ComfyUI Node
Latent Hybrid Inverter (Qwen)
A ComfyUI node in Qwen/Sampling with 12 inputs and 1 output.
Latent Hybrid Inverter (Qwen)
- model
- latent_image
- positive
- LATENT
◄steps20►
◄strength0.50►
◄blend_factor0.50►
◄inverter_seed0►
◄forward_diffusion_seed1►
◄velocity_amplification1.00►
◄velocity_perturb_strength0.20►
◄internal_precisionbfloat16►
◄normalize_outputenable►
CategoryQwen/Sampling
Inputs (12)
| Name | Type | Default | Description |
|---|---|---|---|
| model | MODEL | UNet model used for both inversion and diffusion passes. | |
| latent_image | LATENT | Latent tensor to invert and re-noise. | |
| positive | CONDITIONING | Positive prompt conditioning that guides both passes. | |
| steps | INT | 201–1000 | Solver steps executed for inversion and forward diffusion. |
| strength | FLOAT | 0.500–1 | Strength for both inversion and forward diffusion. The target noise level. |
| blend_factor | FLOAT | 0.500–1 | 0.0 = 100% Forward Diffusion (Stable), 1.0 = 100% Inverter (Creative) |
| inverter_seed | INT | 00–18446744073709550000 | Seed for the creative, perturbed inversion process. |
| forward_diffusion_seed | INT | 10–18446744073709550000 | Seed for the stable, 'perfect' noise anchor. |
| velocity_amplification | FLOAT | 1.00-0.9–2 | Deterministically scales the predicted velocity each step. |
| velocity_perturb_strength | FLOAT | 0.200–5 | Random velocity noise level for the inverter stage. |
| internal_precision | COMBO | bfloat16 | Precision used for inverter UNet autocast. |
| normalize_output | COMBO | enable | Normalizing the inverter output is highly recommended. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| LATENT | LATENT | — |