“How much does it cost to make an AI short drama” is hard to answer with a single number in 2026, because market quotes have split apart. Low-end assembly-line short dramas have fallen to a few hundred RMB per minute, while premium custom productions can still quote RMB 20,000–30,000 per minute—a gap of dozens of times. In 2024, one minute of AI comic drama cost around RMB 15,000. In 2025, it dropped to RMB 100,000 for a full title, or about RMB 1,000 per minute. By 2026, batch-produced titles were down to RMB 400 per minute, with a 100-minute work costing under RMB 40,000. All these numbers are real, but they are not describing the same thing.
The easiest mistake beginners make is confusing a quote with a cost. A RMB 500-per-minute lightweight AI short drama is not just compute. It also includes script adaptation, voice and music, rough editing, and basic image repair labor. If you ignore repeated generation losses, the nominal compute cost per minute may be only a few dozen RMB. Conversely, premium productions cost more not simply because they use more compute, but because a full team of artists, directors, musicians, and editors is tuning assets, optimizing expressions, and pre-visualizing camera movement.
The real budget killer is failed generations. One production postmortem showed that a finished work used 95 storyboard images selected from 1,700 generated images—a ratio of roughly 18 to 1. Generating the same face dozens of times is not unusual, and the compute bill for rejected outputs still has to be paid. The actual ledger of an individual creator is even more direct: compute plus failed outputs at about RMB 1 per second, editing at about RMB 0.7 per second, and a maximum output of five minutes per day. A daily five-minute update therefore costs around RMB 510 per day in compute and editing. So the answer to “how much does it cost to make an AI short drama” depends heavily on how much of your compute budget your workflow spends on rejected outputs.
Reducing that loss is not about finding a cheaper model. It is about reducing the number of times you reintroduce the character. Every new shot that re-describes the character from text gives the model a chance to produce a new face, which raises the failure rate. ElserStudio locks characters, scenes, and props in the World asset layer, and later shots reference those assets instead of regenerating descriptions. That structurally reduces repeated generation caused by drift. On cost transparency, the tool does not mark up compute: users can pay the model provider’s actual price with their own API key, and local project data stays on their own device.
A practical starting point for beginners: do not calculate “how much for a full series” first. Make one episode as a sample, record the actual generated seconds and failure rate from storyboard to final cut, and multiply by the number of episodes using that real coefficient. The answer to how much it costs to make an AI short drama is not in an industry quote sheet. It is in your own episode-one ledger.