ComfyUI Node
Anima LLLite Tiled ControlNet Sampler
All-in-one Anima ControlNet-LLLite tiled sampler. Splits the image into a rows x columns grid of overlapping tiles, then for every tile applies Anima LLLite (tile as control), VAE-encodes, KSamples and VAE-decodes, and finally feathers the tiles back together — replacing a whole manual tile/sample/untile graph with one node. No extra node packs required.
Anima LLLite Tiled ControlNet Sampler
- image
- model
- positive
- negative
- vae
- image
◄sampling_mode▾►
◄seed0►
◄steps20►
◄cfg7.0►
◄sampler_name▾►
◄scheduler▾►
◄denoise0.50►
◄lllite_name▾►
◄strength1.00►
◄start_percent0.000►
◄end_percent1.000►
◄preserve_wrappertrue►
◄rows2►
◄columns2►
◄overlap0.00►
◄overlap_x64►
◄overlap_y64►
◄method▾►
◄color_match▾►
◄color_match_strength1.00►
◄vae_decode_tiledfalse►
◄vae_decode_tile_size512►
CategoryvsLinx/sampling
Inputs (27)
| Name | Type | Default | Description |
|---|---|---|---|
| image | IMAGE | — | |
| model | MODEL | — | |
| positive | CONDITIONING | — | |
| negative | CONDITIONING | — | |
| vae | VAE | — | |
| sampling_mode | COMBO | per_tile: sample each tile fully then stitch (lowest VRAM; can show seams/ghosting). multidiffusion: one sampling pass over the whole image, averaging overlapping tiles in latent space every step — tiles stay in sync so seams and double-exposure are eliminated (slightly more VRAM). In multidiffusion mode the 'method' and 'color_match' fields are not used. | |
| seed | INT | 00–18446744073709550000 | — |
| steps | INT | 201–10000 | — |
| cfg | FLOAT | 7.00–100 | — |
| sampler_name | COMBO | 44 options: euler, euler_cfg_pp, euler_ancestral, euler_ancestral_cfg_pp, heun, heunpp2, +38 | |
| scheduler | COMBO | 9 options: simple, sgm_uniform, karras, exponential, ddim_uniform, beta, +3 | |
| denoise | FLOAT | 0.500–1 | — |
| lllite_name | COMBO | Anima ControlNet-LLLite weights file (from the controlnet folder). | |
| strength | FLOAT | 1.00-10–10 | — |
| start_percent | FLOAT | 0.0000–1 | — |
| end_percent | FLOAT | 1.0000–1 | — |
| preserve_wrapper | BOOLEAN | true | Delegate to any model_function_wrapper already installed upstream instead of overwriting it, so multiple wrapper nodes can stack. Same toggle as the AnimaLLLiteApply node. |
| rows | INT | 21–256 | — |
| columns | INT | 21–256 | — |
| overlap | FLOAT | 0.000–0.5 | Overlap between tiles as a fraction of tile size, added on top of overlap_x/overlap_y. |
| overlap_x | INT | 640–8192 | Extra horizontal overlap in pixels. |
| overlap_y | INT | 640–8192 | Extra vertical overlap in pixels. |
| method | COMBO | Resampling used to keep every decoded tile at a uniform size before stitching. | |
| color_match | COMBO | Per-tile color matching against the source tile, to fix tonal seams (brightness/colour steps between tiles). 'mean_std' re-scales each tile's per-channel mean/std (fast, simple); 'wavelet' keeps the tile's detail but takes the source tile's broad tone (better on textured tiles). | |
| color_match_strength | FLOAT | 1.000–1 | How strongly to apply the color match (0 = off, 1 = full). |
| vae_decode_tiled | BOOLEAN | false | (multidiffusion only) Decode the final full-image latent in tiles instead of one pass, to avoid a single huge VAE decode that can spike VRAM (or spill into slow shared system memory). No effect in per_tile mode, where each tile is already decoded on its own. |
| vae_decode_tile_size | INT | 51264–4096 | (multidiffusion only) Tile size in pixels for the tiled VAE decode. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| image | IMAGE | — |