAdvancedBaseOnly - Category:
Mikey/Sampling - Output node:
False
This node is designed for advanced tiled sampling without the need for smooth steps or a refiner. It focuses on generating or processing images in a tiled manner, optimizing for scenarios where seamless integration of tiles is crucial without additional refinement steps.
Input types¶
Required¶
base_model- Specifies the base model used for sampling, setting the foundation for the generation process.
- Comfy dtype:
MODEL - Python dtype:
str
samples- The initial latent samples to be processed or refined through the sampling process.
- Comfy dtype:
LATENT - Python dtype:
torch.Tensor
vae- The variational autoencoder used alongside the base model to process or refine samples.
- Comfy dtype:
VAE - Python dtype:
torch.nn.Module
positive_cond_base- Positive conditioning for the base model to guide the sampling towards desired attributes.
- Comfy dtype:
CONDITIONING - Python dtype:
str
negative_cond_base- Negative conditioning for the base model to steer the sampling away from undesired attributes.
- Comfy dtype:
CONDITIONING - Python dtype:
str
model_name- The name of the model, typically used to identify different models within a framework or library.
- Comfy dtype:
COMBO[STRING] - Python dtype:
List[str]
seed- Seed for random number generation, ensuring reproducibility across sampling runs.
- Comfy dtype:
INT - Python dtype:
int
denoise_image- Specifies the degree of denoising applied to the image, affecting the clarity and quality of the output.
- Comfy dtype:
FLOAT - Python dtype:
float
steps- The number of steps to run in the sampling process, affecting the detail and quality of the generated image.
- Comfy dtype:
INT - Python dtype:
int
cfg- Controls the conditioning factor, influencing the generation's adherence to the given conditions.
- Comfy dtype:
FLOAT - Python dtype:
float
sampler_name- The name of the sampler used, affecting the diversity and quality of generated samples.
- Comfy dtype:
COMBO[STRING] - Python dtype:
str
scheduler- The scheduler used to manage the sampling process, impacting the progression and variation of samples.
- Comfy dtype:
COMBO[STRING] - Python dtype:
str
upscale_by- The factor by which the image is upscaled, affecting the resolution and detail of the final output.
- Comfy dtype:
FLOAT - Python dtype:
float
tiler_denoise- Specifies the degree of denoising applied to each tile, affecting the consistency and quality of the tiled output.
- Comfy dtype:
FLOAT - Python dtype:
float
tile_size- Defines the size of each tile in the generated image, affecting the granularity of the output.
- Comfy dtype:
INT - Python dtype:
int
Optional¶
image_optional- An optional image input that can be used to influence the sampling process.
- Comfy dtype:
IMAGE - Python dtype:
Optional[torch.Tensor]
Output types¶
output_image- Comfy dtype:
IMAGE - The generated image after the tiled sampling process, reflecting the combined influence of all input parameters.
