BountEx Motion
Subscribe
Showcase

Two kinds of video, made the same way.

Films are story-led: an interview, a brief, a production prompt, a storyboard, a build, a critic loop. Reels are music-led: one visual idea, cuts on the beat, kinetic type, approved at a review gate before anything is built.

The examples are being uploaded now. Every slot below is an explicit placeholder until it lands — nothing here is presented as work we did not do.

Films · Film Kit

Films: 0 of 2 published.

Story-led product and brand films. An interview, a creative brief, a production prompt, a storyboard, then a build and a critic loop.

All 2 slots are placeholders. Nothing here is presented as finished work.

  • Example film — uploading shortly

    PLACEHOLDER 16:9Duration 0:30

    Replace this: product launch film

    Placeholder — a product launch film will go here

    Placeholder entry. Replace with a real film and delete this sentence.

    Model and token count recorded per film as we publish them.

  • Example film — uploading shortly

    PLACEHOLDER 9:16Duration 0:15

    Replace this: vertical social cutdown

    Placeholder — a 9:16 social cutdown will go here

    Placeholder entry. Replace with a real film and delete this sentence.

    Model and token count recorded per film as we publish them.

Reels · Reel Kit

Reels: 0 of 2 published.

Short, music-led motion pieces. One visual idea, cuts on the beat, springy overshoot, kinetic type — approved at a review gate before anything is built.

All 2 slots are placeholders. Nothing here is presented as finished work.

  • Example film — uploading shortly

    PLACEHOLDER 16:9Duration 0:15

    Replace this: kinetic-type reel

    Placeholder — a graphic-mode reel will go here

    Placeholder entry. Replace with a real reel and delete this sentence.

    Model and token count recorded per film as we publish them.

  • Example film — uploading shortly

    PLACEHOLDER 9:16Duration 0:15

    Replace this: UI-pop reel

    Placeholder — a ui-mode reel will go here

    Placeholder entry. Replace with a real reel and delete this sentence.

    Model and token count recorded per film as we publish them.

The honest number

What a film costs to make.

A film is one session of a coding agent reading a brief, writing compositions, and rendering them on your machine. The only metered part is the model, and it is billed by the token.

Between $0.084 and $5.00 a film

Cheapest paid model: DeepSeek V4 Flash at $0.084. Dearest: GPT 5.5 at $5.00. Both are for the token profile below.

Tokens per film
500,000
An average, measured on our own builds — not a guarantee for yours.
Input / output
80% / 20%
An assumption about the shape of a build, not a measurement.
Cache discount claimed
0%
Every figure here is an upper bound; a cached session is cheaper.

9 free models, grouped by what they do with your input

Free

Free is free because the provider gets something too, and the price list says what. All 9 entries come from the OpenCode Zen price list, where they are free for a limited time or as trial endpoints — and free and safe-for-this-input are different questions, so they are grouped by what each provider does with what you send. The wording in each group is that provider's own term, taken from the list rather than summarised by us.

Free, and retains nothing

2 models

Retains nothing and does not train on what you send.

Choose from this group when the input is a client's unreleased material.

Free, and may train on what you send

5 models

Your prompts and completions may be used to improve the model.

For your own product, and for material you are happy to publish.

Free trial endpoints that log your session

2 models

Trial endpoint: your session is logged. Do not send confidential material.

The practical rule: use a free model on your own product and on material you are happy to publish; on a client's confidential material, use a paid model, or one of the 2 free models above whose terms say it retains nothing.

Free models are listed separately and above the paid table, because zero in a cost column reads as a cheap tier rather than a different arrangement.

Paid models, cheapest first

All 13 paid models in the pricing module, at 500,000 tokens a film. Prices read 1 October 2026; the bar is a log scale so the cheap end stays readable, and the number beside it is the exact figure.

