How to Use
Rate my models and LoRAs with the S–F tier system after you have tested it on your preferred workflow.
(your base model, sampler, scheduler, strength, steps, denoise, and prompt)
Use these 5 Bullet Points: Consistency, Orientation, Prompt Accuracy, Ease-of-Use (0–100 steps), and Drift Control.
Comment on a post of mine starting with this format for ideal rating examination:
Rating: S – F
Reason: “Briefly explain the rating such as what stayed consistent, what broke in certain orientations/prompts, how fast it locked in within 0–100 steps, and how much drift you saw.”
Note: As long as your post starts with Rating + Reason, you can add any extra details below (I may read them depending on time). Please keep everything community-friendly and within CivitAI rules/guidelines. Thank you!
S Tier — “Perfect”
Consistency: Extremely consistent; stays strongly on-model/on-style across many prompts.
Orientation: Reliable in Square/Portrait/Landscape with minimal composition breakage.
Prompt Accuracy: Highly predictable; small prompt changes give expected results.
Ease of Use (0–100 steps): Locks in fast; high success rate at typical step counts.
Drift Control: Very low drift; holds identity/style even in harder scenes.
A Tier — “Excellent”
Consistency: Very consistent; only rare misses under extreme prompts.
Orientation: Works well in most orientations; minor issues in specific cases.
Prompt Accuracy: Mostly predictable; a few tokens may need specific wording.
Ease of Use (0–100 steps): Strong success rate at normal steps; occasional reroll needed.
Drift Control: Low drift; noticeable mainly when pushing extremes (angles, lighting, chaos).
B Tier — “Good”
Consistency: Consistent in most prompts; occasional instability in harder scenarios.
Orientation: Generally solid; one format may be slightly weaker (landscape/full-body).
Prompt Accuracy: Mostly accurate; may need careful phrasing or a few negatives.
Ease of Use (0–100 steps): Usually locks in at normal steps; sometimes needs extra steps/rerolls.
Drift Control: Moderate-low drift; holds identity/style with minor softening under changes.
C Tier — “Average”
Consistency: Decent in “safe prompts,” inconsistent across a wide variety.
Orientation: Works in most formats but shows clear weak spots in at least one.
Prompt Accuracy: Prompt-sensitive; better results with careful wording/negatives.
Ease of Use (0–100 steps): Needs more tries or higher steps to get consistent hits.
Drift Control: Moderate drift; identity/style can slip with outfit/scene/lighting changes.
D Tier — “Weak”
Consistency: Inconsistent; can hit, but doesn’t stay stable across batches.
Orientation: Often prefers a specific format (square/portrait); others degrade.
Prompt Accuracy: Less predictable; prompts may produce mixed or off-target results.
Ease of Use (0–100 steps): Frequent rerolls and/or high steps needed for acceptable outputs.
Drift Control: High drift; easier to slide off-character/off-style.
E Tier — “Poor”
Consistency: Frequently off-model/off-style; only occasional correct results.
Orientation: Orientation changes commonly break composition or identity/style.
Prompt Accuracy: Prompts don’t reliably map to output; results feel inconsistent/random.
Ease of Use (0–100 steps): Hard to lock in even at higher steps; very low success rate.
Drift Control: Very high drift; identity/style shifts quickly with prompt changes.
F Tier — “Broken”
Consistency: Rarely produces correct identity/style; mostly unusable.
Orientation: Fails across orientations; unstable framing and features.
Prompt Accuracy: Prompts do not meaningfully control outputs.
Ease of Use (0–100 steps): Does not lock in even at max steps; failure symptoms present.
Drift Control: Extreme drift; outputs collapse into generic or corrupted results.
Note: I hope this rating system helps me improve my Models & LoRAs over time, and also helps the community give clear, consistent ratings for Models, LoRAs, and workflows.