ComfyUI Node
AsymFLUX2 Apply Adapter
Turns a stock FLUX.2-klein-base-9B MODEL into AsymFLUX.2. Drop the adapter (Lakonik/AsymFLUX.2-klein-9B diffusion_pytorch_model.safetensors, ~707 MB) into ComfyUI/models/loras/. Chain: Load Diffusion Model -> this node -> KSampler. Use AsymFLUX2 Empty Pixel Latent for the latent input, AsymFLUX2 Oklab Decode after the sampler.
AsymFLUX2 Apply Adapter
- model
- model
◄adapter▾►
◄shift17.0►
◄adapter_strength1.00►
◄orthogonal_guidance1.00►
◄clamp_denoisedtrue►
CategoryAsymFLUX2/loaders
Inputs (6)
| Name | Type | Default | Description |
|---|---|---|---|
| model | MODEL | FLUX.2-klein-base-9B from a stock Load Diffusion Model node. | |
| adapter | COMBO | AsymFLUX.2-klein adapter safetensors (in models/loras/). | |
| shift | FLOAT | 17.00.1–100 | Flow shift (paper convention; converted to comfy mu = log(shift) internally). |
| adapter_strength | FLOAT | 1.00-2–2 | LoRA strength applied to the rank-256 LoRA. 1.0 = full strength. |
| orthogonal_guidance | FLOAT | 1.000–2 | AsymFlow orthogonal CFG bias strength. 1.0 = upstream demo default. 0.0 = standard CFG. |
| clamp_denoised | BOOLEAN | true | Per-step Oklab gamut clamp on the x0 estimate. Prevents color drift across the sampling loop. Disable if using uni_pc — see README. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| model | MODEL | — |