Scripts that are accurate, grim where the truth is grim, and actually watched
The biology videos that get watched do not soften the material: the parasite, the tumour, the fungus that eats the brain. Accuracy and grimness are the format, and a model that flinches is no use writing it.
Why the mainstream APIs fail here
Hosted models sand the interesting parts off: the parasite becomes a fascinating organism, the mechanism gets a content-warning paragraph. The video then loses to the ones that did not flinch, and those are not always the accurate ones.
How MagmaRouter is used for this
Give the model the topic, the audience and the hook; get a script in beats, with the mechanism stated plainly and the analogy checked for where it breaks. Ask for the three factual claims most likely to be wrong and verify those in the literature. Nothing is retained; your unmade video ideas stay yours.
A working loop
- Hook in the first sentence; ask for ten, pick one.
- State the mechanism plainly; the grim bit is the retention.
- Fact-check the three riskiest claims; check analogies where they break.
Prompts are not stored. A signed receipt of the hash comes back on every call, so you can prove later what MagmaRouter logged. Details are on the privacy page. Per-model prices are on /models and the pricing model is on /pricing.