Module 18: Who uses this: sectors and careers · Lesson 18.10 · about 20 min
Writers, creators and translators
Writers, video makers, designers and translators use local models for drafts, editing, subtitles, translation and images, keeping unpublished work and client texts private, with care for originality and credit.
Module 18 · about 20 minutes · read first: 13.4 and 13.6. You’ll be able to plan a private AI workshop for creative and language work.
1. Why this sector
Unpublished manuscripts, scripts, client translations and brand work are valuable and often confidential. Creators also use AI in long sessions that burn through cloud limits. A local setup keeps work private and has no cap.
2. Top uses
- Writers: brainstorming, outlines, line edits, consistency checks across a long manuscript (with document search).
- Translators: first drafts in common language pairs, glossary checks, terminology lists from past work.
- Video makers and podcasters: transcripts and subtitles (whisper.cpp), show notes, chapter titles.
- Designers: image ideas and mood boards with local image models (Lesson 13.6).
3. Care points
- Originality: check that output does not copy existing text; do your own creative work on it.
- Credit and disclosure: follow publishers’, platforms’ and clients’ AI rules.
- Licences: check image models’ licences for commercial use.
- Quality: small models translate common pairs well and rare pairs poorly; test on your own past work (Lesson 17.1).
A translator’s NDA (non-disclosure agreement) is a promise not to share the client’s text; a local model lets you keep it. A glossary is your list of agreed translations for key terms.
The job. Ravi translates insurance documents from English to Malayalam and Tamil under client NDAs. Nothing may be uploaded.
The glossary goes first. His file of agreed terms ("deductible", "rider", "claim") is put in front of every request, so the model uses the client's words every time.
The model drafts. A local model on his laptop writes a first draft, paragraph by paragraph.
Ravi edits every sentence. He fixes meaning, tone and the odd wrong term. The judgment, and the responsibility, stay his.
The result. His output rises from 2,500 to 4,000 words a day, and no client document leaves his laptop.
Maths and principle
Principle: price your time, not the AI’s.
Equation: effective rate = fee ÷ (drafting hours with AI + editing hours).
Analogy: a carpenter with a power saw still sands and finishes by hand.
Try it: a 4,000-word job at 400, 8 hours by hand, or 2 hours AI draft plus 3 hours editing? (50 an hour against 80 an hour, if quality holds.)
- fee
- what the client pays for the job, the same with or without AI
- drafting hours
- your time preparing the request, the glossary and the draft
- editing hours
- your time checking and fixing every sentence; never skipped
- Physical meaning
- The fee is fixed by the job, so every hour the model saves you raises what each of your hours earns. If the draft is poor and editing takes longer than writing would, the rate falls instead: test on your own past work first.
“If I pressed generate, the work is mine.” Ownership rules focus on human authorship: your choices, arrangement and rewriting (Lesson 17.7). Purely generated text or images may not be protected at all. Do real creative work on the result, and follow your client’s and publisher’s disclosure rules.
4. Worked scenario
Made-up translator. Ravi translates insurance documents from English to Malayalam and Tamil under client NDAs. A local model drafts; he edits every sentence and keeps a glossary file the model reads each time. His output rises from 2,500 to 4,000 words a day, and no client document leaves his laptop.
5. Practice
- Which of your creative tasks would benefit most from a local model?
- Write a disclosure line for AI-assisted work.
- How would you test a model’s quality in your language pair?
Answers
- Your answer; drafts and transcripts are common winners.
- For example: “Drafted with the help of an AI tool; edited and checked by the author.”
- Translate 10 passages you have already translated well, compare blind, and note error types.
New words: NDA, glossary, subtitles, disclosure.
- Local models keep unpublished and client work private and have no usage cap.
- Good uses: outlines and edits, first-draft translation with a glossary, transcripts and subtitles, image ideas.
- Effective rate = fee ÷ (drafting + editing hours): it rises only if quality holds after your edit.
- Ownership needs real human creative work; check originality, licences and disclosure rules.
Check yourself
Sources
- USA: Copyright Office, "Copyright and Artificial Intelligence, Part 2: Copyrightability", January 2025, https://www.copyright.gov/ai/
- whisper.cpp (offline speech to text and subtitles), https://github.com/ggml-org/whisper.cpp
- ComfyUI (local image generation), https://github.com/comfyanonymous/ComfyUI
Your ticks stay in this browser. Save them as a file on the course home.
Updated 2026-11-22. Prices and model names change fast: check the official source before you buy.