Ship a vertical anime episode from a JSON file — on a machine with no GPU.
ManjuKit is the whole storyboard-to-video pipeline: FLUX stills, Piper and Edge-TTS voice, FFmpeg camera moves, burned-in subtitles, finished MP4. Unzip it, edit one JSON, render 1080×1920. Both episodes rendered on 32 CPU cores with no GPU — 35 sec in 4 minutes, 62 sec in 5 — and cost $0 in API spend.
No monthly billing · we never take a cut of your API usage
No GPU. No $2,000 rig.
32 CPU cores and 123GB of RAM is enough. No node graph to wire, no model to babysit.
One character card, every shot.
Every shot pulls from the same character card, so your lead does not get a new face between episodes. Studio adds cross-shot consistency and a per-frame check that re-rolls a frame that drifts.
Script to vertical cut in minutes.
Storyboard JSON → speech → stills → camera move → subtitles → 1080×1920 MP4. Change the script, change the JSON, render again.
Sample episode player
The sample file is not attached to this page build. The finished episode ships inside the download.
The full pipeline plus 12 storyboard templates, the install script, the assembly guide, the finished sample and the cost-per-episode table.
- Full pipeline source code — read it, fork it, run it offline
- 12 storyboard templates + the JSON spec
- Install script and assembly guide
- Two sample episodes (35 sec / 10 shots and 62 sec / 20 shots)
- Cost-per-episode table
Everything in Creator, plus the character-card consistency system, the per-frame visual gate, the 62-second project, a commercial license and 90 days of updates.
- Everything in Creator
- 62-second, 20-shot serialized project
- Character-card consistency system
- Per-frame visual gate — inconsistent frames re-roll
- Commercial license + 90 days of updates
Sample episode
Real output from the pipeline — not a mock-up.
The clip below was rendered by ManjuKit itself: twenty shots, voice, camera move and burned-in subtitles, composed into a 1080×1920 vertical MP4. The same command produced it on a 32-core CPU box with no GPU, in about five minutes, at $0 in API spend.
Sample episode player
The sample file is not attached to this page build. The finished episode ships inside the download.
What you are looking at
- 01Twenty shots, 62 seconds — a full episode beat, cut vertical, 1080×1920, 32.6 MB.
- 02Voice from the JSON — Piper runs offline, Edge-TTS covers more languages; timing is measured from the generated speech, not guessed.
- 03Camera move + hard subtitles — FFmpeg zoompan for the move, ASS for subtitles burned onto the timeline.
- 04$0 in API spend — the whole run stayed inside free tiers (Cloudflare Workers AI and the Pollinations fallback).
- 05Five minutes wall clock for twenty shots on 32 CPU cores and 123GB of RAM, no GPU involved.
Style: anime illustration. Frames in the sample are AI-generated. The dialogue in the current sample is Mandarin — voice IDs are set per character in the storyboard JSON (en-US-* for English), so the same pipeline renders other languages without code changes.
The only comparison that matters
One episode, two ways to get it.
Outsourced animation is quoted per episode and takes weeks to arrive. ManjuKit is paid for once and renders on your own machine, so the cost per episode stops being a recurring line item.
Outsource ≈ $300–2,000 per episode + 2–4 weeks
- A new quote for every episode. The price and the wait come back together, every time.
- 2–4 weeks of turnaround before you can review a cut, and another round if a shot needs to change.
- You do not own the pipeline. The character descriptions, the shot list and the motion setup live in the studio's files.
- The schedule is the studio's. If the episode is late, you are late.
Cost per episode: $300–2,000 · Lead time: 2–4 weeks · Recurring
ManjuKit = $149 once + $0 API + 4 minutes
- Paid once. $149 for Creator, $499 for Studio. No monthly billing, and we never take a cut of your API usage.
- 4 minutes to a finished episode on 32 CPU cores — change the JSON and render the next one the same afternoon.
- You own the source code. The storyboard JSON, the character cards and the pipeline are yours to fork and extend.
- The cost-per-episode table shows the real unit prices behind the $0 API run, line by line.
Paid once: $149 · API on the sample run: $0 · Render: 4 minutes
* The sample run cost $0 in API spend because it stayed inside the free tiers actually used — Cloudflare Workers AI for the FLUX stills and Edge-TTS for the voice. Those free allowances are what the numbers in the delivered cost-per-episode table are built from. If a longer series outgrows the allowance, you pay Cloudflare directly — we do not meter, mark up or take a cut.
Character consistency
One character card carried into every shot, with a per-frame identity check.
The failure that breaks most AI short-drama pipelines is the lead getting a different face in the next shot. ManjuKit pins the character to a fixed seed plus a written character card plus a reference portrait, then measures every rendered frame against that card before it ships.
Fixed seed + character card
The character lives in one place in the storyboard JSON: a description, a fixed seed and a reference portrait. Every shot pulls the same block, so nothing is re-invented per shot.
CLIP similarity check per frame
Each rendered frame is compared to the card with an image-similarity check. The check reports how close the frame is to the reference.
Frames that drift get re-rolled
A frame that falls below the threshold is automatically regenerated with a new seed and checked again. You get the episode, not a pile of near-misses.
The pipeline
Five stages, one command.
Nothing to wire up and no node graph to babysit. Change the JSON, run the pipeline, get an MP4.
Storyboard JSON
One file holds the shots, the lines, the camera moves and the character descriptions. Twelve templates ship ready to render.
storyboard.json
Voice, with real durations
Piper runs offline; Edge-TTS covers more languages. The timeline is built from the length of the generated speech, so nothing drifts out of sync.
