The 70/30 rule: spend on winners, test the rest

The 70/30 rule is a way to split your AI video budget once you have a winner. Spend about 70% of your credits on remaking the format that won, on the video model that made it. Spend the other 30% on tests: new video models and new formats. Move the split anywhere from 60/40 to 80/20 as your account changes, and run every test across at least three reels before you judge it.

Picking a video model and a format is trial and error. No model wins every shot, and no format works for every account. Without a rule, most people do one of two things: they keep testing and never cash in on what works, or they ride one winner until it stops working and have nothing ready to replace it. A fixed split does both at once. This guide gives the split, what counts as a remake and what counts as a test, and one month worked out in Hook Studio credits.

The split is of what you spend on making reels in a month: stills, clips, and the retakes you need. Set it at the start of the month, not reel by reel, so a good week does not eat the test money and a bad week does not eat the remakes.

Before your first winner

The rule starts once something has won. Until then, the whole budget is for tests. Make a few different formats, each across three reels, and stop when one beats your usual numbers on more than one reel. For how to read those numbers and spot a winner, see how to find winning formats.

• Toward 80/20 when you have more than one winner and their remakes keep beating your usual numbers. More of what works pays more than another experiment.
• Toward 60/40 when the remakes of your winner get fewer views each week, so the format is fading and you need its replacement soon.
• Toward 60/40 when a new video model comes out. In 2026 alone, Seedance 2.5 came out on July 31, Kling 4 on October 15, and Google's Gemini Omni 1.2 Flash on November 5.
• Toward 60/40 when you change niche, language, or audience. Your old winners were picked by people who may not be watching anymore.

Where 70/30 comes from

Ad teams split paid budgets the same way. Meta recommends dedicating "no more than 20% of your budget" to its creative testing tool, as Jon Loomer reported. A Meta ads course from iMentiX calls "roughly 70/30" a common working dial, and says "mature accounts with a deep bank of winners might sit 80/20" (iMentiX). TikTok's own ad testing guide says to "treat ad testing as ongoing" and to "change only one element per test."

Those are rules for ad spend. The 70/30 rule on this page is our formula for organic AI video: the money goes into making reels, not into buying views, and the tests are about which model and which format to make them with.

A winner is a format, not a video. The format is the part you can repeat: the kind of hook, the shot plan, the length, the person on camera or none, the text style, and the video model that made the clip. A remake keeps all of that and changes one thing:

• A new hook. The same format with a new first line.
• A new person. Another face from your cast, in the same shot plan.
• A new setting. The café becomes the car or the kitchen, with the same action and framing.
• A new product or topic. The same format sells something else or answers another question.

Keep the model that won. If the winner was a talking clip on Kling 3, its remakes are talking clips on Kling 3. Another model moves, speaks, and acts differently from the same still, so a model change is a test, and it comes out of the 30%.

In Hook Studio, save the winning format as an idea on the idea board, keep the person as a saved avatar, and make each remake in its own session on the same video model. For more ways to turn one good post into new ones, see remix your winners.

Every test changes one thing and keeps the rest, so you know what made the difference. There are two kinds:

• A new video model. Run your winning format on the new model: the same kind of hook, the same shot plan. For the first reel, use the same still and the same line as a reel that already won, so the model is the only change.
• A new format. A skit, a point-of-view reel, a faceless product reel, a talking hook in a new setting. Make it on the model you already trust, so the format is the only change.

For the steps of making each reel, from the still to the finished clip, see how to make AI videos.

One reel is one sample. Its views depend on the hook, the hour, the topic, and who happened to see it first. A good format can flop once, and a lucky post can make a weak one look good. So run each test on at least three reels, with three different hooks, before you decide:

• Two or three of the three beat your usual numbers: it is a winner. It moves into the 70% next month.
• One of the three: run three more before you decide.
• None: drop it, and spend the next test on something else.

On Instagram, a trial reel is a cheap place to run a test: Instagram says "your followers won't see the trial reel in their feed or Reels tab," so a test that flops does not reach your followers' feed. See Instagram trial reels for who can use them.

This is one month on Hook Studio's Growth plan: $50 a month for 5,800 credits. The winning format is a talking hook: two stills and one 5 second talking clip on Kling 3 per reel, with text and an end card. The prices are Hook Studio's on November 14, 2026: a still is 11 credits, and a 5 second clip is 143 on Kling 3 with speech, 96 on Kling 3 silent, 197 on Seedance 2.5 at 720p, and 68 on Seedance 2 Mini at 720p.

The other 183 credits stay free for the realism checks and the renders, which Hook Studio bills on top, at a few credits each. That is 29 reels in the month: 20 remakes and three tests of three reels each.

A test that wins changes next month's math. If the Seedance 2.5 test wins, its remakes cost 219 credits a reel instead of 165, so the same 70% buys 18 remakes instead of 24. Whether the better reel is worth six fewer posts is what your numbers decide. On the Concept plan, $20 for 2,300 credits, the same rule gives you 1,610 credits for 9 remakes and 690 credits for one three-reel test on Seedance 2.5.