One request, many models: API fan-out
Pass several model names in one generate call, get a request id per model, and run a bake-off without writing your own dispatch loop.
7 min read
Open-weight video licenses and what you can ship
Open weights do not mean open usage. Four clauses decide whether a downloaded video model can run a paid campaign — and which entity has to pass the test.
9 min read
The operating week of a solo AI content business
A block week that stops production eating outreach and delivery: named sales days, batch days, and one day for review, invoicing and next week's briefs.
9 min read
Operation AI Comply and claims about your AI
FTC Section 5 applies to what you say your AI does, not just what you generate with it. Capability claims rewritten so they survive a deception review.
8 min read
Output licence terms compared across generators
Assignment, licence-back, revenue thresholds, IP warranties and indemnity availability differ by generator. The five clauses to read before a deliverable.
8 min read
Your upscale made the flicker worse
Sharpening an upscale past a conservative setting amplifies temporal noise. Stabilize the clip first, then upscale gently so flicker dies instead of blooming.
8 min read
Packaging problem or content problem?
Click-through and retention form a two-by-two that tells you whether to repackage or re-edit. Most creators re-edit videos that only needed a new title.
9 min read
Padding cliffs and publishing at true length
Stretching a six-minute idea to clear a runtime threshold leaves a visible seam in the retention graph. How to spot it and compare per-video return honestly.
9 min read
Part 2 inherits Part 1's audience
Finishing Part 1 is a strong interest signal, so Part 2 often arrives without search. Build a Part 1 that earns the handoff, then track their views ratio.
9 min read
Pediatric therapy video without filming children
ABA, speech and OT clinics cannot build a video library on real clients. How to use proxy avatars for parent onboarding and session-preview content instead.
9 min read
Peeking is why your creative tests lie
Early data favours whichever variant posted first and often reverses with volume. Pre-commit a sample size and a stop rule before you look.
8 min read
Pending AI training cases: a user's risk map
NYT v. OpenAI, the Disney and Warner suit against Midjourney, and the Anthropic lyrics cases are all in discovery. What a downstream user should infer.
8 min read
Per-key rate limits as a blast radius control
Per-key request-per-minute limits contain a runaway loop. Give CI, staging, and production different ceilings so one job stalls instead of draining the account.
8 min read
Per minute, per deliverable, or retainer?
The same month of video work quoted three ways, showing where each model earns and leaks, plus a rule for choosing on clip length and revision risk.
10 min read
Picking the chat model behind the agent
The chat model plans; the generate model renders. Change chat_model to pick the brain, and pin generate models separately when the clip has to match.
7 min read
Pickleball club video: open play and league fills
Court hours are perishable inventory. A 15-second clip per session fills off-peak better than any facility tour, plus the releases member footage needs.
9 min read
Pin to endpoints, not agent tool names
Agent tool names move. REST route groups do not. Build integrations against endpoints, and treat tool names as a convenience layer on top.
8 min read
Where Pinterest's AI modified label appears
Pinterest renders its AI modified label bottom-left in close-up view and hides it behind the ellipsis on ads. Here is a Pin layout that avoids the collision.
8 min read
Pinterest labels AI without any metadata
Pinterest labels from IPTC metadata, owner disclosure, and classifiers that catch generated content with no obvious markers. Stripping files changes nothing.
8 min read
Pinterest publishes no downranking data for AI Pins
Everyone assumes labelled Pins get suppressed and Pinterest has said nothing either way. Here is a paired-Pin test that answers it for your own account.
9 min read
Pinterest reference search inside the agent
Describing a look in words is lossy. How to pull real reference images into the agent conversation and brief a generation from a picture instead of a paragraph.
8 min read
Pinterest's see fewer GenAI control and reach
Pinterest's see fewer GenAI control is category-scoped and expanding. Size your exposure from topic mix and labels, not from an unpublished opt-out rate.
8 min read
Place CTAs at 55-75% of runtime
End-screen CTAs reach the smallest remaining audience. Place the ask at 55-75% of runtime, right after a payoff, then A/B the timestamp.
9 min read
Plastic skin is a guidance problem
Airbrushed mannequin skin is high guidance amplifying low-frequency signal. A working CFG band, texture vocabulary, and why a fine-tune beats prompt patches.
8 min read