kubakub canvas plan

Work out the size a model should generate at, so the result scales back to your target without stretching. The work canvas is target / k for one uniform factor k, padded to the model's size grid, never stretched. Feed width/height to the empty latent and the plan to Canvas To Work / Canvas Restore.

By kubakubkub·Created 15 days ago·Updated 3 days ago· 9

kubakub canvas plan

size_from

canvas_plan
width
height
frames
fps
plan_json
report
k
target_width
target_height

◄target_width3840►

◄target_height2160►

◄model_familyflux2►

◄k2►

◄budget_mp0.00►

◄integer_kfalse►

◄target_frames0►

◄pad_aligncenter►

◄max_pad_percent2.0►

◄divisor_override0►

Categorykubakub/2d/regions/canvas

Inputs (11)

Name	Type	Default	Description
target_width	INT	384016–32768	Width of the final image in pixels (the matrix = the full projection image at delivery size).
target_height	INT	216016–32768	Height of the final image in pixels.
model_family	COMBO	flux2	flux2: divisor 16, ~1 MP. ltxav: divisor 32, 8k+1 frames. minimax_h3: divisor 32, 17k+5 frames, 24 fps. sd_8: divisor 8.
k	STRING	2	Downscale factor target -> work. A number (2, 2.5), a fraction (15/4) or 'auto' to pick from the MP budget. It must divide the target into whole pixels.
budget_mp	FLOAT	0.000–16	auto k: target megapixels on the model grid. 0 = the model's trained budget.
integer_k	BOOLEAN	false	auto k: only whole factors (needed for latent x2 paths).
target_frames	INT	00–10000	Video models: frames wanted. Rounded up to the model's frame rule; Canvas Restore trims the extra frames. 0 = still.
pad_align	COMBO	center	Where the grid padding goes: split on both sides, or all at the end (right/bottom) or start (left/top).
max_pad_percent	FLOAT	2.00–50	auto k: skip factors needing more padding (extra area in percent).
divisor_override	INT	00–256	Force another size grid (e.g. 64). 0 = the model's own.
size_fromopt	IMAGE		Take the target size from this image instead of the widgets.

Outputs (10)

Name	Type	Description
canvas_plan	KUBA_CANVAS_PLAN	The plan; connect to canvas to work and canvas restore.
width	INT	Model grid width (padded). Use for the empty latent.
height	INT	Model grid height (padded).
frames	INT	Frames to generate (1 for stills).
fps	FLOAT	Frame rate the model generates at.
plan_json	STRING	The plan as text, for saving or checking.
report	STRING	Readable summary: work size, padding, k, frames.
k	FLOAT	The downscale factor that was used (target = work x k), e.g. 2.0 or 3.75. Useful after 'auto'.
target_width	INT	Width of the final image in pixels (from the widget or size_from).
target_height	INT	Height of the final image in pixels.