Synthetic content — AI-generated images, voice/face synthesis, CGI recreations, generative video and audio — is an enormous creative tool. It lets storytellers imagine, repair, illustrate and delight in ways previously impossible. It also introduces new risks: deception, harm to reputation, bias, and loss of agency. This post walks through practical, concrete steps studios, directors, producers and clients can use to steward synthetic content responsibly — without moralising, and with the goal of keeping great creative work honest and safe.

Before you generate a single frame, answer these:

• What story am I trying to tell, and why does it need synthetic content?
• Who could be helped, and who could be harmed by this piece?
• Is the subject a private person, a public figure, or a historical person (living or dead)?
• If the creative gain is small and the risk of misrepresentation is high, choose another approach.

Practical rules:

Get explicit written consent from any living person you intend to synthesise (voice, face, likeness). Use a short, clear release that specifies usage, territory, duration, and ability to revoke where possible.

For deceased persons or public figures, escalate approvals: family, estate, legal counsel, or a client-side ethics reviewer. Treat these as high-risk uses.

If you’re using or fine-tuning models on private or proprietary data, confirm you have the rights to do so.

Simple principle: when in doubt, get permission or do not proceed.

Audiences deserve to know when content is synthetic.

Add clear, conspicuous labels in-frame (for video), in captions, and in metadata: e.g., “This clip contains synthetic voice/CGI reconstruction.”

Attach provenance metadata to masters (date, model used, inputs, operator, version). Use descriptive alt text for images and transcripts for audio.

Transparency reduces harm and builds trust.

Practical actions:

Maintain an immutable audit log of prompts, model versions, training data provenance (as available), and post-processing steps.

Embed provenance: visible watermarking for consumer contexts plus cryptographic metadata (content credentials) for archival/enterprise contexts.

You don’t need to publish everything publicly, but you must be able to prove how a piece was made.

Never treat models as final decision-makers for sensitive content.

Require human review for any content that depicts real people, reenacts real events, or could influence public opinion.

Set up a simple internal review board (2–3 people) for sensitive cases: a creative lead, a legal/rights person, and an ethics reviewer (could be external).

Red-team: run a quick “misuse test” — could this look like real footage? If yes, rework the approach.

Design style intentionally:

For reconstructions/hypotheses, prefer non-photoreal visual styles (graphic, illustrative, or motion-graphic) so the audience can’t mistake simulation for documentary footage.

If photorealism is required, increase transparency and provenance and budget for extra legal and editorial safeguards.

Hard rule: avoid synthetic recreations that could mislead about real-world events, especially politics, ongoing legal matters, or public safety.

In news or documentary contexts, prefer archival footage, interviews, or clearly labelled illustrative animation.

If synthetic content is necessary to explain an idea, mark it as interpretation and provide sources.

Practical steps:

Audit outputs for stereotyping and skewed representation (e.g., how a synthesised voice or appearance maps onto gender, race, age).

Use diverse reference sets and consult cultural experts when depicting historically marginalised groups.

Document corrective steps taken (and the remaining limitations).

Protect the inputs and the people behind them:

Limit who can access raw recordings, training data, and models.

Encrypt stored assets, log access, and enforce short retention policies for sensitive material.

Don’t share private voiceprints or biometric identifiers without explicit, documented consent.

Before release:

Screen sensitive synthetic pieces to a small, diverse group of uninvolved viewers and ask: “What did you think was real?” If more than a small minority misattribute reality, tighten disclosures or change the treatment.

Test edge cases (low bandwidth, social crop, thumbnail) — how does the content read when resized or recontextualised?

Make ethics part of the price:

Include line items for rights clearance, consent wrangling, provenance capture, red-team testing, and legal review. These are legitimate, billable production tasks.

Contracts should specify who owns synthetic assets, who may reuse or retrain models, and who is liable for misuse.

Have a clear, public remediation policy:

If someone raises an issue, respond quickly: take down the asset if appropriate, publish a correction/explain the provenance, and provide a transparent timeline for fixes.

Maintain a public “ethics notes” page per project for high-sensitivity works (short summary: purpose, methods, reviewers).

Operationalise ethics so it’s repeatable:

Publish a short studio policy (1–2 pages) listing permitted/forbidden synthetic uses, approval steps, and required documentation.

Train producers, creatives and clients on those rules — 30–60 minute sessions and a one-page checklist keep everyone aligned.

We’re not alone. Encourage:

Use of content credentials standards (C2PA or equivalent) and industry watermarking.

Sharing of best practices between studios, platforms, and networks to make responsible approaches the default.

• Intent documented? ✔️
• Consent/releases acquired? ✔️
• Transparency label included? ✔️
• Provenance & audit log stored? ✔️
• Human review completed? ✔️
• Bias/cultural audit passed? ✔️
• Security & retention policy set? ✔️
• Remediation plan ready? ✔️

Ethics here is not a creative handbrake — it’s craft hygiene. It protects your audience, your talent, and your studio’s reputation. When you plan synthetic content with clear intent, documented consent, visible disclosure, and human oversight, you unlock powerful storytelling without trading trust for spectacle.