The stock subscription is the cheap part: the real bill is 90 to 150 searches per video
Twelve minutes is 720 seconds, and at a visual change every 5 to 8 seconds that video needs between 90 and 150 pictures. Every one of them has to come from somewhere. A stock library costs $20 to $80 a month, which spread over 8 videos is $2.50 to $10 per video, and that is the number everybody budgets for. The number nobody budgets for is that filling those slots means running 90 to 150 separate searches.
Time it once and the illusion dies. Curating a full 12 minute episode by hand takes 40 to 90 minutes, which is 20 to 45 seconds per scene and already assumes you accept the second or third result without scrubbing it. Price your own hour at $25 and the curation costs $17 to $37 per video against $2.50 to $10 of subscription. The licence fee was never the expense. The search was, and it is most of what a youtube video editor is actually paid for.
The search also does not always end well. On concrete subjects, a kitchen, a highway, a laptop, the hit rate is high. On abstract ones, inflation, a rivalry, a decision, a specific historical moment, the share of clips that genuinely fit drops to 50% to 70%. In a 120 scene video that is 36 to 60 slots where you settle for something that only almost illustrates the sentence. FalconVid closes the gap from the other side: the pipeline matches the media to the sentence the narration is about to say, so nobody is searching at 1am and nobody settles at scene 90.
- 12 minutes at a cut every 5 to 8 seconds means 90 to 150 pictures to source.
- Subscription: $20 to $80 a month, roughly $2.50 to $10 per video at 8 videos.
- Curation by hand: 40 to 90 minutes per video, which is $17 to $37 of your time.
- Abstract subjects: only 50% to 70% of searches return a clip that truly fits.
The generic clip is the number one abandon trigger between minute 2 and minute 4
The drone over the city at sunset. The hand typing on the laptop. The two people in suits shaking hands in a glass lobby. You have seen them, and your viewer has seen them in forty other videos this month. A clip that reads as footage instead of as the thing being described carries no new information, and a screen with no new information is permission to leave.
Read it on the curve rather than in theory. Long form videos rarely break in the hook when the hook is good, they break in the flat stretch between minute 2 and minute 4, where the narration is still delivering and the pictures have started repeating. As a working benchmark from our own production, not an official YouTube figure, a narrated 12 minute video holds 45% to 60% at minute 2 and 35% to 50% at minute 4, and the size of that gap tracks how many consecutive scenes said nothing.
The mechanism is boring and easy to test: find three consecutive scenes that could belong to any other video on any other subject, and you have found the drop. Note the direction of the effect, because it matters for what you buy. Stock is not the problem. Repeated generic stock is. A specific picture of the specific thing works whether a camera or a model produced it, which is exactly why this decision has to be made per scene and not per channel.
- Generic footage carries no new information, and no new information invites the exit.
- The break in long form sits between minute 2 and minute 4, not in the hook.
- Working benchmark: 45% to 60% held at minute 2, 35% to 50% at minute 4.
- Three consecutive scenes that fit any video is where the curve bends.
The scene by scene rule: three doors, and only one of them is generation
Stock wins on general context and texture. Nature, weather, crowds, an office, an airport, a slow macro shot behind a definition. These scenes carry mood rather than fact, and mood is precisely what a library holds four million takes of. Roughly 40% to 55% of a 12 minute episode is this kind of scene, and paying a generator to reinvent a wave breaking on rocks is money set on fire.
Generation wins on the specific thing the narration just said and that no library ever filmed: the 1954 factory floor with that exact machine, the product that does not exist yet on the shelf, the metaphor built for that one sentence, the historical scene nobody was there to shoot. Roughly 25% to 40% of the scenes in an explainer or a documentary land here, and they are the ones that hold the flat middle of the video.
The third door beats both when the subject is real and identifiable: a public figure, a named product, a recognizable landmark, a news event. There a generated approximation reads as fake and a generic clip reads as evasive, and what the scene wants is the real photo of the real entity. FalconVid runs all three doors in the same pass: a media bank with real entity photos for people, products and places, generated scenes for what does not exist, library footage plus a music and sound effects bank with BGM, SFX and Freesound for everything else.
- Library media: context, nature, cities and texture, about 40% to 55% of the video.
- Generated scene: the exact thing the sentence names, about 25% to 40%.
- Real entity photo: people, products, landmarks and news, where fakery is obvious.
- The choice belongs to the scene, never to the channel or to the whole video.
What a generated scene actually costs, and why the same video has three prices
Comparison articles skip this part because generation is priced per second and stock is priced per month, and the two never meet until you put them on the same episode. In FalconVid the arithmetic is already done for the whole video: the same 12 minute video costs 1,008 credits in economy mode, 8,676 in balanced and 26,760 in premium. At the Starter credit value that is about $3.16, $27.18 and $83.84 for identical runtime, identical script and identical narration.
The 133 times spread between economy and premium is not a support tier and not a feature gate. It is literally how much of the video is generated, and by which engine. Economy leans on the media bank and on images, so most scenes are library footage, real entity photos and stills with slow movement. Balanced starts generating scenes. Premium brings in the expensive engines with motion and audio, Veo 3 with sound and Seedance Pro, on the scenes that carry the episode.
