30 videos a month: the actual math
Credits, render time, queue depth — what it costs to scale, with no fictional creator success story.
The previous version of this post had a creator named Sarah who scaled from 50K to 200K subs in three months with a 40% watch-time boost. Sarah doesn't exist. We made her up. Here's a more useful version: the actual math of producing 30 videos a month in VidFlow.
Credits. A 3:30 finished video costs around 220 credits end-to-end on the current pricing — that's the headline number from `/pricing`. Multiply: 30 videos × 220 credits = 6,600 credits a month. The largest line items inside that 220 are image generation (KIE nano-banana-2 or gpt-image-1.5, dozens of calls per video for shot generation) and video generation (Kling 2.6, the most expensive single op). Voiceover and rendering combined are usually under 20% of the bill.
Calendar time. A 3:30 video that doesn't hit any human-review pauses runs end-to-end in about 25-45 minutes of wall-clock time, most of it spent in the Generation stage waiting on KIE webhooks. If you ship 30 a month and your queue is empty, that's well under an hour of active machine time per video — but you'll need at least one approval gate (Visual Bible, then Review) where your eyes are required. Plan on 20-40 minutes of human time per video, not zero.
Queue depth. The Phase 1.5 BullMQ worker (set `JOBS_GENERATE_SHOT_NEW=1`) is the right path for parallel shot generation. With the synchronous fallback you're serialized within a project; with the worker you can have multiple projects' shots generating concurrently. The queue isn't infinite — vendor rate limits apply, KIE in particular — but it scales much better than the synchronous path.
Where 30/month actually breaks. Three places, in order of likelihood:
1. Visual Bible approval bottleneck. Each project gates here. If you batch 30 projects to bibles-pending, you'll either approve them all in one painful afternoon or let them sit. Build a routine for it. 2. Outlier topic supply. Stage 1 surfaces outlier-scored topics from your channel's history. After a few weeks of high output, your channel has fewer fresh outliers to mine. The system needs new input — research, current events, audience feedback — not just more loops over the same history. 3. KIE webhook backlog. During heavy generation, vendor callbacks land at `/api/kie/callback`. We've yet to break the queue at our current customer count, but if you sustained 30 videos a month for several months in parallel with other heavy users, you'd start to notice latency on individual generations. Watch the worker logs.
What 30 videos a month is not. It's not 'set it and forget it.' The pipeline does the production work; it doesn't do the decision-making. Picking the right title from the slate, approving the bible, kicking back a bad shot in review, picking the thumbnail — those don't go away. They're the parts you should be spending your time on. The pipeline gives you back the editing time, not the judgment time.
The roadmap line that matters. Channel-level bibles (reuse character/location assets across a creator's projects) would significantly reduce the per-project setup work, which is the largest fixed cost of running multiple parallel projects today. It's not built yet.
We'll write a real case study when we have a real creator at 30 videos a month for long enough to learn something from. Until then, this is what we know.