Ultimate Affine KSampler

A ComfyUI node in image/upscaling with 32 inputs and 1 output.

By WASasquatch·Created 12 months ago·Updated 4 months ago· 5

image
model
positive
negative
vae
external_mask
options
noise_options
affine_schedule

IMAGE

◄seed0►

◄steps20►

◄cfg8.0►

◄sampler_nameeuler►

◄schedulersimple►

◄denoise0.20►

◄affine_interval1►

◄max_scale1.2000►

◄max_bias0.0000►

◄patternwhite_noise►

◄affine_seed0►

◄affine_seed_incrementfalse►

◄tile_width0►

◄tile_height0►

◄tile_overlap64►

◄tiled_decodefalse►

◄tiled_tile_size512►

◄tiled_overlap64►

◄tiled_temporal_size64►

◄tiled_temporal_overlap8►

◄batch_size0►

◄merge_frames_in_batchtrue►

◄vae_encode_batch_modeimages►

Categoryimage/upscaling

Inputs (32)

Name	Type	Default	Description
image	IMAGE		Input IMAGE or VIDEO tensor. Supports [B,H,W,C] or [B,F,H,W,C].
model	MODEL		Diffusion model to sample with.
positive	CONDITIONING		Positive prompt conditioning.
negative	CONDITIONING		Negative prompt conditioning.
vae	VAE		VAE used to encode/decode latents during USDU.
seed	INT	00–18446744073709550000	Seed for base sampler (noise).
steps	INT	201–10000	Number of denoising steps.
cfg	FLOAT	8.00–100	Classifier-free guidance scale. Can be a single float or a per-step list (short lists repeat last value).
sampler_name	COMBO	euler	Base sampler algorithm.
scheduler	COMBO	simple	Scheduler for sigma schedule.
denoise	FLOAT	0.200–1	Denoise fraction (<=1).
external_maskopt	IMAGE		Optional external mask to gate affine.
optionsopt	DICT		Base options for Affine (common/full options).
noise_optionsopt	DICT		Pattern-specific overrides layered onto 'options'.
affine_intervalopt	INT	11–100	Apply affine every N steps (1 = every step).
max_scaleopt	FLOAT	1.20000–2	Scale at schedule peak: 1 + (max_scale-1)*t.
max_biasopt	FLOAT	0.0000-2–2	Bias at schedule peak: max_bias*t.
patternopt	COMBO	white_noise	Affine mask pattern.
affine_seedopt	INT	00–18446744073709550000	Seed for affine mask generation (separate from sampler seed).
affine_seed_incrementopt	BOOLEAN	false	Increment affine seed after each application (temporal).
affine_scheduleopt	DICT		WASAffineScheduleOptions dict; interpreted over total steps with repeat-last behavior.
tile_widthopt	INT	0-1–16384	IMAGE-space sampling tile width in pixels. 0 disables IMAGE tiling (single pass). -1 auto-calculates proportional tile size ≤1024 pixels.
tile_heightopt	INT	0-1–16384	IMAGE-space sampling tile height in pixels. 0 disables IMAGE tiling (single pass). -1 auto-calculates proportional tile size ≤1024 pixels.
tile_overlapopt	INT	640–2048	IMAGE-space overlap between sampling tiles (pixels).
tiled_decodeopt	BOOLEAN	false	Use VAE tiled decode to reduce VRAM spikes for large outputs/video.
tiled_tile_sizeopt	INT	51264–8192	Target output tile size (pixels) for VAE tiled decode.
tiled_overlapopt	INT	640–1024	Output-space overlap (pixels) for tiled decode.
tiled_temporal_sizeopt	INT	640–4096	Temporal window size (frames) for video tiled decode. 0 disables temporal tiling.
tiled_temporal_overlapopt	INT	80–512	Temporal overlap (frames) for video tiled decode.
batch_sizeopt	INT	00–4096	Process the batch dimension in chunks of this size to reduce peak VRAM. 0 = process all at once. Applies to non-tiling path.
merge_frames_in_batchopt	BOOLEAN	true	If decoded IMAGE is 5D [B,F,H,W,C], merge F into batch for concat.
vae_encode_batch_modeopt	COMBO	images	images: treat batch dim as separate images [B,C,1,H,W]. frames: treat batch dim as video frames [1,C,B,H,W].

Outputs (1)

Name	Type	Description
IMAGE	IMAGE	—