Cost of one film per paid model, cheapest first, with a link to each model's published price list
Model Vendor Cost per film Relative cost
DeepSeek V4 Flash cheapest paid DeepSeek via ZenOpenCode Zen price list (opens in a new tab)$0.084
GPT 6 LunaRates shown are the ≤ 272K-token tier.OpenAI via ZenOpenCode Zen price list (opens in a new tab)$0.090
Qwen3.8 FlashAlibaba via ZenOpenCode Zen price list (opens in a new tab)$0.11
GLM 5.3 FlashZ.ai via ZenOpenCode Zen price list (opens in a new tab)$0.11
DeepSeek V4.1 Flash (direct)Off-peak is half of peak. Peak is 01:00–04:00 and 06:00–10:00 UTC, Monday–Friday.DeepSeek APIDeepSeek API pricing, off-peak rates (opens in a new tab)$0.12
DeepSeek V4.1 FlashThe model this site is built and verified with.OpenCode ZenOpenCode Zen price list (opens in a new tab)$0.24
MiniMax M3MiniMax via ZenOpenCode Zen price list (opens in a new tab)$0.24
Gemini 3 FlashGoogle via ZenOpenCode Zen price list (opens in a new tab)$0.50
Gemini 3.5 FlashGoogle via ZenOpenCode Zen price list (opens in a new tab)$1.50
Claude Sonnet 5Anthropic via ZenOpenCode Zen price list (opens in a new tab)$1.80
Kimi K3Moonshot via ZenOpenCode Zen price list (opens in a new tab)$2.70
Claude Opus 5Anthropic via ZenOpenCode Zen price list (opens in a new tab)$4.50
GPT 5.5Rates shown are the ≤ 272K-token tier.OpenAI via ZenOpenCode Zen price list (opens in a new tab)$5.00

Bar length uses a logarithmic scale, normalised so the dearest model is the full width. Hover a bar for the exact multiple it represents.

Why the token count is as low as it is

500,000 tokens is a claim about a workflow, not about a model being clever. It is the result of four decisions in how the build runs — all of them engineering, none of them a discount.

We have not measured a percentage saving from any of this, so we do not claim one. It is simply how the workflow is built, and the figure above is what it produces on our own builds.

  1. One phase at a time

    The kit installed on your machine is the shape of the workflow, not the whole method. The substance behind each phase is fetched from the license server when that phase starts, so what sits in the context window is the phase you are in — not the entire library.

  2. A render spike before the build

    One representative composition is built and rendered first, and determinism is checked on the stills, before the full film is built. An approach that cannot render is caught while it is still a test, instead of after it has consumed a whole session.

  3. Artifacts on disk, not in the conversation

    The brief, the production prompt, and the storyboard are written to files and read back from disk when they are needed. The files are the source of truth, so they are not carried in the conversation and re-sent on every turn.

  4. A critic that works from evidence

    Each critic pass inspects specific frames and specific measurements and reports FIXED, PARTIALLY FIXED, or STILL PRESENT. It does not ask the builder to re-describe the whole film, so the review reads a little and changes a lot.

What these figures do not claim

  1. aThe 80%/20% input-to-output split is an assumption, stated here rather than buried. It is not measured per film, and a build that writes more than it reads costs more than the table above.
  2. bNo prompt-cache discount is claimed: the profile assumes a 0% cache-hit rate, while every paid model listed publishes a cheaper cached-input rate. Every figure above is an upper bound, and a real session that hits the cache is usually cheaper.
  3. cPrices are read from each provider's published price list on 1 October 2026, and providers change prices. Each row links to the list its number came from so you can check it against the current one.
  4. dRendering happens on your own machine with your own Chrome and FFmpeg, and is not included above because it is free. The only metered part of a film is the model.

A video agency quotes thousands of dollars for a single film. At these rates the model spend is not the cost — your time is.

The next one on this page can be yours.

One license key installs both workflows into the AI coding tool you already use. You keep the films and the source; the model spend stays in the cents.

See pricingRead the docs