AI Transparency Statement
Version 1.0 · 22 August 2026
In one paragraph
MOVIEWRITER uses third-party AI language and image models to read your screenplay and write readings, reports, suggestions, scores and images from it. Seven of its applications do this; three do not. Anything a model produced is labelled as AI-generated where it is shown, is non-deterministic (a re-run gives a different answer), and is offered as a reading to inform your own judgement — not as fact. Your text is sent to our AI gateway and on to the provider named for the feature; it is not used to train any model. This statement is version 1.0, dated 22 August 2026.
What each application uses AI for
Scriptwax AI text
- What AI does
- Answers a fixed catalogue of about sixty polish questions across twelve dimensions (structure, characters, dialogue and so on), giving each an answer, an optional suggestion and a severity.
- What is not AI
- The question catalogue itself is written by us, not the model.
- Your control
- Results cannot be edited — you read them, rate them, and re-run when you want a fresh pass. Nothing is written back into your screenplay.
- Documents
- Polish Report PDF.
Coverage AI text
- What AI does
- Produces industry-style coverage: a Pass / Consider / Recommend verdict with a note, six category grades with rationale, strengths and weaknesses, a synopsis, comparables, tone, and a pitch for the sell sheet.
- What is not AI
- The six-category framework and the verdict scale are fixed by us.
- Your control
- Every field, grade and the verdict is editable. The AI original is kept alongside your edit so you can revert, and edited fields are marked.
- Documents
- Coverage Report PDF (for you) and Sell Sheet PDF (to send out).
ScriptLens AI text
- What AI does
- Writes a reading of the draft through a chosen lens (Plot, Roles) as prose sections, plus a list of findings with severity and scene references.
- What is not AI
- The lens catalogue and section headings are ours.
- Your control
- Sections are editable with the AI original kept for revert; findings can be marked open, fixed or dropped.
- Documents
- ScriptLens PDF per lens.
Character AI text
- What AI does
- Suggests who the protagonist and antagonist are; suggests a character's want, need, flaw, ghost and arc; finds quotations in the script that support or contradict the intent you declared; scores twenty transformation indicators (0–5) with a rationale; describes each cast role; and writes a thirteen-section reading with findings.
- What is not AI
- Dialogue and scene counts, the Main/Minor classification, the dramatised / asserted / untested status of each claim, and the ECS-100 subtotals are computed by code. Every quotation the model cites is checked against the script text and discarded if it cannot be found.
- Your control
- AI-suggested intent must be read and accepted (or edited) by you before the evidence pass will run; the record keeps whether it was accepted unchanged. Scores and reading sections are editable with the AI original retained.
- Documents
- Character PDF.
Storyliner AI text
- What AI does
- For each storyline, writes a summary, classifies its type, estimates its weight and depth, lists its beats with supporting quotations, and scores it against a set of lenses with a rationale. Also suggests storylines that may be missing or could be cut.
- What is not AI
- The storylines themselves are identified by which characters share scenes — a count, not a model — and the timeline is built from those scenes.
- Your control
- Beats can be edited or deleted; triage (keep / cut) and notes are yours. Developing an AI suggestion copies its rationale into your note, where you should rewrite it in your own words.
- Documents
- Storyliner Map PDF.
Scriptplanner AI text
- What AI does
- Writer-first: the plan, characters, storylines and scenes are yours. AI is used only for the Coherence check — a list of findings (area, severity, detail) about the plan as a whole — and, on request, a suggested fix for a single finding.
- What is not AI
- Nothing in the plan is written by the model; the three-act skeleton comes from a template.
- Your control
- Findings can be dismissed; suggestions can be regenerated. Neither is copied into your plan unless you type it there.
- Documents
- Coherence Report PDF (AI); Story Bible and Outline PDFs (no AI content).
ScriptFAQ AI text
- What AI does
- The Structure analytics — paradigm beat maps (three-act, Hero's Journey, Save the Cat, eight-sequence, Propp), emotional arc, tension curve, scene function and theme grid — are generated by the model and are non-deterministic.
- What is not AI
- Every other analytic (script statistics, dialogue/action ratios, pacing, cast size, presence timelines, sentiment, profanity, locations) is computed directly from the text.
- Your control
- AI analytics are not editable; re-run for a fresh pass. Each result is marked ✨ AI and carries a note saying so.
- Documents
- ScriptFAQ dossier PDF.
ScriptQA AI text
- What AI does
- Every answer is generated by the model from the writer's question plus the scene's text and card (or the character's lines, action mentions and metadata). Answers are instructed to cite scene numbers and to say where the script is silent, but they remain a model's reading and are non-deterministic.
- What is not AI
- The questions are the writer's own words. The scene and character lists, and the material the model reads, come directly from the script.
- Your control
- No answer is kept without the writer choosing Keep; Discard deletes it, Ask again replaces it, and kept answers can be deleted at any time. Only kept answers appear in the dossier PDF.
- Documents
- ScriptQA Q&A dossier PDF.
Storyboard AI images
- What AI does
- Generates panel images from your shot description (with the world, character descriptions and any reference images you attach folded into the prompt) and reference portraits of people, places and things from their descriptions.
