Framer AI AgentsDesign and publish professional sites with AI
Dali by Lulu
The prediction MCP that helps you save the AI generation tax
15 followers
The prediction MCP that helps you save the AI generation tax
15 followers
The prediction MCP that helps you avoid the AI generation tax. Most AI generation failures are prompt failures. You can't tell the difference until after you've burned the token. Dali scores your prompt before you generate — so you never waste a credit on a bad prompt again. Every wasted generation has a real cost (a Seedance retry is ~$6) — the live dashboard tracks what the community has saved by catching bad prompts before they burned a credit.
Curious how the scoring actually works under the hood. Is it running the prompt through a smaller model to predict output quality, or is it more of a heuristic check on the prompt structure itself?
@smet1510713 Honest answer: today it's structured heuristics, not a model guessing quality — versioned, per-generator rubrics that score the prompt's structure against what each model demonstrably responds to, plus antipattern detection per model. So it's deterministic and inspectable: same prompt, same rubric version, same score. The next layer (in progress) grounds the rubric weights in real ad-performance outcomes, so dimensions get weighted by what actually converted — not just what generates prettier. Rubric format is public in the repo if you want to go deeper.
The dashboard idea is genuinely useful, knowing what failed before spending credits saves real money on the pricey video models. Scored a test prompt at 62 and it did flag something I'd have missed on my own.
@slastntepeq6zm Thank you! That "flagged something I'd have missed" moment is exactly the job — knowing before you spend the credit, not after. Curious: which generator were you scoring for? Rubrics are per-model, so knowing where people feel the save the most helps me decide which one to deepen next.