FeaturesTimeMachineModels BenchmarksPricingDocsGet a key
Public Preview · local runtime · no cloud rendering

Run large AI video models locally — without hand-tuning memory.

Veizik is a hardware-aware local AI media runtime powered by the LimML engine: it checks your hardware and reports a runnable support tier and recommended execution profile. Public Preview — internally verified on Linux + NVIDIA 24 GB class; Windows WSL2 is experimental.

Get free key Check compatibility
Free to install & verify on your own GPU Linux · NVIDIA 24 GB class (internally verified) Windows WSL2 · experimental
9.55 GB
LTX-13B peak VRAM · RTX 3090 (internal)
6
engine paths checked at block level
3
free CLI commands live: doctor · login · status
0
media or prompts leave your machine
Preview
Linux + NVIDIA verified · WSL2 experimental
0 /100
paid Founding seats claimed
Real renders

Made locally with the LimML engine

Rendered on one RTX 3090 we own, with every number below measured on that run — not estimated. Full conditions travel with each clip so you can judge it the way you'd judge your own hardware.

Espresso pour · seeded image-to-video

measured
1216×704 · 73 frames @ 24 fps (3.04 s) · 40 steps · bf16 · warm run
GPURTX 3090 · 24 GB · 420 W cap
Video modelLTX-Video (image-to-video)
Seed stillFLUX.1-dev · 32 steps
Render wall time657 s · 9.0 s/frame
Peak VRAM (render)20.6 GB
Seed still cost1940 s · 24.2 GB peak (CPU-offload)

Prompt, seed and every option are published with the run manifest. The seed still is what fixes composition — text-to-video at the same settings drifted off the brief, so the lineup is seeded.

seed 42 · veizik-run-report-v1

The seed still that pinned this shot: view the FLUX frame. Earlier internal clips below.

Character animation

960×544 · internal render · Public Preview

LTX · product motion

1280×704 · internal render (fp8)

Wan · camera move

832×480 · internal render (fp8)
What it does

What's live today, and what's still coming

Public Preview. The tags below say exactly what the downloadable build does today versus what is still internal or planned — see Known Issues for the honest ledger.

Hardware check & profile

veizik doctor scans your machine and prints a runnable support tier per model family, so you know what fits before you render.

live in the download

Low-VRAM execution

Capacity-compute tiering (hot / warm / cold placement) targets fitting large video models on a single 24 GB GPU. LTX-13B peak 9.55 GB on an RTX 3090 (internal).

internal · VRAM measured

Native CUDA DiT engines

Six model-family engine paths are numerically checked at the block/engine level (rel_L2 ~1e-6–1e-7 vs reference) — a component-level check, not an end-to-end render benchmark.

block-level numerical check

Local & licensed

veizik login redeems a free key for a server-signed entitlement. Your media and prompts stay on your machine; only license data is exchanged.

live in the download

TimeMachine Render

Branch from a saved checkpoint: keep the good prefix, re-render only the failing tail. Validated internally on GPU; ships as a Preview build asset — not in the current download yet.

planned · Preview build

ComfyUI integration

Goal: run your existing ComfyUI graph under Veizik with low-VRAM + native engines. Under preview validation — run / serve land in the next release.

upcoming preview
Planned · Preview build

Branch instead of re-render

Status: TimeMachine is validated internally on GPU and ships as a downloadable Preview build asset — it is not in the current public download yet. The diagram below describes the intended behavior.

The biggest waste in AI video is "the first part is good but the rest breaks, so I re-render everything." TimeMachine addresses that — one checkpoint infrastructure serves both crash-resume (same settings) and creative branching (new settings).

base render · denoise steps 5 10 15 ◀ branch 20 30 prefix 0..15 — REUSED (kept) tail 16..30 — new style/cfg real GPU (LTX): re-ran 8/16 steps · 50% saved timeline stored: 807 KB · sparse latents only
  • Keep the good part. Composition, faces, product placement, lighting are preserved — not thrown away.
  • Fix only what's wrong. A clip that collapses after ~6 s regenerates only its tail — a fraction of the work.
  • A/B fanout. Several endings (warm / cyberpunk / luxury) from one branch point, in parallel across the farm.
  • Auto-find "from where" prototype A collapse detector to flag face drift, flicker, and motion jumps and suggest a branch point — prototype, not in the build.
  • Low storage. Sparse latent checkpoints — a branchable timeline is ~kilobytes, not a copy of every frame.
Compatibility

Model families with a native engine path

These families have a native CUDA DiT engine path, each numerically checked at the block/engine level (rel_L2 ~1e-6–1e-7 vs reference — a component check, not an end-to-end render benchmark). A universal fallback path for other models is under development.

