Don't chase trends when replicating AI long-form dramas

Recently, a friend in content creation asked me if the $125k/episode AI long-form drama model is replicable. I didn't engage with the market cap angle. As someone who works on automotive production line automation integration, I'm used to looking at the actual takt time of the line first. Whether a shot can be delivered stably is worth more than the two characters for "limit up" in the news.

This post organizes an AI long-form drama feasibility verification checklist to help judge where a project gets stuck.

First, let's define some terms. AI long-form drama refers to continuous series generated by artificial intelligence for visuals and character performances. AIGC simply means machine-generated content. An asset table lists characters, scenes, and props item by item; think of it as a parts list for a production line.

The facts given in the news are: Journey to the West (sequel) has 30 episodes, 40 minutes each, with a cost of $125k per episode. 109 characters and 143 scenes were generated by AI.

I ran through the verification process once, comparing two approaches. Directly asking AI with a prompt like "Help me evaluate the feasibility of an AI long-form drama" yields nice-sounding words, but the source of the numbers is unclear. It's good for inspiration, not for project initiation. First, build a verification table, fill the material numbers into the spreadsheet, then let the agent answer strictly based on the table. The output will land on fields like 30 episodes, 40 minutes per episode, 109 characters, 143 scenes, and $125k per episode, allowing you to see rework, consistency, and delivery takt times.

Here's how you can set up the verification table.

Open VSCode, click File, then click Open Folder. Select an empty folder, e.g., ai-drama-check. If this folder appears in the file tree on the left, it's open. Think of VSCode as an editor for writing text and tables.

Next, click File > New File, type notes.md, and press Enter. Paste the material numbers: 30 episodes, 40 minutes per episode, $125k cost per episode, 109 characters, 143 scenes. Press Ctrl+S to save (on Mac, press Cmd+S). When the small dot in the top right corner disappears, it means the save was successful.

Then click File > New File, type assets.csv. CSV is a comma-separated values table file. Paste the header row, with fields including type,name,source,check_item,status,owner. type is the asset type (write Character or Scene); source is the data source; check_item is the question to verify; status is set to Pending Confirmation; owner is who is responsible.

Fill in two example rows. On the next line, enter role,Sun Wukong,notes.md,Is the character appearance consistent across episodes,Pending Confirmation,Director. On the following line, enter scene,Huaguo Mountain,notes.md,Is the scene lighting stable,Pending Confirmation,Art Director. After saving, seeing two rows of data confirms the format isn't broken.

If Chinese characters appear garbled, press Ctrl+Shift+P, type Save with Encoding, and select UTF-8 without BOM. I've fallen into this trap before. Previously, opening a CSV in Excel turned Chinese characters into a bunch of symbols. UTF-8 without BOM is an encoding method that correctly recognizes Chinese, English, and punctuation without adding an extra file header marker.

Click the agent icon in the sidebar and start a new conversation. Input: Answer only based on notes.md and assets.csv, do not add new numbers. Please list three delivery risks based on asset count, episode count, duration, and cost per episode. You should expect it to answer according to the table fields. I've been using agents for about a week; the most annoying thing is that they tend to invent numbers. You must lock down the sources.

Click Chat > New chat, input a project name, e.g., Journey to the West Verification. Drag assets.csv into the message box and click Send. Once you see the file card, click Open in Excel to check for misalignment. I used Teams for a while; viewing spreadsheets together is clearer than screenshotting WeChat groups.

After running through this, you'll get a simple judgment. The materials mention 109 characters and 143 scenes, totaling 252 basic assets. 30 episodes at 40 minutes each means each asset isn't done after one generation; it needs to maintain the same face, same lighting, and same movement path across episodes. $125k per episode looks low, but have you calculated the integration costs? Rework, consistency checks, manual retouching, and copyright confirmation will all eat into the takt time.

Three pitfalls to avoid. Don't ask AI to write a complete script from the start. It might casually fabricate traditional drama cost figures that aren't in the materials, which shouldn't be written into the plan. The number of commas in each CSV row must be consistent. Missing one comma causes Excel to shift columns when opening. Don't understand AI generation as requiring no human intervention. I've seen enough on production lines: robots can work, provided inputs are consistent, points are accurate, and there's a fallback for exceptions. AI long-form dramas are the same; material consistency and exception handling are key to implementation.

For the next step, pick one 40-minute episode, manually break it down into 20 shots, see how many characters and scenes correspond to each shot, and record the number of generation failures. This way, you can understand the news and know exactly where an AI content production line gets stuck.