TheFluxTrain

World models for film production

We're building a world model for film

The studio is live. Next: a model that can hold a film together, not just a clip.

0/9

Optional — reference uploads as @Image 1, @Image 2, … in your story

Try an example

Why this, now

Clip generators still cannot make a film

After generating a lot of videos through agents, we kept hitting the same wall. Current video models drift on character, struggle to stitch clips into a fluent sequence, and make it painful to edit one beat without regenerating the rest. They are useful for short shots. They are not built for long-form film production.

That is why we are building a world model for film production: a persistent model of the cinematic world that can simulate how a scene evolves under a director's instructions.

A world that remembers

Characters, locations, wardrobe, and camera state should live in the model, not get rediscovered in every clip.

The next beat, not a whole clip

Instead of hallucinating 10 or 30 seconds at once, predict the next short state from previous frames, references, and a director instruction.

Fix one part without starting over

Long-form film needs local edits. Regenerating an entire shot because one gesture is wrong is the tax current generators charge.

The studio today

Watch a brief become a finished cut

Our film platform is live. These shorts show a story or promo brief turning into scenes on a timeline.

10s demo
StoryScenesGenerate
01

Story brief to scenes

Describe the family film or promo you want — Director Zero builds scenes, connects the pipeline, and runs generation.

Try a promo brief
10s demo
BriefShotsTimeline
02

Idea to timeline

Turn a story or product brief into shots, then land on a timeline ready to trim, subtitle, and export.

Create a family story video

The studio today

See the film platform in action

A walkthrough of how teams go from a brief to scenes on a timeline. The world model is the research we are adding underneath this studio.

0:00 / 0:00

Story & promo templates

Start from a story path—not an empty canvas

Browse curated templates for personalized story videos, product promos, and more—preview a card or jump straight into the editor when a starter is linked.

The studio today

From brief to finished cut in three steps

Start from a family story theme or a product promo brief. Templates load the pipeline so you swap inputs, generate scenes, and finish on a timeline while the world-model research continues.

1

Choose a story path

Start with a personalized family story video or a product promo brief. Templates load scenes and wiring so you are not staring at an empty canvas.

2

Add your story inputs

Drop in family photos, a product plate, or a plain-language brief. Leave the pipeline alone until you need fine control.

3

Generate, edit, export

Review scenes on a timeline, trim and subtitle if needed, then share a family short or export a promo cut. Continuity across a longer film is the world-model problem we are solving next.

Under the hood

Power tools when you need them

For teams iterating film scenes, ad variants, or SKUs, tab-group nodes fold repeated subgraphs into one place. Utility nodes, iteration sinks, and in-node generation carousels keep the graph lean while you ship story and promo cuts.

Any serious pipeline gets complex fast. You soon hit the same pattern over and over: identical node groups with different inputs — beats for film, variations for ads, or SKU swaps for ecommerce.

Example: you split a story into scenes. Each scene needs the same stages (prompt expansion, reference locks, generations, motion), only the brief and assets change. Ten scenes can mean the same subgraph copied ten times, which clutters the canvas and makes tweaks painful.

Flow Studio uses a dedicated tab-group node that sits on the canvas like any other node—except it embeds a tabbed workspace inside it. You shape one sub-flow once, duplicate it per scene (or shot / variation), and switch tabs to edit inputs. Nodes from inactive tabs hide so you always see one clean slice inside that group—without isolating it from the rest of the workflow.

That hiding is for focus, not isolation: nodes outside the tab group stay fully interactive with ports inside it—shared references, global prompts, mux nodes, and sinks keep connecting the same way they would on an open canvas.

You can also drop multiple tab-group nodes on one graph when different stages need their own iteration loops—for example scenes in one block and soundtrack or localization variants in another—instead of cramming unrelated tabs into a single control surface.

There is also a growing set of utility nodes for everyday operations—text splitting, image crop and resize, and more—so routine transforms stay on the canvas instead of bouncing through extra tooling.