Spread across 120 scenes the per scene price becomes readable: about 2 credits in economy, roughly half a cent, 69 credits in balanced, about $0.22, and 224 credits in premium, about $0.70. Now the subscription is finally comparable. A stock library sells you the right to search 120 times; an entire economy episode of 120 scenes costs under a dollar of generation and nobody searches at all. When you create video with ai, the price is never the runtime, it is how many scenes are generated and by which model.
- Same 12 minute video: 1,008 credits economy, 8,676 balanced, 26,760 premium.
- In dollars at the Starter credit value: about $0.63, $26.06 and $84.22.
- Per scene across 120 scenes: about 2, 69 and 224 credits.
- The 133x spread is generation and engine choice, not features or support.
What FalconVid does here: three media sources, one pass, and the Studio for the scene that missed
The pipeline decides per scene instead of per video. It writes the script, splits it into scenes on the sentence rather than on a timer, and for each sentence picks the source: library footage, a real entity photo from the media bank, or a generated scene, with the quality mode you chose deciding how far the generation goes and which engine runs it. The music and sound effects bank with BGM, SFX and Freesound sits inside the same pass, so no scene waits on a second subscription or a licence hunt.
It runs in parallel. A researcher, a scriptwriter, a narrator, an editor and a sound designer work at the same time instead of standing in a queue, which is why the finished video lands in up to 30 minutes rather than in an evening, with simultaneous generations set by plan: 2 on Starter, 5 on Pro, 10 on Business, 25 on Agency and 50 on Scale. The only thing you approve is the calendar.
And the obvious objection has a boring answer. If scene 63 came back wrong, you open the first version in the Studio, swap that one media for another take, change the track, shorten the intro, or open the timeline and do it yourself. You are correcting one scene out of 120, not rebuilding a video. That is the whole difference between reviewing ai generated videos and editing them, and it is why an ai that edits your video is only useful when the media choice was already right on scene 62 and scene 64.
- Media picked per sentence: library, real entity photo or generated scene.
- Music and sound effects bank included, no second subscription and no licence chase.
- Parallel production, finished video in up to 30 minutes, 2 simultaneous generations on Starter to 50 on Scale.
- The Studio swaps a single scene, or you open the timeline yourself.
The mix that works on a 12 minute video, with the arithmetic of the bill
The recipe that holds on a 120 scene episode: 50 to 65 scenes of library media for context and texture, 30 to 45 generated scenes on the sentences that need the exact thing, 10 to 20 real entity photos wherever a person, product or place is named, and 5 to 10 charts, maps or data cards. Nobody notices the seams between the sources. Everybody notices when all 120 scenes came out of the same drawer.
Cost follows the mix, which is why in FalconVid the quality mode is a dial per project and not a plan upgrade. A channel publishing 7 videos a month can run 6 in economy at 1,008 credits and 1 in balanced at 8,676, which is 14,724 credits and fits inside a Starter month. Starter is $47 with 15,000 credits, a realistic 10 to 12 videos a month mixing economy with one in balanced, or 14 in pure economy, which is about 3 hours of video.
Most channels never touch the lever, and it is the cheapest upgrade available. Same channel, same calendar: the routine weekly upload goes economy, the episode you expect to travel goes balanced, the one that carries the quarter goes premium. Top up packs that never expire cover the spikes, Boost 2,000 credits for $9, Power 6,000 for $24 and Mega 14,000 for $49, so one premium episode never forces a plan change.
- 120 scene recipe: 50 to 65 library, 30 to 45 generated, 10 to 20 real entity photos, 5 to 10 cards.
- 6 economy videos plus 1 balanced is 9,530 credits, inside a Starter month.
- Starter $47 with 15,000 credits: about 10 to 12 videos a month mixing modes, or 14 in pure economy.
- Packs that never expire: Boost 2,000 for $9, Power 6,000 for $24, Mega 14,000 for $49.
The 2026 verdict: neither library only nor generated only, and the ceiling belongs to the manual route
Forced into one answer: mixed, decided scene by scene. Library media for the 40% to 55% that is context, generation for the 25% to 40% that is the specific thing the sentence names, real photos for identifiable entities. A channel that is 100% stock reads as recycled by minute 3. A channel that is 100% generated pays up to 133 times more than it needs to, and risks drift on scenes a $20 library already solved perfectly.
Yes, generating all 120 scenes with the premium engines is an $84 episode, and that is worth doing perhaps once a month on the video meant to travel. That is exactly why the three modes exist, why nobody sane runs a whole month in one of them, and why the mode is chosen per video rather than per channel. The engine choice is the honest answer to a cost question, not a paywall.
The real ceiling was never the money, it was 40 to 90 minutes of curation multiplied by every video you publish. One person can hand pick 120 scenes carefully for one video a week, maybe two, before the search becomes the job and the drone shot starts repeating. That ceiling belongs to the manual route, and it disappears the moment the FalconVid pipeline matches media to sentences in parallel and delivers the finished video in up to 30 minutes. Starter runs 1 channel with 2 simultaneous generations, Scale runs 50 channels with 50 simultaneous generations, and a faceless channel keeps the same care at four videos a week that it had at one.
- The answer is mixed: library for context, generation for the specific, photos for real entities.
- 100% stock reads as recycled, 100% generated pays up to 133 times more than needed.
- Premium at 26,760 credits, about $84, is a once a month decision, not a monthly setting.
- The 40 to 90 minutes of curation per video is the manual ceiling, and it is the part that disappears.