π Artifact Risk Predictor - Analyze Quality Before Decoding
Analyzes latent space for signs of common generation artifacts:
β’ Overbaked faces (too many steps/high CFG)
β’ Melted hands (high-frequency noise patterns)
β’ Flat contrast (collapsed dynamic range)
β’ Noisy/grainy results (sampling artifacts)
β’ Oversaturated regions (CFG overshoot)
β‘ What This Node Does:
Provides INFORMATIONAL risk assessment - does not stop or control workflow.
Use the risk_score output with switches, routers, or conditional nodes
to build self-correcting workflows that can retry or adjust parameters.
π― Risk Score Output:
β’ 0.0-0.2: β Low risk - probably good
β’ 0.2-0.4: β οΈ Mild risk - watch for issues
β’ 0.4-0.6: π‘ Moderate risk - likely problems
β’ 0.6-0.8: π High risk - expect artifacts
β’ 0.8-1.0: π΄ Critical risk - probably unusable
π‘ Example Use Cases:
β’ Pass risk_score to a switch node to conditionally decode
β’ Log warnings for quality monitoring in batch generation
β’ Use with retry nodes to regenerate high-risk latents
β’ Build adaptive workflows that adjust CFG based on risk
π§ Detection Methods:
β’ Latent statistics (mean, std, range)
β’ Frequency analysis (detect noise patterns)
β’ Dynamic range checks (contrast collapse)
β’ Parameter correlation (overbaking detection)
β’ Channel balance analysis
βοΈ Risk Thresholds:
β’ Conservative: Flag more potential issues (fewer false negatives)
β’ Balanced: Standard detection (good default)
β’ Permissive: Only flag obvious problems (fewer false positives)
βΉοΈ Note: This node analyzes and reports - it does not prevent decoding.
Combine with other nodes to create conditional/adaptive workflows.
By ArctenoxΒ·Created 8 months agoΒ·Updated about a month agoΒ· 1
