- Robustify encoding cache in A1111PromptNodeusing composite keys (CLIP ID + normalization + text) to prevent stale embeddings.
- Improve cache management by pre-calculating missing prompts and evicting old entries before the encoding loop starts.
- Optimize memory usage by limiting tensor padding to only the prompts within the current generation's schedule.
- Implement linear step index scaling in hooks.py for accurate scheduling across variable sampler step counts.
- Refactor long error messages into a dedicated StepCountRequiredError
- Robustify autocomplete UI by resetting internal selection state on backend fetch errors to prevent stale tag insertion.
imi
f97e4e6fb4