100 ideas to 5 finals: a funnel budget on Sume, stills then 1080p

You can test 100 video ideas and finish the best 5 at 1080p for $20.75 on Sume, against $212.00 for making all 100 as finished clips. The funnel is 100 low-tier Ideogram 4.5 stills at $0.0375, 20 promoted to 5-second Wan 3.0 clips at 480p at $0.32, 5 finals as 8-second Wan 3.0 clips at 1080p at $2.00, plus a voice line and a one-minute render for each final (read 2026-10-06).

The 100, 20 and 5 are a plan, not a measurement. They are the attrition you decide to allow, and you can change them. The point is that each stage is priced by its own unit, and only the last stage costs real money, so most ideas die when they are cheap.

Why stills first

A still is the cheapest way to look at an idea. At $0.0375 you can look at 26 stills for a dollar, and a bad composition shows up at that price rather than at $2.00 for a 1080p clip. The still also becomes the first frame of the clip when you promote it, since Wan 3.0 supports image-to-video, so the early money is not wasted.

The mix is adjustable. If your ideas are strong, promote 30 and finish 8: that is $3.75 plus $9.60 plus $16.00 plus $0.16 plus $0.80, or $30.31. If they are weak, cut the first stage to 50 ideas and the whole funnel costs $18.88 for the same 5 finals.

• Set the funnel counts before you spend, and write down the rule that promotes a still.
• Never promote more than your finish budget allows: 20 drafts at $0.32 are $6.40, a small number, but 20 finals at $2.00 are $40.00.
• Keep every still and draft with its prompt, so a later brief can reuse the winners.
• Failed jobs are refunded, and an idea you reject costs nothing beyond the stages it already passed.

A funnel calculator

The function below prices a funnel from stage counts and the billed prices above. Change the counts to model a stricter or a looser funnel.

def funnel(stills: int, drafts: int, finals: int) -> float:
    still = 0.0375
    draft = 0.32   # Wan 3.0, 480p, 5 seconds
    final = 2.00   # Wan 3.0, 1080p, 8 seconds
    voice, render = 0.02, 0.10
    total = stills * still + drafts * draft
    total += finals * (final + voice + render)
    return round(total, 2)


print(funnel(100, 20, 5))
print(funnel(100, 30, 8))
print(funnel(50, 20, 5))
print("all finals:", funnel(0, 0, 100))

The cost of skipping the funnel

Making all 100 ideas as finals costs 100 clips at $2.00, 100 voice lines and 100 renders, $212.00. That is ten times the funnel for the same 5 ideas that would have shipped. The funnel's saving is large only because most ideas are not worth a final, so measure how many of yours are before you decide the shape.

Check the live model limits in the Video Router catalog, the card tiers in the images docs and the render rate in the Timeline docs.

Choosing the promotion rule

The funnel only works if the promotion rule is boring and written before you look at the stills. A rule such as 'the subject is readable at thumbnail size and the composition has room for text' can be applied by anyone in seconds and does not drift as you tire. Stricter rules shrink the second stage, and a clip draft at 480p is cheap enough that you can afford a looser rule early and a tight one at the final.

Keep the draft stage honest too. A 480p draft shows you motion and framing, but it does not show you 1080p detail, so do not reject an idea for softness that the final will fix. Reject it for the thing the draft can show: pacing, motion that fights the subject, or a first frame that does not hold.