ComfyUI Node
Flow Matching Stage
A ComfyUI node in latent/upscaling with 21 inputs and 6 outputs.
Flow Matching Stage
- model
- positive
- negative
- latent
- latent
- presampler_latent
- next_seed
- model
- positive
- negative
◄seed0►
◄steps16►
◄cfg4.5►
◄sampler_name▾►
◄scheduler▾►
◄scale_factor1.00►
◄noise_ratio0.00►
◄skip_blend0.50►
◄denoise1.00►
◄upscale_methodbicubic►
◄enable_dilated_samplingdisable►
◄reduce_memory_useenable►
◄dilated_downscale2.00►
◄dilated_blend0.25►
◄dilated_min_steps1►
◄dilated_seed_modederive►
◄dilated_denoise0.50►
Categorylatent/upscaling
Inputs (21)
| Name | Type | Default | Description |
|---|---|---|---|
| model | MODEL | Flow-matching diffusion model to drive refinement. | |
| positive | CONDITIONING | Positive conditioning for CFG. | |
| negative | CONDITIONING | Negative conditioning for CFG. | |
| latent | LATENT | Latent to upscale and refine in this single stage. | |
| seed | INT | 00–18446744073709550000 | Seed controlling re-noising for this stage (and typically the sampler noise). |
| steps | INT | 161–256 | Denoising steps for this stage. |
| cfg | FLOAT | 4.50–20 | Classifier Free Guidance strength. |
| sampler_name | COMBO | Sampler backend leveraged during refinement. | |
| scheduler | COMBO | Noise schedule applied during denoising. | |
| scale_factor | FLOAT | 1.000.1–8 | Spatial scale factor applied to the latent grid for this stage. |
| noise_ratio | FLOAT | 0.000–1 | Flow-style re-noise amount (0 = keep latent, 1 = replace with pure noise). |
| skip_blend | FLOAT | 0.500–1 | Skip blend weight (0 = all denoised, 1 = all pre-sampler latent). |
| denoise | FLOAT | 1.000–1 | Denoising strength supplied to the sampler. |
| upscale_method | COMBO | bicubic | Resampling kernel for spatial upscaling. NOTE: ComfyUI's `lanczos` path uses PIL and is unsafe for LATENT tensors; this node will fall back to `bicubic` if selected. |
| enable_dilated_samplingopt | COMBO | disable | Optionally run a dilated refinement pass for global coherence (experimental). |
| reduce_memory_useopt | COMBO | enable | Enable to reduce VRAM use by avoiding extra tensor clones where possible. |
| dilated_downscaleopt | FLOAT | 2.001–4 | Factor used when downscaling for the dilated pass (>=1.0). |
| dilated_blendopt | FLOAT | 0.250–1 | Blend weight of the dilated refinement result (frequency-domain blend). |
| dilated_min_stepsopt | INT | 11–256 | Minimum sampling steps for dilated refinement. Actual steps = max(min_steps, steps // 2). Increase for lightning models to ensure adequate sampling (e.g., 4). |
| dilated_seed_modeopt | COMBO | derive | Seed mode for dilated sampling. 'derive' adds 10,000 to base seed, 'same' uses identical seed. |
| dilated_denoiseopt | FLOAT | 0.500–1 | Denoising strength for dilated sampling. Lower values preserve more spatial structure from the original. |
Outputs (6)
| Name | Type | Description |
|---|---|---|
| latent | LATENT | — |
| presampler_latent | LATENT | — |
| next_seed | INT | — |
| model | MODEL | — |
| positive | CONDITIONING | — |
| negative | CONDITIONING | — |