Nodes/ComfyUI-ContextAnchoredTileRefine/Context-Anchored Tile Upscale (VL)
ComfyUI Node

Context-Anchored Tile Upscale (VL)

A ComfyUI node in image/upscaling with 19 inputs and 1 output.

By Blakeem·Created about a month ago·Updated 3 days ago· 17
Context-Anchored Tile Upscale (VL)
  • image
  • model
  • clip
  • vae
  • upscale_model
  • negative
  • IMAGE
seed0
sampler_namedpmpp_2m
schedulersgm_uniform
steps20
cfg3.5
denoise0.50
upscale_by2.00
max_tile_width1536
max_tile_height2048
context_anchor32
context_overlap32
anchor_sourcesource image
vlm_methodvision tokens and captions
Categoryimage/upscaling

Inputs (19)

NameTypeDefaultDescription
imageIMAGEThe image to upscale and then refine.
modelMODELThe diffusion model that denoises each tile.
clipCLIPMust be a vision-language text encoder (Krea 2 family). There is no positive prompt input, since each tile is conditioned on the image itself.
vaeVAEThe VAE that encodes and decodes each tile.
seedINT00–18446744073709550000Noise is drawn once for the entire image and then sliced for each tile.
sampler_nameCOMBOdpmpp_2m44 options: euler, euler_cfg_pp, euler_ancestral, euler_ancestral_cfg_pp, heun, heunpp2, +38
schedulerCOMBOsgm_uniform9 options: simple, sgm_uniform, karras, exponential, ddim_uniform, beta, +3
stepsINT201–10000
cfgFLOAT3.50–100
denoiseFLOAT0.500–1
upscale_byFLOAT2.000.01–8The upscale multiplier. The optional upscale_model runs first when one is connected.
max_tile_widthINT1536256–16384Hard cap on the width the model ever sees per sampled crop, including the context_overlap and context_anchor rings. Set to the largest width the model supports.
max_tile_heightINT2048256–16384Hard cap on the height the model ever sees per sampled crop, including the context_overlap and context_anchor rings. Set to the largest height the model supports.
context_anchorINT320–512Pixels around each tile that are frozen and shown to the model as context, then cropped away.
context_overlapINT320–512Overlapped context that is diffused from both sides and then blended. It anchors the tiles to each other, like context_anchor anchors each tile to its surroundings.
anchor_sourceCOMBOsource imageWhat fills context_anchor. source image keeps the result true to the input. live canvas adds more detail and drifts further from the input.
vlm_methodCOMBOvision tokens and captionsWhether each tile is conditioned on a caption of itself, on its slice of the entire image's vision encode, or on both. The name in parentheses is the caption preset it asks. Copy settings.toml to settings.user.toml to write your own tile prompts.
upscale_modeloptUPSCALE_MODELOptional upscale model, run over the entire image before any tiling.
negativeoptCONDITIONINGOptional negative conditioning. Unconnected it is an empty encode of this node's CLIP.

Outputs (1)

NameTypeDescription
IMAGEIMAGE