- Python dtype:
PIL.Image
- Comfy dtype:
Usage tips¶
- Infra type:
GPU - Common nodes: unknown
Source code¶
class MikeySamplerTiledAdvancedBaseOnly:
# there is no step skipped, so no smooth steps are required
# also no refiner for this
@classmethod
def INPUT_TYPES(s):
return {"required": {"base_model": ("MODEL",),
"samples": ("LATENT",), "vae": ("VAE",),
"positive_cond_base": ("CONDITIONING",), "negative_cond_base": ("CONDITIONING",),
"model_name": (folder_paths.get_filename_list("upscale_models"), ),
"seed": ("INT", {"default": 0, "min": 0, "max": 0xffffffffffffffff}),
"denoise_image": ("FLOAT", {"default": 1.0, "min": 0.0, "max": 1.0, "step": 0.01}),
"steps": ("INT", {"default": 30, "min": 1, "max": 1000}),
"cfg": ("FLOAT", {"default": 6.5, "min": 0.0, "max": 1000.0, "step": 0.1}),
"sampler_name": (comfy.samplers.KSampler.SAMPLERS, ),
"scheduler": (comfy.samplers.KSampler.SCHEDULERS, ),
"upscale_by": ("FLOAT", {"default": 1.0, "min": 0.1, "max": 10.0, "step": 0.1}),
"tiler_denoise": ("FLOAT", {"default": 0.25, "min": 0.0, "max": 1.0, "step": 0.05}),
"tile_size": ("INT", {"default": 1024, "min": 256, "max": 4096, "step": 64})},
"optional": {"image_optional": ("IMAGE",),}}
RETURN_TYPES = ('IMAGE', )
RETURN_NAMES = ('output_image', )
FUNCTION = 'run'
CATEGORY = 'Mikey/Sampling'
def phase_one(self, base_model, samples, positive_cond_base, negative_cond_base,
upscale_by, model_name, seed, vae, denoise_image,
steps, cfg, sampler_name, scheduler):
image_scaler = ImageScale()
vaedecoder = VAEDecode()
uml = UpscaleModelLoader()
upscale_model = uml.load_model(model_name)[0]
iuwm = ImageUpscaleWithModel()
# step 1 run base model
start_step = int(steps - (steps * denoise_image))
#print(f'base model start_step: {start_step}, last_step: {last_step}')
sample1 = common_ksampler(base_model, seed, steps, cfg, sampler_name, scheduler, positive_cond_base, negative_cond_base, samples,
start_step=start_step, last_step=steps, force_full_denoise=False)[0]
# step 3 upscale image using a simple AI image upscaler
pixels = vaedecoder.decode(vae, sample1)[0]
org_width, org_height = pixels.shape[2], pixels.shape[1]
img = iuwm.upscale(upscale_model, image=pixels)[0]
upscaled_width, upscaled_height = int(org_width * upscale_by // 8 * 8), int(org_height * upscale_by // 8 * 8)
img = image_scaler.upscale(img, 'nearest-exact', upscaled_width, upscaled_height, 'center')[0]
return img, upscaled_width, upscaled_height
def upscale_image(self, samples, vae,
upscale_by, model_name):
image_scaler = ImageScale()
vaedecoder = VAEDecode()
uml = UpscaleModelLoader()
upscale_model = uml.load_model(model_name)[0]
iuwm = ImageUpscaleWithModel()
# step 3 upscale image using a simple AI image upscaler
pixels = vaedecoder.decode(vae, samples)[0]
org_width, org_height = pixels.shape[2], pixels.shape[1]
img = iuwm.upscale(upscale_model, image=pixels)[0]
upscaled_width, upscaled_height = int(org_width * upscale_by // 8 * 8), int(org_height * upscale_by // 8 * 8)
img = image_scaler.upscale(img, 'nearest-exact', upscaled_width, upscaled_height, 'center')[0]
return img, upscaled_width, upscaled_height
def run(self, seed, base_model, vae, samples, positive_cond_base, negative_cond_base,
model_name, upscale_by=2.0, tiler_denoise=0.4, tile_size=1024,
upscale_method='normal', denoise_image=1.0, steps=30, cfg=6.5,
sampler_name='dpmpp_sde_gpu', scheduler='karras', image_optional=None):
# if image not none replace samples with decoded image
if image_optional is not None:
vaeencoder = VAEEncode()
samples = vaeencoder.encode(vae, image_optional)[0]
if denoise_image > 0:
# phase 1: run base, refiner, then upscaler model
img, upscaled_width, upscaled_height = self.phase_one(base_model, samples, positive_cond_base, negative_cond_base,
upscale_by, model_name, seed, vae, denoise_image,
steps, cfg, sampler_name, scheduler)
img = tensor2pil(img)
else:
img = self.upscale_image(samples, vae, upscale_by, model_name)
img = tensor2pil(img)
# phase 2: run tiler
tiled_image = run_tiler_for_steps(img, base_model, vae, seed, cfg, sampler_name, scheduler, positive_cond_base, negative_cond_base, steps, tiler_denoise, tile_size)
return (tiled_image, )