How to spend H3's nine reference slots
MiniMax H3 takes nine reference images, three videos and three audio clips. A working slot budget for a character-plus-product series, and what to drop first.
9 min read
Mine rewatch spikes for your next thumbnail
A bump above the retention line is a moment the audience chose. Turn spikes into thumbnails, clips, and re-hooks, then test them against a control.
8 min read
Missing required params are why agents stall
A stalled agent turn is usually a missing required input, not a bad prompt. Diagnose the tool schema and model preflight before rewriting anything.
8 min read
More steps will not fix a soft image
Cranking sampling steps almost never fixes softness. A three-question diagnosis separating resolution ceiling, sampler mismatch and guidance from step count.
9 min read
Motion blur and shutter language in prompts
Motion blur names an outcome, not an instruction. Shutter and exposure phrasing for crisp versus smeared movement, and when to add the blur in post instead.
9 min read
Motivated lighting: name the source in frame
Name a light that exists in the shot and shadows land consistently. Motivated-light phrasing for interiors, night scenes and product work, plus the tells.
10 min read
MSP video: what your SOC does at 2am
Vendor logo walls do not differentiate an MSP. A dashboard screen-capture explainer of your response process does, with two rules that keep it publishable.
9 min read
Muse Video ranks fifth and nobody can use it
Meta's Muse Video sits fifth on Arena's text-to-video board and has never shipped. A four-question filter that strips unusable models out before you plan.
8 min read
Name two foley events, and no more
Native-audio models place a limited number of sound events per clip. Name two specific, countable sounds and you get both; name six and you get mush.
9 min read
Naming background tasks so re-runs reuse them
A background task name is the saved-workflow key. Name it by intent so a later re-run updates one recipe instead of spawning a v2 row.
7 min read
Negative hooks vs benefit hooks: run the pair
Loss-framed openers usually beat benefit-framed ones, but the size of the gap is yours to measure. A paired-post protocol that produces your own number.
9 min read
New York's synthetic performer ad disclosure
Since 9 June 2026, ads reaching New York consumers with a synthetic human performer need a conspicuous disclosure. Scope, exemption and penalties.
9 min read
YouTube won't monetize these AI persona niches
YouTube will not monetize an AI persona giving health, legal, or financial advice. Reporting a named source can stay. A disclaimer does not fix the avatar.
9 min read
Night scenes that don't turn to mud
Low-light prompts come back grey because nothing in frame motivates the exposure. The one-hard-source-plus-one-practical recipe and the post pass order.
8 min read
Nine caption families across 45 presets
Versely's 45 caption presets are nine families of five. Map them by backing, position and word timing to shortlist three candidates in a couple of minutes.
8 min read
Nine specs a model page should publish
The nine facts that decide whether a video model fits a workflow, what each omission actually costs you, and how sparse most published model records really are.
9 min read
'No clear winner' is a result
An inconclusive Test & Compare result is information, not a glitch. Why variants were too similar or impressions too thin, and what to change before you re-run.
9 min read
NO FAKES is still a bill, not a law
S.4591 advanced out of Senate Judiciary on 18 June 2026 and has not been enacted. What it would do, and why state law is the operative regime today.
8 min read
No federal likeness right, fifty state regimes
The US has no federal right of publicity, so likeness exposure follows the subject and the audience. How to set a governing-law posture that actually holds.
8 min read
On-screen text in generated video: what holds up
Text a video model draws inside the frame has to be right 200 times, not once. The size threshold where letterforms break, and the clean-plate rule.
9 min read
One timeline call or a chain of edit tools
Chained cut, attach, and caption calls multiply round trips and intermediate files. Use one timeline render when the output is a single finished video.
9 min read
One generation with cuts, or three clips stitched
Directing three shots inside one generation versus generating three clips and cutting them, compared on continuity, per-shot control and what a retake costs.
8 min read
One model for image, video and audio?
FLUX 3, MiniMax H3, Wan 3.0 and Gemini Omni Flash train all three modalities at once. When one model beats a best-of-breed stack, and three jobs where it loses.
8 min read
One-prompt campaign generators: what they miss
Brief-in campaign-out agents genuinely solve variant volume and resizing. Offer logic, brand exceptions, claims, and the hook still need a human.
9 min read