Native DiT

Step-Video 30B

rel 2.35e-6 · single 24 GB
Native DiT

HunyuanVideo

rel 3.6e-6 · dual/single
Native DiT

Wan 2.1

rel 5.53e-7
Native DiT

LTX-Video

rel 1.94e-7 · measured
Native DiT

CogVideoX

rel 8.43e-7
Native DiT

FLUX.1-dev

rel 4.58e-7 · image
planned

Llama / Qwen

LLM · OpenAI-compatible endpoint
Planned

Universal fallback

other models · in development
Works with your stack
runs on the tools you already use — no rewrite, no lock-in
upcoming

ComfyUI

integration · next release
runtime

NVIDIA CUDA

sm_80+ · fp16 TC
pipeline

diffusers

T2 fallback path
framework

PyTorch

2.x · bf16 / fp16
OS

WSL2 / Linux

Windows or native
planned

OpenAI-compatible

LLM endpoint · planned
Where we are

What is measured — and what isn't yet

We separate claim types on purpose. Measured internally so far: the LTX render times and peak VRAM below, plus block/engine-level numerical equivalence. Still blank: the other model families, and any external reproduction — nobody outside has verified these yet, which is what the Preview is for. Method: benchmark-method.md.

Render (RTX 3090, 24 GB · 420 W cap · internal)Wall timePeak VRAMOOM
LTX image-to-video · 1216×704 · 73f · 40 steps657 s20.6 GBnone
LTX text-to-video · 1216×704 · 73f · 40 steps597 s20.6 GBnone
FLUX.1-dev seed still · 1216×704 · 32 steps1940 s24.2 GBnone
CogVideoX · HunyuanVideo · Wan · Step-Videomeasuring…measuring…
Warm runs, fixed seed, one card. Roughly 9 s per frame at this resolution and step count — so a 3-second clip is about 11 minutes, not seconds. The FLUX seed used CPU offload, which trades speed for headroom; that is why it costs more wall time than the video itself. The remaining families are still being measured and stay blank until they are.
Engine pathBlock/engine rel_L2
Step-Video 30B2.35e-6block-level
LTX-Video1.94e-7block-level
CogVideoX8.43e-7block-level
HunyuanVideo3.6e-6block-level
Wan 2.15.53e-7block-level
FLUX.1-dev4.58e-7block-level
Numerical equivalence vs the reference implementation at the block/engine level — not an end-to-end generation benchmark.
VRAM budget · single RTX 3090 (24 GB · internal)LTX-13B peak 9.55 GB
9.55 GB 24 GB
One internally measured VRAM figure (LTX-13B). External reproduction is exactly what this Preview is for — report your run.
Why local

Why run it on your own hardware

Cloud video tools bill a monthly subscription plus metered GPU credits that scale with how much you render. Veizik is a local runtime instead: it runs on the GPU you already have, keeps your media on your machine, and takes the manual memory tuning out of running large models.

No per-render bill

Cloud video tools bill a monthly subscription plus metered GPU credits that grow with usage. Veizik renders on the GPU you already own — there's no per-render charge, only the electricity the card draws.

your GPU · electricity only

No manual memory tuning

Skip the model / VRAM / offload trial-and-error and the OOM restarts. veizik doctor scans your machine and reports a runnable support tier per model family, so you know what fits before you render.

doctor · live auto-apply · planned

Big models on a 24 GB card

Capacity-compute tiering targets fitting large video models on a single 24 GB GPU — LTX-13B peaked at 9.55 GB on an RTX 3090 (internal). Frames and prompts are never uploaded to a rendering cloud.