[1/5] TTS
FLUX stills
Cloudflare Workers AI image-to-image, seeded from the character card. The sample run produced 10 out of 10 frames.
[2/5] FRAMES
Camera move + subtitles
FFmpeg zoompan pushes and pans over the stills; ASS burns the subtitles onto the timeline. The look is dynamic-comic motion, not full character animation.
[3/5] MOTION · [4/5] SUBS
Vertical MP4 out
Everything is composed into a 1080×1920 H.264 file — 19.6 MB for the 35-second episode, 32.6 MB for the 62-second one.
[5/5] COMPOSE → out/EP01.mp4
Run it with one command: python3 manju_pipeline.py. There is no GUI to learn and nothing to click through — which also means the output is reproducible, so a series looks like a series.
What you receive
Everything the sample was built with.
The download is the same pipeline, the same templates and the same documentation used to render the sample episode above — not a reduced demo.
Creator
$149 one-time- Full pipeline source code — storyboard JSON → MP4, read it and run it offline
- Storyboard JSON spec — every field documented, so you can write your own files
- 12 storyboard templates — ready to render as-is on day one
- Install script — walks through the dependencies step by step
- Assembly guide — the 8-layer, 24-block breakdown of how the pieces fit together
- Two sample episodes — 35 sec / 10 shots and 62 sec / 20 shots · 1080×1920
- Cost-per-episode table — the real unit prices behind the $0 API run
Studio
$499 one-time- Everything in Creator — the full pipeline and all 12 templates are in there too
- 62-second, 20-shot serialized project — the multi-episode format, engineered
- Character-card consistency system — fixed seed + character card + reference portrait
- Per-frame visual gate — every frame checked, inconsistent frames re-rolled automatically
- Commercial license — use the output commercially
- 90 days of updates — new templates, pipeline patches and model adapters
Both tiers are one-time downloads. No monthly billing, and we never take a cut of your API usage — if you outgrow the free tier, you pay the provider directly. Prices in USD.
Questions
Straight answers.
Will my character look like the same person in every shot?
That is the thing that breaks most AI pipelines, which is why it is the main reason to look at Studio. One description in the JSON is injected into every shot, and Studio adds the character card plus a per-frame check that re-rolls any frame that drifts. It is not perfect — nothing is — but it moves you from "a new actor every episode" to "consistent enough to publish". If consistency is your first blocker, that is the tier to buy.
Does the English voiceover work?
The pipeline runs Piper offline and Edge-TTS for multi-language voices, English included. You pick the voice per character in the JSON. It will not sound like a voice actor, but for vertical shorts with burned-in subtitles it is clean and identical across every episode — which a human voice actor is not when you are on episode 40.
I don't know how to code.
You don't need to. The setup is unzip → edit one JSON → run one command. The install script walks through the dependencies, the assembly guide shows how the layers fit together, and there are 12 templates you can render as-is without touching anything. If you can edit a text file and paste a command, you can ship an episode. If it will not run on your machine, reply to your purchase email and we will get it running with you.
Why is this cheaper than one outsourced episode?
It depends what it is compared to. One outsourced episode is $300–$2,000 and takes 2–4 weeks; ManjuKit is $149 once, and the API spend on the sample run was $0. If you are making one video, outsource. If you are making a series, the math changes quickly — the cost-per-episode table in the download does that arithmetic with real unit prices. Start with Creator if you want the pipeline and the templates; Studio is worth it when consistency is what is blocking you.
Is this another monthly AI subscription?
No. It is a one-time download with the full source code included, and there is no monthly billing. Nothing is metered on our side. The pipeline runs on free-tier APIs (Cloudflare Workers AI for the stills, Edge-TTS for the voice); if you outgrow the free allowance, you pay that provider directly.
Do I need a GPU?
No. The sample episode was rendered on 32 CPU cores and 123GB of RAM with no GPU involved. Because there is no GPU, the motion comes from FFmpeg zoompan — a camera push and pan over generated stills. That is the dynamic-comic look, not full character animation, and we would rather say so here than in a review.
Is there lip sync, sound effects or music?
Not in the sample build. You get voice, camera move and burned-in subtitles; there is no lip sync and no music bed in the shipped sample. The 62-second, 20-shot project in Studio is where the serialized format work lives, and the guide covers where a music and SFX layer attaches.
Is the $0 API spend real?
Yes, and the cost-per-episode table breaks it down: the sample run used free tiers of Cloudflare Workers AI for the FLUX stills and Edge-TTS for the voice, and both came to $0. It is not a claim that every workflow is free forever — it is what that specific 10-shot run cost, and the table shows the unit prices so you can project your own series.
Refund policy
A one-time purchase, with a window to change your mind.
Full refund within 7 days of purchase, provided the download has not been retrieved yet. ManjuKit is a digital download — once the files are on your machine we can no longer collect them back, so the window is tied to the download rather than to a form.
7 days, not downloaded — full refund
Reply to your purchase email and we will refund the full amount to the original payment method.
Already downloaded, but it will not run
Email us before giving up on it. If the pipeline will not run on your machine, we will help you get it running first — that is what the install script and the assembly guide are for, and it is where most questions land.
How refunds are processed
Refunds are issued to the original payment method and usually land within 5–7 business days, depending on your bank.
What you keep paying for
Nothing on our side. The commercial license and Studio's 90-day update window belong to the tier you bought, and they start on the purchase date.
Third-party API costs
The sample ran on free tiers and cost $0. If your own usage outgrows a free allowance, you pay that provider directly — we do not bill, meter or take a cut of it.
One download. Every episode after it.
Creator $149 for the pipeline and the templates. Studio $499 adds the character-card system, the per-frame gate, the 62-second project, a commercial license and 90 days of updates.