ComfyUI Node
TS Langevin Inpaint
Inpaint sampler that spends inner Langevin steps at every noise level so the repainted region agrees with the pixels around it. Works with any model family; feed it a latent carrying a noise mask.
TS Langevin Inpaint
- model
- positive
- negative
- latent_image
- LATENT
◄seed0►
◄steps20►
◄cfg1.0►
◄sampler_name▾►
◄scheduler▾►
◄denoise1.00►
◄think_steps5►
◄guidance16.0►
◄step_size0.20►
◄beta1.0►
◄friction15.0►
CategoryTS/Image/Retouch
Inputs (15)
| Name | Type | Default | Description |
|---|---|---|---|
| model | MODEL | — | |
| positive | CONDITIONING | — | |
| negative | CONDITIONING | — | |
| latent_image | LATENT | Latent with a noise mask (white = repaint). | |
| seed | INT | 00–18446744073709550000 | — |
| steps | INT | 201–200 | — |
| cfg | FLOAT | 1.00–100 | — |
| sampler_name | COMBO | 45 options: euler, euler_cfg_pp, euler_ancestral, euler_ancestral_cfg_pp, heun, heunpp2, +39 | |
| scheduler | COMBO | 11 options: simple, sgm_uniform, karras, exponential, ddim_uniform, beta, +5 | |
| denoise | FLOAT | 1.000–1 | — |
| think_steps | INT | 50–20 | Inner Langevin steps per noise level. 0 disables the refinement and leaves an ordinary masked sample. |
| guidance | FLOAT | 16.00.1–50 | How hard the masked region is pulled towards agreement with the known pixels. Higher binds tighter, too high can oscillate. |
| step_size | FLOAT | 0.200.001–1 | Langevin step size. Larger converges faster, less stably. |
| beta | FLOAT | 1.00.01–5 | Time-step ratio of the masked branch. Lower it to compensate a high guidance value. |
| friction | FLOAT | 15.00–50 | Damping of the inertial dynamics. Lower converges faster and is less stable. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| LATENT | LATENT | — |