A model with no generation ability drives external tools

This shows, screen by screen, a setup that puts a text model unable to produce scenes on its own at the command line and directs it through conversation to handle everything on an external generation platform, from logging in and authentication to model selection and re-runs. Three pieces are worth taking away. The first is an arrangement that separates intelligence from execution: the brain side only collects and preprocesses material and chooses tools, while the actual generation is left to external models, so swapping one side out doesn't shake the other. The second is a cost gate that, right before a re-run, first gives an estimate of the credits it will consume and gets approval, a working example of a mechanism that any agent holding consumable resources must have. The third is a reverse constraint that forbids perfection, with the finding that realism only emerges once you suppress the generation model's tendency to idealize lighting and textures. However, it was a human eye that found the flaws, there is no automated checking mechanism anywhere, and no figures for cost or failure rate are given.

This is a note: my own write-up after reading several sources. It has no source link; events checked against a source are kept as records.