FlashVSR+ Tiled Options

A ComfyUI node in FlashVSR+ Tiled with 23 inputs and 1 output.

By hdmotionpersonal·Created a day ago·Updated a day ago· 1

options

◄spatial_tilingfalse►

◄spatial_tile_height512►

◄spatial_tile_width512►

◄spatial_overlap128►

◄spatial_blend▾►

◄spatial_gaussian_sigma0.25►

◄temporal_compression▾►

◄temporal_merge▾►

◄temporal_gaussian_sigma1.00►

◄decode_every_frames112►

◄cache_offload_mode▾►

◄offload_original_cachefalse►

◄tiny_decoder_mode▾►

◄tiny_decoder_channels_lasttrue►

◄color_correctionfalse►

◄color_correction_levels5►

◄color_correction_chunk_size16►

◄ffn_chunk_tokens8192►

◄vae_tiledtrue►

◄vae_tile_height240►

◄vae_tile_width416►

◄vae_overlap120►

◄video_preset▾►

CategoryFlashVSR+ Tiled

Inputs (23)

Name	Type	Default	Description
spatial_tiling	BOOLEAN	false	Run Frames or Tiny-decoder VIDEO inference on overlapping spatial tiles. VIDEO stores temporary float32 segments on disk.
spatial_tile_height	INT	512256–8192	Tile height in output pixels.
spatial_tile_width	INT	5120–8192	Tile width in output pixels. Set independently from height; 0 follows tile height for older workflows.
spatial_overlap	INT	1280–4096	Overlap in output pixels; step 8, less than half of both tile dimensions. Tile dimensions stay multiples of 128.
spatial_blend	COMBO		Choose how overlapping spatial tiles are feathered together.
spatial_gaussian_sigma	FLOAT	0.250.05–1	Set the Gaussian falloff width for gaussian spatial blending. Lower values concentrate weight near tile centers.
temporal_compression	COMBO		Set effective temporal compression. Lower values process more repeated temporal samples and cost more; final frame count stays unchanged.
temporal_merge	COMBO		For temporal compression 1 or 2, choose Gaussian averaging or the first decoded sample in each repeated group.
temporal_gaussian_sigma	FLOAT	1.000.01–10	Set the Gaussian averaging width for temporally merged samples.
decode_every_frames	INT	1120–16384	Output video-frame interval for Tiny incremental decoding. Must be divisible by 4 × temporal_compression. 0=auto/full clip; spatial VIDEO tiling requires a positive interval.
cache_offload_mode	COMBO		Choose when to offload the DiT KV cache from GPU memory. Auto estimates memory and selects a best-effort policy.
offload_original_cache	BOOLEAN	false	Also offload the original input cache. This can save VRAM at the cost of extra transfers.
tiny_decoder_mode	COMBO		Choose Tiny decoder execution mode. Auto selects a mode; serial uses less memory, parallel may be faster.
tiny_decoder_channels_last	BOOLEAN	true	Use channels-last memory layout in the Tiny decoder where supported.
color_correction	BOOLEAN	false	Match output colors to the low-quality input reference. In spatial video tiling this is applied per tile.
color_correction_levels	INT	51–8	Set the number of color-correction pyramid levels.
color_correction_chunk_size	INT	161–128	Set the number of frames processed together during color correction.
ffn_chunk_tokens	INT	81920–262144	Limit feed-forward tokens per chunk to reduce peak VRAM. Zero disables chunking.
vae_tiled	BOOLEAN	true	Decode the Wan VAE in spatial tiles to reduce peak VRAM.
vae_tile_height	INT	24032–2048	VAE decoder tile height in output pixels (Wan VAE uses 8 latent cells per pixel block).
vae_tile_width	INT	41632–2048	VAE decoder tile width in output pixels (Wan VAE uses 8 latent cells per pixel block).
vae_overlap	INT	1200–1024	Set overlap between Wan VAE decoder tiles in output pixels.
video_preset	COMBO		Choose the H.264 encoding speed and compression trade-off for the saved video.

Outputs (1)

Name	Type	Description
options	FLASHVSR_PLUS_OPTIONS	—