Iteration sinks automatically route each output into the input for the next command and can run that loop as long as you need, which cuts down noisy chains of one-off nodes when you are stepping through variations or batch steps.

Generations from a model show up as a carousel inside the generator node, so you can flip through results, promote keeps, and manage assets without losing your place on the graph—better iteration with less hunting across panels.

Without tab groups

Shared blocks (refs, style, VO, …) fan out into every copy of the scene pipeline—edges multiply with each repeat.

Shared input nodes on the left send dashed feeds into three repeated scene stacks; each stack duplicates the same subgraph, implying many more scenes off-screen. Full canvas (excerpt) Shared inputs Bible / refs Look / LUT VO stem Scene 1 Brief Generate Motion Scene 2 Brief Generate Motion Scene 3 Brief Generate Motion … ×10
With tab groups on the canvas

Same shared blocks wire once into the tab group; one scene-visible pipeline—duplicate tabs instead of duplicating the spaghetti.

Three shared nodes on the main canvas connect with bundled dashed lines into a tab group; tabs S1 through S10 sit above a single visible Prompt to Generate to Video chain for scene two. Main canvas + embedded tab group Shared inputs Bible / refs Look / LUT VO stem Tab group (one sub-flow × many tabs) S1 S2 S3 S10 Visible sub-flow (Scene 2 only) Prompt Generate Video Shared hooks stay on the canvas; tabs duplicate structure + per-scene inputs only.
  • Less clutter Stop tiling identical stacks across an infinite canvas—fold them into one bounded workspace.
  • Copy structure Duplicate tabs to reuse wiring; swap prompts, references, and seeds per scene without rewiring edges by hand.
  • Focused editing Inactive tabs hide their nodes so you edit one slice at a time—without breaking connections to nodes on the main canvas.
  • Multiple groups Use several tab-group nodes when different stages need separate iteration loops (scenes, stems, locales, and more).

Reuse the story pipeline

Same story recipe, endless video variants

Once a personalized story or promo pipeline looks right, duplicate it for the next birthday theme, product SKU, or campaign hook—without rebuilding from scratch.

  • Family characters and product heroes stay consistent across scenes.
  • Swap briefs and inputs; keep the story-to-video wiring locked.
  • Download JSON anytime for teammates or client handoffs.

Use the studio

Start from a story template

Pick a personalized family story path, a product promo brief, or the full agency stack. Generate scenes, then finish the cut on a timeline. We're building the world model so those films can hold together longer.

How we got here

From LoRA training to a world model for film

September 2024 to now: training, storybooks, Omni Context, filmmaking, agents, then the limit that made the research obvious.

  1. August 2026

    Now

    World models for film production

    After generating a lot of videos through agents, the limit was obvious. Current video models still struggle with consistent clips, fluent long-form sequences, and fixing one part of a shot without regenerating the rest. That is the problem we set out to solve. We will keep you updated.

  2. June–July 2026

    Agentic film platform

    We completed the agentic platform for film production: story to scenes, generation, and a timeline you can actually edit.

  3. February 2026

    Agentic flows

    We started building agentic flows and Flow Studio workflows so a film could be planned and run as a graph, not a pile of one-off generations.

  4. November 2025

    Filmmaking stack

    We moved into filmmaking and added the video, image, and voice models needed to produce shots, not only stills.

  5. September 2024

    Image LoRA training

    We started as a platform for training image LoRA models so people could keep a face, product, or character consistent without a research setup.

Testimonials

What our customers are saying.

TheFluxTrain has been a game changer for my work flow. Training character models is a breeze with its auto-captioning feature, and their unique editor lets me create multi-character scenes seamlessly. No other platform offers an editor like this. I highly recommend it!

Karla Congson

Karla Congson

CEO at Agentiiv

I Actually used it, and I like the ease of the training mode, and it's very great for consistency, support replied super fast! and I really look for upcoming features. Please try make the final results like super realistic so that I can use it for ai influencers. Lots of love from a fellow colleague.