LTX-13B peak 9.55 GB · internal

You own the runtime

It runs locally and keeps working offline inside your license grace window. Media, prompts, and project data stay on disk — only license activation talks to veizik.com, and no output is sent anywhere.

local · offline grace

Every number above is measured internally and left blank where it isn't — see exactly what is measured, and what is still measuring…, in measured status.

Preview in 3 steps

Install, check your hardware, activate.

1 · Install

curl -fsSL https://veizik.com/install.sh | sh on Linux or Windows WSL2 (git + Python 3.10+). No key needed to install.

2 · Check your hardware

veizik doctor scans this machine and prints the support tier per model family — so you know what runs before you commit.

3 · Activate a free key

veizik login <key> redeems a free key from veizik.com for a signed entitlement. Universal t2v render is experimental today.

How it compares

Local runtime vs cloud tools

VeizikCloud video (Runway / Pika)Raw ComfyUI
Runs on your own GPU yesno — cloudyes
30B video on 24 GB no OOMn/aneeds manual low-VRAM tuning
Checkpoint branch (keep prefix)TimeMachine preview buildre-renderre-render
Per-render costyour electricitymetered / clipyour electricity
Data leaves your machine stays localuploadedstays local
PriceFree Preview · Founding $9/momonthly cloud planfree / DIY
Founding 100

The first 100 paid subscribers

  • Founder price locked for 12 months
  • Founder feedback access
  • Priority compatibility support
0 / 100
paid seats · counts real paid subscriptions only
Plans

The CLI is the product. The license is the capability.

The public runner is the distribution channel — free to install, free to inspect. What you buy is a local runtime license: commercial use rights, stable profiles, private Model Adapters, advanced quantization, Capsules, Queue/batch, Recover & TimeMachine, the API bridge, update rights, and how many machines may run at once. No cloud credits. You supply the GPU.

Metering unit = concurrent run nodes. You are never charged for each machine you install on — install on several computers; the license decides how many may render at the same time.

STARTER PREVIEW shipped
$0 · 7 days
Trial starts at your first successful render — not at activation, so install trouble never burns your days.
  • 1 personal seat · install on 1 computer · run on 1 at a time
  • doctor hardware + per-model support scan
  • Limited sample & Capsule access
  • Experimental render path (Linux + NVIDIA)
  • Watermarked · commercial use excluded
Start free preview
FOUNDING CREATOR checkout live
$9 /mo · or $79/yr
Free until general availability · then 12 months of Founder pricing
  • 1 personal seat · install on 2 computers · run on 1 at a time
  • Commercial output rights · watermark removed
  • Stable pinned profiles (not experimental)
  • Creator Model Adapters & Capsules
  • Automatic runtime-pack updates
  • Device change 3×/year, self-service
Get Founding Creator — $9/mo
$79/year (~27% off) · cancel anytime · Polar handles billing
FOUNDING PRO RUNTIME PASS checkout live
$249 · first year
Pro Preview access for 24 months · private release channel
  • 1 personal seat · install on 3 computers · run on 2 at a time
  • Everything in Founding Creator
  • Advanced FP8 / INT8 quantization private preview
  • Queue & batch execution development
  • Advanced GPU Oracle planning private preview
  • Early access to Recover / TimeMachine / API private preview
Get Founding Pro — $249/yr
Preview features above ship as they are released — the tags say where each one stands today.
STUDIO NODE organization contract
Custom
Priced on seats + concurrent run nodes
  • Named seats · negotiated concurrent run nodes
  • Dedicated / private Model Adapters
  • Internal distribution of runtime packs
  • Central license server development
  • Usage logs, audit trail, API bridge development
Talk to us
Contract only · not self-serve.

Founding Creator checkout is live — use the same email at checkout as your free key and payment upgrades that key automatically. Founding Pro and Studio Node are not purchasable today; they open after Preview validation, and the links above only register interest. Sales: sales@veizik.com.