- What is not AI
- You can upload your own drawings or photographs instead of generating; the board records for each image version whether it was generated or uploaded.
- Your control
- Every generated attempt is kept with its prompt, vendor and model; you choose which version a panel shows.
- Documents
- Storyboard, scene, panel, cast and contact-sheet PDFs.
Screenplays No AI calls
- What AI does
- Makes no AI calls. It holds your drafts and exports them.
- What is not AI
- Character tags shown here may originate from the Character app's AI suggestions.
- Your control
- If you paste AI-generated text from a sibling app into a draft, it becomes part of your text. Declare it with the "contains AI-assisted text" setting on the screenplay's title page; the declaration travels with every export and the public copy.
- Documents
- Screenplay PDF (with a statement page when declared) and Fountain export (with an "AI disclosure" key when declared).
Publisher No AI calls
- What AI does
- Makes no AI calls. It publishes a pinned draft to a share link and binds the sibling apps' reports into one Bible PDF.
- What is not AI
- The Bible's chapters (Coherence, Storyliner, ScriptLens, Character, Scriptwax, Coverage) are AI-generated readings produced earlier in their own apps.
- Your control
- What you publish is your decision and is visible to anyone with the link. If the screenplay carries the author's "contains AI-assisted text" declaration, the public landing page, reader and PDF all say so.
- Documents
- Public screenplay PDF (with the author's declaration when made); the Moviewriter Bible PDF (with its own statement page).
Which models, and whose keys
All AI requests go through our gateway at ai.vizihost.com, which we operate. It holds the provider keys, prices each call, and forwards the request to the provider. The defaults are:
| Provider | Used for | Default model | |
|---|
| Anthropic | Text — all text-based features (default) | claude-sonnet-4-20250514 | provider terms |
| OpenAI | Text (alternative) and image generation | gpt-4o · gpt-image-1 | provider terms |
| xAI | Text (alternative) and image generation | grok-3 · grok-imagine-image | provider terms |
New accounts run on our provider keys for a 30-day free trial with a spend cap. After that you add your own provider keys (AI ▾ → My AI keys); they are stored encrypted here and at the gateway, used only for your calls, and removed from the gateway when you deactivate or delete them. With your own keys the provider bills you directly under its own terms.
What is sent, what is kept
- Sent to the provider: the material the feature needs — usually the full text of the screenplay or plan you are working on, your settings for that feature, and for images your prompt plus any reference images you attached. To make repeated passes cheaper, the script text may be sent as a provider prompt-cache prefix, which the provider holds for the short window its terms describe.
- Kept by us: the model's response, in the application it belongs to; and a call record with the application, model, token counts, cost, whether it ran on our keys or yours, the time, and the prompt text (first 2,000 characters on this server; the full prompt at the gateway). You can see your own call records under AI ▾ → My AI usage.
- Not done: we do not use your content or the model's output to train any model, and we have no arrangement for a provider to do so.
- Retention and deletion: call records are deleted after 12 months here and at the gateway; deleting your account erases your content, keys and call records on this server and asks the gateway to erase its copies in the same step. Details in the Privacy Policy.
How AI output is labelled
- On screen: results generated by a model carry an AI marker (a badge, a note, or the model name) where they are displayed; where you have edited an AI result the marker says so and the original is kept for revert.
- In documents: every PDF exported from the AI-using applications, and the Moviewriter Bible, carries an AI Transparency Statement in its front matter (a page after the cover, or a block under the title) naming what in that document is AI-generated and which model produced it, a "contains AI-generated content" note in the page footer, and the same declaration in the PDF's Subject and Keywords metadata so software can detect it. Documents with no AI content say so the same way. The screenplay text itself carries no automatic marker — software cannot tell where pasted text came from — so for screenplays the declaration is the author's: a "contains AI-assisted text" setting on the screenplay (Screenplays → edit → Title page) that prints a statement page in the screenplay PDF, an "AI disclosure" key in the Fountain file, and a notice on the public Publisher pages. A screenplay without the declaration makes no claim either way.
- Watermarks and file marking: text from these models carries no watermark — the label is ours. Images generated through OpenAI include embedded Content Credentials (C2PA) added by OpenAI, which we store byte-for-byte so they stay valid. Images generated through xAI arrive with no marking, so we embed our own in the file: an XMP packet carrying the IPTC "trained algorithmic media" digital-source type, the model name, and a link to this statement — readable by exiftool, Photoshop and most asset managers. Neither is a pixel watermark: both survive copying and renaming but not re-encoding, so images placed inside a PDF are labelled on the page instead.
Limits you should know
- Models misread, invent, and contradict themselves between runs. Every score, verdict and quotation is a model's best reading of your text and should be checked against the page before you rely on it or send it to anyone.
- The transformation rubric (ECS-100) is an editorial framework, not a validated psychometric instrument.
- Image models may produce likenesses you did not intend; do not use reference images of real, identifiable people without their consent.
- AI features need a working provider key and an internet connection; they are unavailable when the trial cap is reached and no key of your own is active.