@officexapp/vidfarm-devcli 0.21.27 → 0.21.29

package/.agents/skills/vidfarm/regimes/ugc-testimonial.QA_REGIME.md

@@ -0,0 +1,82 @@
+---
+name: ugc-testimonial
+video_type: UGC / talking-head testimonial — a person on camera vouching for a product
+checks:
+  duration_sec: 12-60
+  aspect: 9:16
+  first_frame_visual: required
+  first_frame_text: required
+  captions: required
+  audio: required
+  font_regime: required
+  max_text_cards: 2
+  max_simultaneous_text: 1
+---
+
+# UGC / Testimonial QA Regime
+
+For the format where a person talks to camera about a product they use. The entire value of the format is that it **doesn't look produced** — so most of this regime is about what NOT to add.
+
+## The one test
+
+> **Would this survive being posted from the speaker's own account, with no brand involved?**
+
+If the answer needs the brand's permission, budget, or logo, it isn't UGC — it's an ad wearing UGC's clothes, and the audience reads that in under a second. The failure is not "it looks cheap"; it's "it looks funded," and funded means skippable.
+
+## Structure
+
+| Beat | Runtime | Job |
+|---|---|---|
+| **Situation** | 0:00–0:03 | The problem, as a thing that happened to *them*. Not a claim about the product |
+| **Turn** | 0:03–0:10 | What they tried, what changed. The product enters here, never in the first line |
+| **Demonstration** | the middle | Screen or hands. The product doing the thing, on their real account/desk/kitchen |
+| **Honest limit** | before the close | What it doesn't do / who it isn't for. **The credibility beat** |
+| **Close** | final beat | What they'd tell a friend. Not a CTA read |
+
+**The product enters second, never first.** A testimonial that opens on the product is a commercial. It opens on the *situation* — and situations are also what survives a cold algorithm (see the `hooks` regime).
+
+## Rules
+
+### Rule 1 — first person, specific, and falsifiable
+
+"It saves me time" is an ad line. "I used to spend Sunday night on this and now I don't" is a testimonial. Every claim should be a thing that happened, with a when and a how much. Vague praise reads as paid; specific mundane detail reads as real.
+
+### Rule 2 — one honest limit, said out loud
+
+The beat that makes everything else believable. It is not a disclaimer and it does not go at the end as fine print — it goes in the body, in the speaker's own voice, and it should cost you something real. A testimonial with no limit is a review nobody believes.
+
+### Rule 3 — never claim an outcome the viewer's situation can't reproduce
+
+Results-may-vary in tiny type does not fix a headline result. If the number depends on the speaker's audience, budget, or luck, either say what it depended on or don't say the number.
+
+### Rule 4 — production floor: deliberately low
+
+- **Audio is the only thing allowed to be good.** Bad audio is the one production sin the format doesn't forgive; everything else *should* look handheld.
+- **No b-roll montage, no music swell, no logo sting, no lower-third name card.** A brand logo on screen converts this from a testimonial into an ad.
+- **Captions verbatim**, weight 700–900, safe zone, one line at a time. Never kinetic VFX captions — produced captions are the single loudest "this was made by a marketing team" signal.
+- **At most two standalone text cards** in the whole video, and only to surface a number the speaker said out loud.
+- **Frame 0:** the speaker already mid-sentence, in frame, with a caption up. Not a title card, not a product shot, not black.
+
+### Rule 5 — disclosure is not optional
+
+Paid, gifted, or affiliate relationships get disclosed on screen and in the caption. Beyond the legal exposure, an undisclosed relationship that surfaces later costs more trust than the video ever bought.
+
+## Bulk-generation notes
+
+The natural variant axis here is **the situation, not the script**: five different opening situations pointing at the same turn and demonstration. Varying the adjectives instead of the situation gives you five videos that are one video — and the algorithm treats near-duplicates accordingly.
+
+If you're generating variants with different speakers/avatars, hold the honest-limit beat constant. It's the load-bearing one, and it's the one an AI rewrite is most likely to quietly soften into nothing.
+
+## Pre-flight checklist
+
+- [ ] The first line is a situation that happened to the speaker, not a product claim
+- [ ] The product enters in the second beat or later
+- [ ] Every claim is specific enough to be falsifiable (a when, a how much, a what happened)
+- [ ] There is one honest limit, in the body, in the speaker's own voice
+- [ ] No outcome is promised that the viewer's situation couldn't reproduce
+- [ ] The demonstration shows the real thing on a real account/desk, not a mockup
+- [ ] Audio is clean; nothing else looks produced
+- [ ] No logo sting, no music swell, no lower-third, no kinetic captions
+- [ ] Any paid/gifted/affiliate relationship is disclosed on screen and in the caption
+- [ ] Frame 0 shows the speaker mid-sentence with a legible caption
+- [ ] In a batch: this variant's *situation* differs from its siblings, not just its wording