Jan Höppner

Jan Höppner

Co-founder at Narrify AI

Flexible options for everyone

Pricing Plans

Choose a subscription for regular use or buy one-off credits as you go.

Monthly
Annual 47% off

Free

Free
  • 200 credits one time
  • Agentic video editing forever
  • Agentic AI workflow creation
  • Cursor and Claude integration
  • Image, video and audio models

Starter

$11/mo
billed monthly
  • 5000 credits/mo
  • ~2-3 model trainings/mo
  • 48hr email support
  • Flow Studio + 20+ workflow templates (ads, infographics, film)
  • Image gen — Nano Banana, Flux 2 Pro, GPT Image 2 & more
  • Story Studio — inpaint, edit, reference & LoRA editor
  • Custom training — face, product, character, influencer (Flux 1/2, Qwen, LTX)
  • Auto-captioning, Character Builder & Virtual Model
  • Video — text, image, reference, multi-shot & interpolation
  • Talking avatars, motion transfer & actor replacement
  • LoRA-guided video & video extend
  • AI Video Editor — timeline, subtitles & social export
  • Voice — TTS, dubbing, voice change, SFX & noise removal
  • Storybooks & shareable links
  • Action Studio, tools, folders, CLI & API

Pro

Recommended
$19/mo
billed monthly
  • 10000 credits/mo
  • ~4-5 model trainings/mo
  • 24hr email support
  • Flow Studio + 20+ workflow templates (ads, infographics, film)
  • Image gen — Nano Banana, Flux 2 Pro, GPT Image 2 & more
  • Story Studio — inpaint, edit, reference & LoRA editor
  • Custom training — face, product, character, influencer (Flux 1/2, Qwen, LTX)
  • Auto-captioning, Character Builder & Virtual Model
  • Video — text, image, reference, multi-shot & interpolation
  • Talking avatars, motion transfer & actor replacement
  • LoRA-guided video & video extend
  • AI Video Editor — timeline, subtitles & social export
  • Voice — TTS, dubbing, voice change, SFX & noise removal
  • Storybooks & shareable links
  • Action Studio, tools, folders, CLI & API

Limited time — unlimited Gemini Omni Flash

Advance

$39/mo
billed monthly
  • 25000 credits/mo
  • ~10-11 model trainings/mo
  • 24hr priority support
  • Flow Studio + 20+ workflow templates (ads, infographics, film)
  • Image gen — Nano Banana, Flux 2 Pro, GPT Image 2 & more
  • Story Studio — inpaint, edit, reference & LoRA editor
  • Custom training — face, product, character, influencer (Flux 1/2, Qwen, LTX)
  • Auto-captioning, Character Builder & Virtual Model
  • Video — text, image, reference, multi-shot & interpolation
  • Talking avatars, motion transfer & actor replacement
  • LoRA-guided video & video extend
  • AI Video Editor — timeline, subtitles & social export
  • Voice — TTS, dubbing, voice change, SFX & noise removal
  • Storybooks & shareable links
  • Action Studio, tools, folders, CLI & API

Limited time — unlimited Gemini Omni Flash

Cosmo

$189/mo
billed monthly
  • 125000 credits/mo
  • 50+ model trainings/mo
  • 24hr priority support
  • Flow Studio + 20+ workflow templates (ads, infographics, film)
  • Image gen — Nano Banana, Flux 2 Pro, GPT Image 2 & more
  • Story Studio — inpaint, edit, reference & LoRA editor
  • Custom training — face, product, character, influencer (Flux 1/2, Qwen, LTX)
  • Auto-captioning, Character Builder & Virtual Model
  • Video — text, image, reference, multi-shot & interpolation
  • Talking avatars, motion transfer & actor replacement
  • LoRA-guided video & video extend
  • AI Video Editor — timeline, subtitles & social export
  • Voice — TTS, dubbing, voice change, SFX & noise removal
  • Storybooks & shareable links
  • Action Studio, tools, folders, CLI & API

