ComfyUI Node
TS Shifted Sigmas
Sigmas for few-step turbo LoRAs of Qwen Image 2.1: the raw timesteps the LoRA was distilled on, bent by the same resolution-dependent shift the diffusers pipeline uses, plus the final zero. Connect to SamplerCustom or SamplerCustomAdvanced with cfg 1 and the euler sampler.
TS Shifted Sigmas
- latent
- sigmas
◄nodes1.0, 0.9375, 0.875, 0.75, 0.5, 0.25►
◄base_shift0.50►
◄max_shift0.90►
CategoryTS/Utils
Inputs (4)
| Name | Type | Default | Description |
|---|---|---|---|
| latent | LATENT | The latent that goes into the sampler. Only its size is read: a bigger picture gets a stronger shift. | |
| nodes | STRING | 1.0, 0.9375, 0.875, 0.75, 0.5, 0.25 | Raw timesteps from noise (1.0) down, comma separated; fractions like 1/6 and the model card's square brackets are fine. The default is viggle-turbo 6 steps (v0.2.1 and v0.3). Add or drop steps at the noisy end only and keep 0.875, 0.75, 0.5, 0.25. |
| base_shift | FLOAT | 0.500–5 | Shift at 256 image tokens (256x256 px). Qwen Image 2.1: 0.5. |
| max_shift | FLOAT | 0.900–5 | Shift at 8192 image tokens (about 1448x1448 px). Qwen Image 2.1: 0.9. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| sigmas | SIGMAS | One sigma per node, then 0. Steps = number of nodes. |