Nodes/ComfyUI-ContextAnchoredTileRefine/Tile Test: Settings
ComfyUI Node

Tile Test: Settings

A ComfyUI node in image/upscaling/tile testing with 1 input and 13 outputs.

By Blakeem·Created 2 months ago·Updated 3 days ago· 25
Tile Test: Settings
    • global_style_instruction
    • global_style_max_tokens
    • tile_caption_instruction
    • tile_caption_max_tokens
    • tile_tags_instruction
    • prompt_tags_instruction
    • tile_tags_verification_statement
    • caption_megapixels
    • canvas_tokens
    • crop_tokens
    • tile_tags_verification_threshold
    • tile_tags_position_threshold
    • prompt_tags_verification_threshold
    ◄presettags►
    Categoryimage/upscaling/tile testing

    Inputs (1)

    NameTypeDefaultDescription
    presetCOMBOtagsWhich settings file preset the outputs are read from, of the caption or the tags kind. The list is built when ComfyUI starts, so a new or renamed preset needs a restart. An edited wording reaches the outputs on the next queued run.

    Outputs (13)

    NameTypeDescription
    global_style_instructionSTRINGThe question the VL model is asked about the entire image for the style caption, with {PROMPT} left in.
    global_style_max_tokensINTThe generation budget for the style caption.
    tile_caption_instructionSTRINGThe question the VL model is asked about each tile by the caption kind, with {PROMPT} left in. A tags preset outputs an empty text.
    tile_caption_max_tokensINTThe generation budget for each tile caption. A tags preset outputs 0.
    tile_tags_instructionSTRINGThe question that asks the VL model to list the things in each tile as tags. A caption preset outputs an empty text.
    prompt_tags_instructionSTRINGThe question that asks the VL model to list the things the prompt names, with {PROMPT} left in. A caption preset outputs an empty text.
    tile_tags_verification_statementSTRINGThe statement each candidate tag is scored against on its tile, with {TAG} left in. A caption preset outputs an empty text.
    caption_megapixelsFLOATThe picture size in megapixels the VL model reads for each caption, from the [vision] table.
    canvas_tokensINTThe vision rows a tile takes from one encode of the entire image, from the [vision] table.
    crop_tokensINTThe vision rows a tile takes from the encode of its own crop, from the [vision] table.
    tile_tags_verification_thresholdFLOATThe score a tag the tile's own list names must reach on the entire tile to be kept. A caption preset outputs 0.
    tile_tags_position_thresholdFLOATThe score a tag must reach on one of the six strips of a tile for that strip to hold it. A caption preset outputs 0.
    prompt_tags_verification_thresholdFLOATThe score a thing from the prompt that the tile's own list lacks must reach on the entire tile to be kept. A caption preset outputs 0.