ComfyUI Node
Krea2 TaylorSeer Lite
TaylorSeer Lite for Krea2. It forecasts one final model output between periodic full refreshes, skipping all transformer blocks with only a tiny cache. Euler/simple is recommended; increase fresh_interval for more speed at a larger quality cost. Order 0 is the tested 10-step Krea2 Turbo default.
Krea2 TaylorSeer Lite
- model
- model
◄warmup_steps3►
◄fresh_interval2►
◄tail_full_steps2►
◄max_order0►
CategoryKrea2 Training Free Improvements/acceleration
Inputs (5)
| Name | Type | Default | Description |
|---|---|---|---|
| model | MODEL | — | |
| warmup_steps | INT | 32–100 | — |
| fresh_interval | INT | 22–100 | — |
| tail_full_steps | INT | 20–100 | — |
| max_order | INT | 00–2 | — |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| model | MODEL | — |