Ouroboros Critic Settings 🐍

Critic (vision judge) settings bundle for the Ouroboros loop — model + scoring rules.

By Kinburg·Created 2 months ago·Updated 3 days ago· 1

config

critic_settings

◄criteriaoverall_quality: style matches the prompt, no artifacts, no excess noise, correct proportions, good color reproduction anatomy: all required limbs present, no extra limbs, correct placement, natural pose, proportional body prompt_compliance: how accurately the image follows the generation prompt camera: camera angle and camera settings match the intent text: if the prompt requests text — present, character-accurate, correct color/font/size/placement (if the prompt has NO text, give the top score)►

◄score_min1►

◄score_max5►

◄rubricStrictly and critically evaluate the generated image.►

◄system_promptYou are a strict, consistent image-quality judge inside an automatic refinement loop. You are shown ONE image plus evaluation criteria and the prompt it was generated from. Score it objectively, then give concrete, actionable advice on how to change the GENERATION PROMPT (not the image directly) to raise the weakest criteria, and list negative-prompt terms for the flaws you see. Reply with ONLY a single JSON object — no prose, no markdown, no fences.►

◄advice_styleone or two concrete, actionable sentences►

◄samples1►

◄image_downscale1.0►

CategoryKinburg-Nodes/Bestiary/Ouroboros

Inputs (9)

Name	Type	Default	Description
config	KINBURG_LLM_CONFIG		A 'Local LLM Settings (GGUF)' WITH a 'Vision Settings (GGUF)' (mmproj) — the critic needs vision.
criteria	STRING	overall_quality: style matches the prompt, no artifacts, no excess noise, correct proportions, good color reproduction anatomy: all required limbs present, no extra limbs, correct placement, natural pose, proportional body prompt_compliance: how accurately the image follows the generation prompt camera: camera angle and camera settings match the intent text: if the prompt requests text — present, character-accurate, correct color/font/size/placement (if the prompt has NO text, give the top score)	Criteria the critic scores each image on (one per line, 'name: description'). Pre-filled; clear for a single overall score.
score_min	INT	10–100	—
score_max	INT	51–100	—
rubricopt	STRING	Strictly and critically evaluate the generated image.	Free-text guidance / tone for the critic.
system_promptopt	STRING	You are a strict, consistent image-quality judge inside an automatic refinement loop. You are shown ONE image plus evaluation criteria and the prompt it was generated from. Score it objectively, then give concrete, actionable advice on how to change the GENERATION PROMPT (not the image directly) to raise the weakest criteria, and list negative-prompt terms for the flaws you see. Reply with ONLY a single JSON object — no prose, no markdown, no fences.	The critic's persona (system prompt). Default pre-filled; blank → built-in default.
advice_styleopt	STRING	one or two concrete, actionable sentences	How the advice should read, e.g. 'a detailed paragraph' or 'one short sentence'.
samplesopt	INT	11–9	Self-consistency: judge each image N times and take the median score (N× critic cost).
image_downscaleopt	FLOAT	1.01–8	Shrink the image the critic sees by this factor before judging (2 = half-size, 4 = quarter). Fewer image tokens → faster judging and less VRAM, at the cost of fine detail — good when pixel-level precision isn't needed. 1.0 = full resolution (never upscales past the Vision Settings 'image_max_side' cap).

Outputs (1)

Name	Type	Description
critic_settings	KINBURG_CRITIC	—