Storyboard with stills first: how many images cost one AI clip
One 15-second Seedance 2.5 clip at 720p costs as much as 1,733 higgsfield-soul or 231 flux-2-pro stills. A cost-ratio table for screening with images.

A common way to control video spend is to look at stills first: generate a handful of frames for each idea, drop the weak ones, and only then spend on motion. This post is about the cost ratio only. It does not claim a still predicts how a clip will turn out, because motion adds many things an image cannot show.
Prices are Sume billed rates read on 2026-10-03.
How many stills equal one clip?
Each cell divides a clip's price by a still's price and rounds down. A five-second Wan 3.0 clip at 480p ($0.3125) costs about as much as eight flux-2-pro stills ($0.0375 each).
| Still model | Price per still | vs Wan 3.0 480p, 5 s | vs Wan 3.0 720p, 10 s | vs Seedance 2.5 720p, 15 s, 9:16 |
|---|---|---|---|---|
| higgsfield-soul | $0.005 | 62 | 250 | 1733 |
| gpt-image-2.5 low | $0.0075 | 41 | 166 | 1155 |
| gpt-image-2.5 medium | $0.0165 | 18 | 75 | 525 |
| flux-2-pro | $0.0375 | 8 | 33 | 231 |
| seedream-5-lite | $0.0437 | 7 | 28 | 198 |
| nano-banana-2 | $0.10 | 3 | 12 | 86 |
When does screening with stills pay off?
It pays off when the clip is expensive and the question is compositional: product placement, framing, color, a pose. A 15-second Seedance 2.5 clip at 720p is $8.667. Screening ten ideas with flux-2-pro stills costs $0.375, and the saving is real if it removes even one rejected clip.
It does not pay off when the question is about motion, pacing or audio. For those, a short, low-resolution clip on Wan 3.0 is the cheaper screen: $0.3125 for five seconds at 480p.
A suggested order
Start with the cheapest stills to explore, move to a mid-priced still model once a direction survives, and pick a clip model last. Then render the clip at 480p, and the final at delivery resolution. If your final model supports a first frame, a still you approved can seed it, as in the Halloween first-frame post.
The numbers here move whenever a rate card changes, so reread the image price list before you plan a large run.
Sources
Related posts
More in Use cases
- Strava video longer than 30 seconds: cut the part you want
Strava crops activity videos to the first 30 seconds and shows 6 on mobile, 12 on web. Pick your 30-second moment with Sume video trim, $0.02 per cut.
- Swap a person in a 4-second video: Recast needs 5 seconds
A 4-second clip is below h3-max-recast's 5-second minimum. Genjutsu takes 4 to 30 s, Gemini Omni edit 3 to 10 s, and Avatar Face Swap about 4 to 15 s.
- Swap multiple characters in a video with AI: 2 to 4 people
Swap several people in one clip with H3 Max Recast on Sume: one photo per person, left to right by default, up to four, and a price that ignores headcount.
- Swap the presenter in a product demo for each market with Recast
Plan one source demo and several Recast jobs, one per presenter photo: cost for five variants, idempotency keys, and the voice caveat, using documented limits.
Written by Sume