Costs and the estimate

Every generation gets a risk band, every line of the estimate two figures, and nothing spends before its stage's estimate is agreed.

Video and stills are priced in Higgsfield credits (cr). Suno, ElevenLabs and Google are priced in their own units, named on every line.

Risk bands. Give each generation in your shot plan a band, low or high, with its reason ("high: readable face, hands on the jar"). Many attempts per kept shot are normal [proven]. The band figures are this Guide's default (decided), set from one studio's published attempts per kept shot (see Sources). The ceiling figure comes from creators' reports: adjust it once your own take log shows your real numbers.

The credits for a shot are the generation's seconds × the band rate:

Seedance 2.5 costs 3 cr a second at 480p and 16 at 1080p (read Oct 2026; see Video with Seedance → Platform facts and settings).

• Each draft is one attempt, and so is each direct 1080p take. A 1080p finish counts as part of the draft it came from, not as a new attempt. If a finish loses its draft's motion, the shot switches to direct 1080p takes inside the same budget.

• A calibration record [official guidance]: Higgsfield's own team reports about 8,500 cr for a three-story ad of about 2 minutes, unused takes included. Treat it as a reference point, not a target.

• The ceiling is a stop line, not a spend target.

• Batches to expect: when each shot gets one new take in each production batch, a shot needs about as many batches as its expected drafts: about 5 for a low-risk shot, about 10 for a high-risk one. Later batches are small. Every batch adds review time, so the estimate states the count and the schedule leaves room for it.

A worked 30 s film. 49 s of generations in six clips (7 s low, 12 s high, 8 s high, 7 s low, 9 s high, 6 s low) is 2,418 cr expected and 4,807 at the ceiling. At about $0.05 a Higgsfield credit (read Aug–Sep 2026), that is roughly $121 / $240 in US dollars. Building the packshot instead of generating its 6 s low-risk clip: 2,232 / 4,435 cr. With storyboard frames, anchors and take review (about 200 cr) added to the generated film's 2,418 / 4,807: 2,618 / 5,007 cr, roughly $131 / $250. The dollars are for context only; the estimate keeps credits and money apart.

The same film, all in. Each line keeps its own currency, and none of them is added to another. Every figure comes from its line's home; where the home gives a rate rather than a total, the rate is shown, and "—" means the home gives no separate ceiling.

Stills. Expect 2 attempts per image, 3 with a face, plus at most two edit passes, so the ceiling is 4 generations for an image and 5 with a face. Storyboard frames go to the treatment's frame budget, anchors to the anchors budget, and internal frames to their shot's video stage.

GPT Image's version: a new project uses the latest version on the platform (2.6 since 5 Nov 2026), and a project stays on the version its first approved GPT Image asset was made on.

What the estimate holds, expected and ceiling on every line:

• video: each generation's seconds × its band rate, the key reactions' pairs and a feed edit's native 9:16 openings included; the sample batch's second draft per sample shot;

• speaking shots: every Arabic line on Route 2, the default (a speaking plate made with the sound off, followed by lip sync to the exact approved recording). For each close-up hero Arabic line, add a second price line for Route 1 at the line's own plate length (more if draft mode refuses audio references), kept outside the totals until you choose it for that line (see Speaking shots and lip sync → Route 1, for close-up hero lines);

• anchors: sheets per wardrobe look and state; master, reverse and coverage plates; product views and texture anchors; a style anchor where needed;

• the place motion checks, about 15–27 cr per candidate place (see Photoreal anchors (Nano Banana Pro) → The place motion check). The face motion tests are not in this estimate: they run, and are approved, before the treatment is sent;

• internal frames, only where a method names one; take review, if the analysis is charged;

• voice, music and effects, each in its own currency.

Never add the currencies together. Higgsfield credits, Suno credits, ElevenLabs credits and US dollars are four separate totals. Voice runs on your own ElevenLabs account: price it in US dollars on the API, or in ElevenLabs credits in the app, and name the unit on the line. Every line gives its unit price with its source and date.

Price before you spend. In the Higgsfield app, the Generate button shows the job's credit cost for the settings you have chosen, before anything runs (as seen in creators' screen recordings, November 2026; the layout may move).

Read it once per distinct setting (model, mode, resolution, draft or not, and duration), and write it with the date in your price ledger. The figure covers every output the job makes: set two outputs and it doubles. Or price from the prices already in your ledger, and say in the estimate which you used. For Seedance, the number of references, audio and aspect don't change the price. (Higgsfield's interface for programs also takes a cost-only request, which returns the price and creates no job. It works only on single requests, not batches.)

Go-ahead per stage, on both figures (the expected figure and the ceiling):

1. The treatment's frame budget: when the treatment starts.

2. The face motion tests (about 95–171 cr per face, judged on a 1080p finish; see Writing the treatment → Face motion tests) and the place motion checks: before they run.

3. The anchors budget: with the plan.

4. The video budget: only after the client approves the anchors presentation.

5. Audio and finishing items: each in its own currency.

The balance: check it before each stage's first job and before any job over about 200 cr, and reconcile spend after each batch. After a timeout, check the job list before resubmitting: a blind resubmit can charge twice. A generation that fails and returns nothing is refunded; a completed result that misses the intent is charged.

Stage {n}, {name}: estimate, {date}
Prices: Seedance 2.5 480p draft 3 cr/s, 1080p 16 cr/s ({source}, {date}); {other tools}.
| Item | Tool and setting | Qty | Unit price | Expected | Ceiling |
| {shot}, {s} s, {band} | Seedance 2.5, drafts then 1080p | {s} s | {31 / 62 or 62 / 123} cr/s | {cr} | {cr} |
| Higgsfield credits, total | | | | {cr} | {cr} |
| US dollars, total (ElevenLabs API, Gemini) | | | | ${x} | ${y} |
| ElevenLabs credits, total (app only) | | | | {x} | {y} |
| Suno credits, total | | | | {x} | {y} |
Balance now: {cr}. After this stage at the ceiling: {cr}.
Batches to expect: the sample batch ({n} shots, two drafts each), then about 5 production batches; the high-risk shots may run to about 10.

Sources

• PJ Accetturo on the Kalshi spot (newsletter) and on Hell Grind (newsletter); Paul Trillo (article); The Gardening Club (notes) (attempts per kept shot, the studio behind the band figures).