ComfyUI Node
LTX2 Efficient Video Sampler
A ComfyUI node in video/ltx2 with 26 inputs and 1 output.
LTX2 Efficient Video Sampler
- model
- latent_video
- positive
- negative
- sigmas
- LATENT
◄seed0►
◄steps20►
◄cfg4.00►
◄sampler_name▾►
◄denoise1.00►
◄optimization_presetThermal Auto-Scale►
◄target_temp70►
◄frame_stride1►
◄attention_window4►
◄freeze_ratio0.30►
◄thermal_throttletrue►
◄interpolation_methodslerp►
◄optimization_engineAdaptive Cache (2-4x speedup)►
◄engine_cache_ratio0.5►
◄engine_merge_ratio0.25►
◄context_slice_methodKeep Last 4096 (Slot 2 - LTX Connector)►
◄max_shift2.05►
◄base_shift0.95►
◄terminal0.10►
◄stretchtrue►
◄frame_rate25.00►
Categoryvideo/ltx2
Inputs (26)
| Name | Type | Default | Description |
|---|---|---|---|
| model | MODEL | — | |
| latent_video | LATENT | — | |
| positive | CONDITIONING | — | |
| negative | CONDITIONING | — | |
| seed | INT | 00–18446744073709550000 | — |
| steps | INT | 201–10000 | — |
| cfg | FLOAT | 4.000–100 | — |
| sampler_name | COMBO | 44 options: euler, euler_cfg_pp, euler_ancestral, euler_ancestral_cfg_pp, heun, heunpp2, +38 | |
| denoise | FLOAT | 1.000–1 | — |
| optimization_preset | COMBO | Thermal Auto-Scale | 7 options: Quality - Fast (RTX 3080+), Quality - Balanced (RTX 3060/3070), Quality - RTX 2060 6GB, Quality - Cool (RTX 2060 6GB), Quality - Ultra Cool (Low Power GPUs), Thermal Auto-Scale, +1 |
| target_temp | INT | 7050–85 | — |
| frame_stride | INT | 11–32 | Use 1 for best quality |
| attention_window | INT | 41–32 | — |
| freeze_ratio | FLOAT | 0.300–1 | — |
| thermal_throttle | BOOLEAN | true | — |
| interpolation_method | COMBO | slerp | 4 options: linear, slerp, motion, none |
| optimization_engine | COMBO | Adaptive Cache (2-4x speedup) | Temporal attention optimization engine |
| engine_cache_ratio | FLOAT | 0.50–1 | Fraction of later steps to cache (AdaCache) |
| engine_merge_ratio | FLOAT | 0.250–0.5 | Fraction of tokens to merge (TokenMerge) |
| context_slice_method | COMBO | Keep Last 4096 (Slot 2 - LTX Connector) | For Gemma+LTX Connector: use Last 4096. For T5+Gemma: use First 4096. |
| sigmasopt | SIGMAS | Connect LTXVScheduler output here to override internal scheduler | |
| max_shiftopt | FLOAT | 2.050–100 | FlowMatch max shift (same as LTXVScheduler) |
| base_shiftopt | FLOAT | 0.950–100 | FlowMatch base shift (same as LTXVScheduler) |
| terminalopt | FLOAT | 0.100–0.99 | Sigma terminal value after stretching |
| stretchopt | BOOLEAN | true | Stretch sigmas to [terminal, 1] range |
| frame_rateopt | FLOAT | 25.000–1000 | Frame rate for conditioning (same as LTXVConditioning) |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| LATENT | LATENT | — |