Rate Card & Plan Usage

Model / FeatureCost
Starter
$11/mo
Pro
$19/mo
Advance
$39/mo
Cosmo
$189/mo
Video
Google Veo 3.1 Fast 720p382 credits / 4s 13 videos 26 videos 65 videos 327 videos
Google Veo 3.1 Fast 1080p382 credits / 4s 13 videos 26 videos 65 videos 327 videos
Google Veo 3.1 720p1020 credits / 4s 4 videos 9 videos 24 videos 122 videos
Google Veo 3.1 1080p1020 credits / 4s 4 videos 9 videos 24 videos 122 videos
Kling O3 Standard 720p428 credits / 4s 11 videos 23 videos 58 videos 292 videos
Kling O3 Pro 1080p428 credits / 4s 11 videos 23 videos 58 videos 292 videos
Seedance 1.5 Pro (480p)342 credits / 4s 14 videos 29 videos 73 videos 365 videos
Seedance 1.5 Pro (720p)771 credits / 4s 6 videos 12 videos 32 videos 162 videos
Seedance 1.5 Pro (1080p)1735 credits / 4s 2 videos 5 videos 14 videos 72 videos
Seedance 2.0 Fast (480p)342 credits / 4s 14 videos 29 videos 73 videos 365 videos
Seedance 2.0 Fast (720p)771 credits / 4s 6 videos 12 videos 32 videos 162 videos
Seedance 2.0 (480p)342 credits / 4s 14 videos 29 videos 73 videos 365 videos
Seedance 2.0 (720p)771 credits / 4s 6 videos 12 videos 32 videos 162 videos
Seedance 2.0 (1080p)1735 credits / 4s 2 videos 5 videos 14 videos 72 videos
Seedance 2.5 (480p)262 credits / 4s 19 videos 38 videos 95 videos 477 videos
Seedance 2.5 (720p)589 credits / 4s 8 videos 16 videos 42 videos 212 videos
Seedance 2.5 (1080p)1326 credits / 4s 3 videos 7 videos 18 videos 94 videos
Happy Horse (text / image / ref, with audio)510 credits / 4s 9 videos 19 videos 49 videos 245 videos
Happy Horse (text / image / ref, no audio)408 credits / 4s 12 videos 24 videos 61 videos 306 videos
Happy Horse Video Edit642 credits / 4s 7 videos 15 videos 38 videos 194 videos
Gemini Omni Flash (720p, synced audio)331 credits / 4s 15 videos 30 videos 75 videos 377 videos
FLUX 3 (720p)541 credits / 5s 9 videos 18 videos 46 videos 231 videos
FLUX 3 (1080p)924 credits / 5s 5 videos 10 videos 27 videos 135 videos
FLUX 3 Draft (720p)191 credits / 5s 26 videos 52 videos 130 videos 654 videos
FLUX 3 Extend (720p)1306 credits / 5s 3 videos 7 videos 19 videos 95 videos
FLUX 3 Extend (1080p)1689 credits / 5s 2 videos 5 videos 14 videos 74 videos
FLUX 3 Draft Enhance (1080p)924 credits / 5s 5 videos 10 videos 27 videos 135 videos
Image
Nano Banana 250 credits / image 100 images 200 images 500 images 2500 images
Nano Banana Pro60 credits / image 83 images 166 images 416 images 2083 images
Z-Image Base50 credits / image 100 images 200 images 500 images 2500 images
WAN 2.7 Pro50 credits / image 100 images 200 images 500 images 2500 images
GPT Image 2135 credits / image 37 images 74 images 185 images 925 images
Flux 2 Pro50 credits / image 100 images 200 images 500 images 2500 images
LoRA Models
Flux 1 Dev25 credits / image 200 images 400 images 1000 images 5000 images
Training
Flux 1 Training (2000 steps)2858 credits / train 1 trainings 3 trainings 8 trainings 43 trainings
Flux 2 Training (2000 steps)13000 credits / train 0 trainings 0 trainings 1 trainings 9 trainings
Qwen Image Training (2000 steps)3250 credits / train 1 trainings 3 trainings 7 trainings 38 trainings
LTX Video LoRA (2000 steps)7800 credits / train 0 trainings 1 trainings 3 trainings 16 trainings
Voices & Audio
Text to Speech (1000 chars)154 credits 32 units 64 units 162 units 811 units
Voice Changer (60s)153 credits 32 units 65 units 163 units 816 units
Voice Isolator (60s)153 credits 32 units 65 units 163 units 816 units
Dubbing (60s)344 credits 14 units 29 units 72 units 363 units
Sound Effects89 credits 56 units 112 units 280 units 1404 units

