1000ish

I wanted to see if I could get a model to describe an image in exactly 1,000 words. Asking for that count wasn’t enough, so I built a small loop that counts the draft and asks the model to rewrite the ending until it fits.

I let the model write a first draft of roughly 1,050 words without asking it to do exact-count arithmetic. Then I keep the opening intact, choosing a sentence boundary between 800 and 900 words, and ask for a replacement ending with the remaining word budget. If the first draft hits 1,000 words and ends with a complete sentence, I keep it as is.

A small validation tool counts each proposed ending and tells the model how many words it needs to add or remove. It gets up to three attempts. Both the server and browser check the assembled text before marking it complete. If the retries fail, I leave the draft visible rather than pad it or cut it mid-sentence. I count whitespace-separated words, so contractions and hyphenated words each count as one.

I wanted the correction to be visible. The interface dims the original ending, backspaces it, and reveals the replacement. I buffer and validate that replacement before animating it, so you’re watching a presentation of the revision rather than live model edits. The “-ish” disappears only after the final count checks out.