What to log on every generation for compliance
A per-asset record of model, version, prompt, references and human edits answers registration, regulator and client-audit questions. The fields to capture.
9 min read
What to tell clients about using AI in delivery
Contract language covering AI production, where audience disclosure is actually required, and the framing that sells iteration speed over cheap labour.
9 min read
What to write instead of a negative prompt
Negative prompts are a no-op on flow-matching models and leak the negated noun elsewhere. Positive re-specification and the three constraints that replaced it.
9 min read
What turbo tiers actually remove from a model
Fast, lite and mini variants cut sampling depth, resolution, reference slots or options, and each cut leaves its own artifact. Predict which shots survive.
9 min read
What unlimited generation offers actually cap
Time-boxed unlimited promotions cap resolution, model subset and queue before they cap volume. Six levers to check, and how to find the real ceiling.
9 min read
What 'you own the output' actually means
Vendor terms can assign you an output file. They cannot make it copyrightable. How to separate contractual ownership from protectability before you ship.
9 min read
What YouTube counts as mass-produced video
Mass-produced is a two-part test in YouTube's July 2026 wording, and volume is not one of them. A definition you can act on, plus a batch pipeline that passes.
9 min read
When a 4K export is worth the credits
Most social delivery wastes the 4K premium. The three cases that earn it, the reframe arithmetic behind them, and a per-destination resolution rule.
8 min read
When a client contract forbids AI tools
Blanket no-AI clauses are usually aimed at a risk you are not creating. A decision tree for complying, negotiating a scoped exception, or declining.
9 min read
When an open model line goes closed
Qwen's image line went from Apache-2.0 to no licence in one version. A stack design that survives licence reversal: what to pin, mirror, and never depend on.
9 min read
When the agent picks the wrong sub-agent
Misrouting is usually ambiguous phrasing, not a broken agent. Word the first sentence so only one of the six domains can claim the job.
9 min read
When to hire your first AI specialist
The one-specialist-per-five-creatives ratio is a hypothesis, not a benchmark. Four trigger conditions drawn from your own throughput and rework data instead.
9 min read
When to subcontract a human UGC creator
Where AI UGC still loses: hands-on unboxing, food, worn-product fit and verifiable testimonial. A decision table plus a pass-through that protects margin.
9 min read
When your AI ad has no copyright
A competitor is running your hero frame with their logo on it and you own nothing. The five levers that never depended on copyright, ranked by what works.
9 min read
Where motion coherence breaks as clips get longer
A 15-second ceiling is rarely 15 usable seconds. The three ways long clips fall apart, a ladder test to find your safe duration, and where to place the cut.
9 min read
Which image models hold architectural perspective
Bowed lines, disagreeing vanishing points and drifting facade counts sink generated architecture. The traits that help, and the prompt that fixes interiors.
9 min read
Which stock agencies still accept AI submissions
Shutterstock and Getty refuse contributor AI outright. Adobe Stock accepts it with a declaration — here is the one viable channel and what it actually pays.
8 min read
Which video models hold up with crowds in frame
Background people break more shots than hero subjects. The pixel budget that predicts failure, plus staging tactics for street, stadium and venue shots.
9 min read
Which video models speak more than English
Native audio is a capability flag, not a language list. A one-generation test for any model, and the rule for generating natively versus dubbing after.
8 min read
Which voice models accept inline audio tags
Four speech providers parse bracketed delivery cues and four take mood as a parameter. The two camps, four incompatible syntaxes, and how to convert a script.
10 min read
Who indemnifies what in an AI contract
Indemnity splits two ways: you cover tool terms-of-service breaches, the client covers distribution and labelling. A two-way split to drop into an agreement.
9 min read
Who owns the prompt library you built?
Prompts, reference sets and workflow graphs built on a client job are unassigned in most contracts. Clause language that keeps method and gives outputs.
9 min read
A batch of eight is not eight single runs
Batch size changes the noise each sample gets. A batched call and eight parallel calls diverge, even with the same seeds. Know when that variance is harmless.
9 min read
Why a seven-second clip costs more
Tiered models sell duration in whole steps, so seven seconds buys the ten-second price. How to write scene lengths that land inside a band instead of past it.
7 min read