One bootstrap CLI for every tier — no separate builds. The public binary carries the parser, updater, doctor, license client, telemetry client, pack loader, signature verifier and the public Adapter interface; a server-signed entitlement unlocks the matching private runtime pack (LimML native runtime, advanced kernels, Adapters, Capsules, GPU Oracle planner, Recover, TimeMachine, Queue, API bridge). The runtime — not the UI — enforces features. Billing by Polar (merchant of record; USD, tax included). Feature states are labelled research / development / private preview / public preview / shipped; render-time figures are published only where actually measured (LTX today) and left blank elsewhere.

Quickstart

Public Preview usage

# install (git + python 3.10+)
curl -fsSL https://veizik.com/install.sh | sh

veizik doctor                   # scan GPU + per-model support tier
veizik login <key>              # redeem a free key for a signed entitlement
veizik status                   # show current tier & entitlement
# experimental universal render — Linux + NVIDIA, your own torch/diffusers env
# measured on our RTX 3090: ~9 s/frame at 1216x704, 40 steps (see the status section)
veizik t2v "a barista pouring latte art, warm cafe" \
  --model ltx --w 1280 --h 704 --frames 49 --steps 40
# NOT in the current download — landing in upcoming Preview releases:
#   veizik run my_workflow.json     # ComfyUI integration (next release)
#   veizik branch --from-step 15    # TimeMachine (Preview build asset)
#
# track status: github.com/veizikhq/veizik  ·  KNOWN_ISSUES.md
Questions

FAQ

No. Veizik runs entirely on your own GPU — nothing is uploaded and no cloud render credits are involved. A license is a local runtime license; you supply the hardware.

The Preview is internally verified on Linux x86_64 with an NVIDIA 24 GB-class GPU; Windows WSL2 + NVIDIA is experimental. We don't claim every GPU works — run veizik doctor first (it's free and needs no key) and it tells you what it detects. Apple Silicon and smaller memory classes are planned.

Not yet. ComfyUI integration (run / serve) is under preview validation and lands in the next release — it is not in the current download. The goal is that your existing graph runs under Veizik with low-VRAM + native engines; we'll say it's live only when the downloadable build does it.

It branches a render from a saved denoise-step checkpoint: the approved prefix is reused and only the failing tail is regenerated with a new prompt / style / CFG. It has been validated internally on GPU. It ships as a Preview build asset and is not in the current public download yet.

No — Veizik is a proprietary commercial runtime. Installation and evaluation are free; production use requires a license. The engine (LimML) is not distributed as source.

Permission is a server-signed entitlement token, not the key itself. Activation binds to a device fingerprint, and feature gating lives in the runtime executor — not just the UI. You can issue and revoke keys above.

Step-Video 30B, HunyuanVideo, Wan 2.1, LTX, CogVideoX, and FLUX.1 each have a native CUDA DiT engine path, numerically checked at the block/engine level (rel_L2 ~1e-6–1e-7 vs reference — a component check, not an end-to-end render benchmark). The experimental t2v render path today targets LTX-class models on Linux + NVIDIA.

Live

Get an API key

LIVE

Free keys render watermarked, non-commercial. Your key authorizes via a server-signed entitlement — issued below, enforced in the runtime.

Your API key (shown once — copy it now):
Set once: export VEIZIK_API_KEY=… then veizik doctor.

Keys are stored hashed; the raw key is shown only once. Revoke anytime — the entitlement stops validating.

Roadmap
Coming soon

Being finalized

NEXT

ComfyUI integration

Run your existing ComfyUI graph under Veizik with low-VRAM + native engines — run / serve, under preview validation.

NEXT

TimeMachine Preview build

Checkpoint branching in the downloadable build: keep the good prefix, re-render only the failing tail. Validated internally on GPU.

LATER

Pro & Studio tiers

Batch, A/B fanout, multi-GPU and team seats open after Preview validation. Your free account carries over.

Live in the current download: doctor, login/status, and a free entitlement, plus an experimental universal t2v path. Everything above is on the way — we mark it live only when the build does it.

Stay in the loop

Get launch & billing updates — or ask us anything

Drop your email to hear when billing opens and Preview builds ship, or send a compatibility question — it reaches us directly.

No payment, no spam. Launch, billing-open, and Preview-build notices only. Unsubscribe anytime.
or email support@veizik.com