1.5.0

FSampler Changelog:
2026-01-01
v1.5.0
Sigma-Aware Extrapolation
New sigma_aware toggle (FSampler Advanced / FSampler Select) — uses actual sigma coordinates for Lagrange interpolation
Improves skip prediction accuracy with non-uniform schedulers (bong_tangent, karras, exponential, beta, etc.)
Enabled by default in the simple FSampler node
Denoised Extrapolation
New extrapolate_denoised toggle — extrapolates the model's denoised (clean image) output instead of epsilon
Denoised converges smoothly toward the final image, can produce more stable skip predictions
Enabled by default in the simple FSampler node
Composes with sigma_aware (orthogonal: sigma_aware controls weights, denoised controls what quantity is extrapolated)
New Node: FSampler Select
Outputs SAMPLER + SIGMAS for use with ComfyUI's SamplerCustom (KSamplerCustom) node
Lets you plug FSampler's skip-aware sampling into the standard modular workflow alongside any guider (CFGGuider, DualCFGGuider, BasicGuider, etc.)
Wire sampler → SamplerCustom's sampler slot, sigmas → SamplerCustom's sigmas slot
All FSampler skip/adaptive controls available; model/conditioning/cfg/noise handled by SamplerCustom
Live Preview
FSampler and FSampler Advanced now show live image previews during sampling (Latent2RGB, auto-enabled)