ComfyUI-FlowMatch-Advanced

Custom nodes to make ComfyUI sampling closer to ai-toolkit flow-matching behavior for Flux, Qwen-Image, and Z-Image.

Nodes

• FlowMatch Sigmas (ai-toolkit style)
  Outputs SIGMAS for SamplerCustomAdvanced using ai-toolkit-style formulas and defaults.
• DiT FlowMatch Model (ai-toolkit style)
  Patches model_sampling with a real Comfy sampling object (Comfy-native patching), not a callback.

Model Presets

• flux: dynamic shift (base_shift=0.5, max_shift=1.15, max_seq_len=4096)
• qwen: dynamic shift (base_shift=0.5, max_shift=0.9, max_seq_len=8192)
• z-image: static shift (shift=3.0)

Recommended Workflow (Closest to ai-toolkit)

1. Load your model and LoRA.
2. Use DiT FlowMatch Model (ai-toolkit style) on the model (set model_type=auto unless needed).
3. Use FlowMatch Sigmas (ai-toolkit style) to generate SIGMAS.
4. Run SamplerCustomAdvanced:
   • sampler: euler (or res_multistep if that is your trained setup)
   • sigmas: from FlowMatch Sigmas (ai-toolkit style)
   • guider: CFGGuider (typically cfg=1.0 for flow models)
5. Decode with VAE.

Notes

• This is designed for the SamplerCustomAdvanced path.
  KSampler cannot directly inject this exact custom sigma schedule.
• force_aitk_timesteps=true uses 1.0 -> 1.0/steps timesteps before shift math, matching ai-toolkit behavior more closely.
• Default steps=25 is set for closer ai-toolkit-style sample quality (you can still drop to 20 for faster Flux previews).