emberframe-emberframe-nodes/example_workflow_notes.md

Example Workflow Notes

Split-Sigma Sampling

Scheduler
  -> SplitSigmas

SplitSigmas high_sigmas
  -> Sigmas Resample
  -> ZIB EmberFrame SamplerCustomAdvanced Capture sigmas

ZIB output
  -> ZIT EmberFrame SamplerCustomAdvanced Capture latent_image

SplitSigmas low_sigmas
  -> ZIT EmberFrame SamplerCustomAdvanced Capture sigmas

PiD Final-Latent Decode

ZIT output or denoised_output
  -> EmberFrame Normalize ZImage/Flux Latent
  -> PiD Decode latent

Use:

PiD Decode backbone = zimage
PiD Decode pid_ckpt_type = 2k or 2kto4k
PiD Decode scale = 4
PiD Decode pid_steps = 4
PiD Decode cfg_scale = 1
PiD Decode sigma = 0

PiD Staged Intermediate Decode

ZIT pid_latent
  -> PiD Prepare latent

ZIT pid_sigma
  -> PiD Prepare sigma

PiD Prepare
  -> PiD Sample
  -> PiD Finalize

Start diagnostics with:

capture_mode = from_end
capture_step = 1
capture_tensor_source = x
sigma_source = next
capture_debug = true

For Z-Image / Flux PiD, keep PiD scale at 4.

Detailer File Notes

z-image_5.1_emberframe_pid.json uses the Impact detailer pass. Its workflow note lists the detector files and links for person_yolov8m-seg.pt, face_yolov8m.pt, Eyes.pt, and sam_vit_l_0b3195.pth.

z-image_5.2_emberframe_pid.json uses the SAM3 detailer pass. Its workflow note links the required models/sam3/sam3.pt file to the gated facebook/sam3 Hugging Face repo and notes that ComfyUI needs authenticated access if auto-download is used.