What is the "Score"?
Last updated
When writing Selects prompts, you can reference a "score" to control how strictly Spingle filters your footage.
An AI confidence rating. As Spingle analyzes your footage, it evaluates each segment (quality, stability, lighting, composition, and content relevance) and assigns a confidence score, typically from 0.0 (lowest) to 1.0 (highest).
A higher score means Spingle is more confident the segment is technically sound and valuable.
A lower score means the clip has imperfections or is borderline, but could still be usable depending on your creative needs.
Tell Spingle how picky to be by including a score threshold in your Selects prompt.
For fewer, higher-quality selects (stricter): use a higher threshold.
Example: `"Keep only shots above a score of 0.8, prioritizing stable movement and clear focus."`
For more options to review (looser): use a lower threshold.
Example: `"Select usable clips above a score of 0.3, focusing on candid moments, even with minor shake."`
Score and the shaky threshold work together. The score (set in your prompt) controls overall quality strictness. The shaky threshold (a slider in Selects) controls how aggressively Spingle cuts camera motion. Combine both with clear descriptive criteria for the best control.
Start moderately: if unsure, try a mid-range score like 0.5 or 0.6 and adjust.
Combine with description: the score sets strictness, while your text guides what the AI prioritizes.
Iterate: re-prompting is fast thanks to the local cache, so experiment. Too few clips? Lower the score. Too many? Raise it or add specific criteria.
Context matters: the ideal threshold varies with your source footage and project.
Refine your description: clearly define the type of good clip you want, beyond the score alone.
Report it: if a genuinely good clip is consistently missed even with a low score and clear description, use the Report Feedback button so we can improve the AI.
Crafting Effective Prompts - full prompting guide with score examples
Why are AI Results Sometimes Inaccurate? - understanding AI limitations
Last updated