Start a film on the studio we already ship

World-model research is underway. The agentic filmmaking tools are live.

Get Started

No credit card required

Frequently asked questions

A world model here is a persistent representation of characters, environments, objects, and camera state. Instead of asking a video model to invent an entire clip from scratch, it predicts the next short beat of a cinematic world from previous frames, references, and a director's instruction.
We tried. After generating a lot of videos through agents, the same failures showed up. Characters drift between clips, shots are hard to stitch into a fluent sequence, and fixing one gesture often means regenerating the whole clip. Those models are useful for short shots. They are not built for long-form film.
Yes. Our agentic film platform is live. Write a story or brief, run Flow Studio or Director Zero, generate scenes, and finish on a timeline. The world model is the research we are doing so those films can stay consistent for longer.
It is early. We will share demos and updates as they exist. The studio you can use today does not wait on that work.
{ "Flow Studio and Director Zero are the production layer": "agents, workflows, models, and a timeline. The world model is the visual engine we want underneath that layer, so continuity does not depend on stitching unrelated clips." }
Yes. Write your story or logline on the homepage, pick Short Film or Auto mode, and the pipeline handles script, storyboard, animation, voiceover, and final cut. Optional reference images help lock characters or products.
No. Reference images are optional. Add photos when you want consistent characters, family members, pets, or product shots across scenes.
Short Script develops and refines your story from an idea. Short Film runs the full automated pipeline—script through final animated cut with voiceover. Auto picks the best path from your input.
No. The automated pipeline produces a finished cut. You can trim, arrange, and subtitle on the timeline when you want small tweaks—no node canvas required for the guided short-film path.
Yes. Family stories use optional photos and themes like bedtime or birthday. Product promos start from a plain-language brief about your product, audience, and hook. Both finish as shareable short films.
No, you need to buy credits to use the platform. We offer both subscriptions (monthly or annual) and one-off credit passes.
You can choose a subscription plan for regular use (monthly or annual with up to 47% off) or buy one-off credit passes valid for 1 year. Both options are available on our pricing page.
Usually 3-9 photos are enough to train a model, depending on the complexity of the subject.
Training a model costs 2000+ credits. Actual cost depends on training settings. After that, each image generation costs 25 credits. Video generation uses more credits.
Yes! All plans include every feature on the platform — image-to-video, text-to-video, motion transfer, talking avatars, and an AI Video Editor to trim, split, add subtitles, and render videos. Premium models use more credits per generation; see the Rate Card on our pricing page.
Yes. Use our cartoon creation tool to generate a cartoon character from a text description. You get 8 images in different poses, then you can train a model on them for consistent generations.
Our video editor lets you trim clips, split videos, add text overlays and shapes, arrange a timeline, and render in various resolutions including vertical (TikTok/Shorts) and square (Instagram). You can save and reopen projects.
On a case-by-case basis. Reach out to us at saquib@zust.ai
Yes. Reach out to us at saquib@zust.ai