ComfyUI Node
360° KSampler
A ComfyUI node in DiT360/sampling with 15 inputs and 1 output.
360° KSampler
- model
- positive
- negative
- latent_image
- LATENT
◄seed0►
◄steps20►
◄cfg3.5►
◄sampler_name▾►
◄scheduler▾►
◄denoise1.00►
◄circular_padding16►
◄enable_yaw_lossfalse►
◄yaw_loss_weight0.10►
◄enable_cube_lossfalse►
◄cube_loss_weight0.10►
CategoryDiT360/sampling
Inputs (15)
| Name | Type | Default | Description |
|---|---|---|---|
| model | MODEL | Connect your diffusion MODEL (e.g., FLUX UNet + DiT360 LoRA). | |
| seed | INT | 00–18446744073709550000 | Noise seed. Use a fixed value for reproducible results; use randomize/increment to explore variations. |
| steps | INT | 201–10000 | Diffusion steps. Recommended: ~20. Increase (30-50) for more detail; decrease for speed. |
| cfg | FLOAT | 3.50–100 | Classifier-free guidance. Recommended: ~3-5 for FLUX. Too high can cause harsh contrast or blown highlights. |
| sampler_name | COMBO | Sampler algorithm. Recommended: euler. Change only if you know what you prefer. | |
| scheduler | COMBO | Step scheduler. Recommended: simple (FLUX default). | |
| positive | CONDITIONING | Positive prompt conditioning. | |
| negative | CONDITIONING | Negative prompt conditioning. | |
| latent_image | LATENT | Starting latent. Use 360° Empty Latent for text2img, or a latent from img2img. | |
| denoise | FLOAT | 1.000–1 | Denoise strength. Recommended: 1.0 for text2img; <1.0 for img2img to preserve structure. |
| circular_padding | INT | 160–128 | Latent-space circular padding on X for seam reduction. Recommended: start at 0; if seam, try 8-16 (up to 32). Higher values cost VRAM/time. |
| enable_yaw_loss | BOOLEAN | false | Yaw-rotation consistency loss during sampling. Use if the panorama changes noticeably when yaw-rotated; ~2x slower. |
| yaw_loss_weight | FLOAT | 0.100–1 | Yaw loss strength. Recommended: 0.05-0.2. Higher values can over-constrain and reduce detail. |
| enable_cube_loss | BOOLEAN | false | Pole distortion reduction via cubemap reprojection loss. Use if poles look smeared/warped; ~1.5x slower. |
| cube_loss_weight | FLOAT | 0.100–1 | Cube loss strength. Recommended: 0.05-0.2. Increase slowly if poles still look distorted. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| LATENT | LATENT | — |