Issue 002 · June 2026/Independent vendor research

See what AI-video platforms actually deliver before you subscribe.

We tagged 132 1-star reviews across Runway, Higgsfield, Krea, Pollo, Pika, Luma, Sora, and Vidu. Between 42% and 78% cite billing predation, not bad output. AVA surfaces the real per-second cost, unlimited-tier gating, and first-try success rates before you commit. After you subscribe, the same engine scores every prompt against 105 failure modes.

No credit card on the free tier · Chrome · Brave · Edge · Arc

Not ready to install? Get the digest in your inbox.

One email when the vendor changelog flags something.

We watch silent pricing-page edits, 'unlimited' policy walks, and credit-burn rule shifts across 11 AI-video vendors. You get a note only when a change matters. Maybe twice a month, often less.

One email when we launch + maybe one followup. No marketing spam, ever. Unsubscribe one-click.

Generations that fail

~33%

Aggregate across 8 video providers

Catalogued failure modes

105

Across 8 providers

Trustpilot 1-stars tagged

132

8 vendors · 8 months

Cost to install AVA

$0

Free tier · no card

Same prompt · four platforms · four failures

Each platform breaks the same prompt in a different way.

Kling 1.6

Same prompt

Google Veo 3.1

Same prompt

Seedance 2.0

Same prompt

Vidu Q2 Pro

Same prompt

Broken motion, garbled signs, melted hands, color drift. AVA learns the per-model failure pattern and warns you which provider will choke on which prompt — before the credit is spent.

What AVA does

Check the vendor. Score the prompt. Keep the credit.

Four signals run inside the extension — before you subscribe and while you draft — so the money you spend produces a clip you keep.

01 · Vendor reality check

See what subscription actually buys.

Per-second cost (credits × failure rate), unlimited-tier gating rules, first-try success rates, and pricing-change history. Built from 132 tagged Trustpilot 1-stars across 8 vendors plus 8 months of fetch-intercept data.

02 · Prompt indicator

A live failure score on every prompt.

Red, yellow, green — trained on 105 catalogued failure modes across Runway, Luma, Sora, Pika, Kling, Veo, Hailuo, Vidu. Updates per keystroke. No cloud calls. Green light before the credit is spent.

03 · Rewrite suggestions

Names the phrase pulling the score down.

"Fingers in frame." "120fps slow-mo." "Neon text." Each red prompt gets a concrete rewrite, not a generic warning. Edit until the indicator turns.

04 · Personal failure history

Your repeating mistakes, surfaced.

Every generation feeds your private history. "You've hit hand-anatomy failure 7 times on Runway this month." The aggregate forecast learns which platform survives your prompt habits.

The math vendors do not publish

Your real cost per clip isn't the list price.

It is the list price divided by how often a generation works on the first try. AVA estimates that ratio before you click Generate so you can pick the vendor — and the prompt — that actually clears.

Effective cost per usable clip

list_price × (1 / first_try_success_rate)

On the vendors we track, a 35% first-try success rate turns a $0.50 disclosed cost into a $1.43 real cost. The disclosed price stays the same. Every red prompt you scrub before you click Generate moves you back toward $0.50.

list_price

What the vendor advertises. AVA reads the live page so you do not have to translate credits to dollars.

first_try_success_rate

Estimated from 105 catalogued failure modes plus your personal history. Climbs when your prompt scores green; the indicator names which phrase is dragging it down.

Full per-platform breakdown in the 132-review research corpus.

Under the hood

How it works.

  1. 01

    Check the platform before you subscribe.

    Install the extension and pick a vendor you are considering. AVA surfaces per-second cost, unlimited-tier gating, first-try success rates, and the public change history. Free on every tier.

  2. 02

    Score every prompt you draft.

    After subscribing, AVA runs your prompt through 105 known failure modes per vendor — Runway, Luma, Sora, Pika, Kling, Veo, Hailuo, Vidu — and emits a red/yellow/green score in browser, zero API roundtrips.

  3. 03

    Rewrite until the score turns.

    When red, AVA names the specific phrase pulling the score down and offers a rewrite. Edit, watch the indicator climb, generate when the light is green.

  4. 04

    Build your private failure pattern.

    Every generation feeds your history. After a few weeks the aggregate tells you which vendor is most likely to succeed on this specific prompt today.

Pricing

Free checks the vendor. Pro scores every prompt.

Vendor reality checks are unlimited on every tier. Pro unlocks unlimited prompt scoring, personal failure history, and silent-policy-change alerts.

Free

$0forever
  • ·Unlimited vendor reality checks
  • ·50 prompt scores per month
  • ·Red / yellow / green indicator
  • ·Concrete rewrite suggestions
  • ·Coverage across 8 video providers
Install free
Most chosen

Pro

$19per month
  • ·Unlimited prompt scoring
  • ·Personal failure-history dashboard
  • ·Change alerts on pricing + tier gating
  • ·Cross-model comparison per prompt
  • ·L2 deep analysis on borderline scores
See Pro

Business

$79per month
  • ·Everything in Pro
  • ·API access to failure-mode catalog
  • ·Team dashboards (per-seat history)
  • ·Cross-platform change feed
  • ·Priority email support
Talk to sales

Interactive L1 engine

Test your Runway prompt.

Paste what you are about to send to Runway. The heuristic engine that runs inside the extension checks it in real time — no install required.

Prompt Input
5
5s
L1 Diagnostic Output

Paste a prompt to start diagnosis
← or load a sample

L1 checks heuristic rules only. Install the extension for failure-frame marking and audit reports.

Privacy-first architecture

Free to use. Nothing to trust us with.

Sign in with Google or email — that is it. Tokens are AES-GCM encrypted locally. No cloud AI vision calls. No prompts stored on our servers.

Encrypted at rest

Session tokens are AES-GCM encrypted before local storage. No plaintext credentials touch disk.

No cloud vision

All diagnosis is local L1 heuristics plus your manual marks. Zero AI-vision calls on our backend.

In-browser inference

Heuristic rules fire client-side with zero latency. The warning appears as you type — no network roundtrip.

Auditable source

Extension source is readable. No obfuscation. CustomEvent-only page communication — no postMessage XSS surface.

Already paying a vendor that doesn't deliver?

Every subscription without AVA is a bet on what the marketing page said.

Install in thirty seconds. Sign in free. AVA checks vendor reality on Runway, Higgsfield, Krea, Pollo, Pika, Sora, Luma, and Vidu — and scores every prompt you write after.

No credit card on the free tier · Chrome · Brave · Edge · Arc