VFX Engine for Nuke/Help

Help and documentation

Install the plugin, sign in, and generate your first AI element inside Nuke. If anything here does not answer your question, email info@vfxengine.com and we will help.

Quick start

01
Request access. On the Nuke page, click Request beta access. You will make a free VFX Engine account if you do not have one. We onboard in small batches, so there may be a short wait.
02
Get approved. When we approve you, 50 free credits land in your account and you get an email. The download button on the Nuke page unlocks.
03
Download and install. Download the plugin (approved accounts only), unzip it somewhere permanent, and double-click the installer for your operating system.
04
Sign in from Nuke. Launch Nuke, open the VFX Engine (Beta) panel, and click Sign in with VFX Engine. Approve in your browser.
05
Generate. Drop a model node, wire a Prompt node into its Prompt arrow, wire a source, and click Generate. Confirm the exact cost and the result drops in as a Read node.

Install

The download is a single zip. Unzip it somewhere permanent (for example your home folder), then run the installer for your operating system. There is nothing to pip install.

Windows: double-click install.bat.
macOS: open Terminal, type sh and a space, drag install.command into the window, and press Enter. (macOS blocks double-clicking downloaded scripts. To allow it instead: System Settings, Privacy & Security, Open Anyway.)
Any system: from a terminal, run python install.py.

The installer copies the toolset to its permanent home at ~/.nuke/vfxengine-toolset and points ~/.nuke/init.py there, which Nuke reads however it is launched. Where you unzipped never matters, and the unzipped folder can be deleted afterward. Restart Nuke and you will see a VFX Engine (Beta) menu. It appears in three places: the top menu bar, the node toolbar down the left side, and the Tab search in the node graph. The sign-in panel opens from that menu and can be docked into your layout like any other panel.

Updating is the same action: download the new zip and run its installer, then restart Nuke. It replaces the previous version in place and keeps your sign-in and settings. The plugin tells you when an update is available: the panel shows a notice, and the node mentions it after a render.

Sign in

Open the VFX Engine (Beta) menu and choose Open Panel, then click Sign in with VFX Engine. A browser window opens; approve it and you are done, with no keys to copy. The panel is your account home: it shows your live credit balance and every model available to you with its live price. Pick a model and click Add node to drop it into your Node Graph (or create nodes straight from the menu); generation itself always happens on the node. My results opens your generation history on vfxengine.com, and Examples opens finished shots with full Nuke projects to download. The first time you generate, you accept a short beta agreement so your consent is on record.

The nodes

The beta ships generation, finishing, and matte nodes, plus two prompt tools. Create any of them from the VFX Engine (Beta) menu, the left toolbar, or by pressing Tab and typing its name: the node lands at your cursor and wires itself to your selection. These docs track the plugin release, currently v1.5.0; when a node changes, this page changes with it.

Nano Banana 2. Dependable image generation and editing. A great first node.
Nano Banana Pro. Higher-fidelity Nano Banana for hard briefs: complex composition, text, and character consistency.
GPT Image 2. Strong instruction following and text rendering, with reference-image support.
Seedance 2.0. Video generation and restyling up to 15 seconds, with native audio. Wire your footage into its Reference video input to keep your camera move and action.
Gemini Omni. Any-to-any video: combine up to seven images and an optional reference video with your prompt. Wiring a reference video switches billing to a flat per-video rate.
Kling 3.0 / Kling O3. One video node each: wire an image into Image 1 for image-to-video, or leave it empty for text-to-video. The node routes to the right mode automatically.
Grok Imagine 1.5. The most affordable video node, and the strongest at holding a face or product through a move: wire a still into Image 1 and it stays faithful to that frame instead of redrawing it. Leave Image 1 empty for text-to-video. Up to 15 seconds with native audio, at 480p or 720p, priced per second, so 480p is the cheap way to test a move before committing.
WAN 2.7. One node for text-to-video, image-to-video, and text-to-image. Pick Video or Image with the Output knob; in Video mode, wiring Image 1 gives image-to-video.
Topaz Image / Topaz Video. Premium upscaling to finish a plate in place: wire a Read (or a clip) into the source, pick 2x or 4x, and the upscaled result drops back in as a Read. No prompt needed; video is trimmed to 15 seconds per run.
SAM 3 Image / SAM 3 Video. Text-driven matting: type what to isolate (for example "the car" or "person on the left") and get back either a clean cutout (RGBA) or a black-and-white alpha matte, chosen with the Output Mode knob. The matte mode feeds a Merge or Keymix like any other roto.
Prompt / Prompt Enhancer. Drive any model node's prompt from the graph. A Prompt node holds text; a Prompt Enhancer improves it (Regular prose or structured JSON, 1 credit). Wire either into a model node's Prompt arrow.

Image nodes: Image 1 + Ref arrows. Every image you wire is a reference, read in wire order. There is no separate plate slot: your prompt decides what happens to them. Ask for a change and you get an edit; ask for a new scene and you get a fresh image built from what the references establish (a character, a product, a look). Leave everything unwired to generate from the prompt alone. Click + Ref input to add more reference arrows (a character sheet, a logo, a lighting reference); the node keeps one spare ready as you wire, up to ten. Wire whichever arrows suit you: if you use only a Ref and leave Image 1 empty, the node simply uses it.

Seedance 2.0: one node, two ways to guide it. Wire First and/or Last to pin exact output frames and let the model interpolate between them. Or click + Ref input and wire the Reference video (it drives all motion and camera) alongside up to four Ref images that teach the look and identity across the whole shot; stills that look like start or end frames bind to those positions. A First or Last frame cannot combine with the Reference video (the provider rejects the mix), so the node refuses that wiring with a clear message before anything is charged. Reference mode renders at up to 4K (the 480p/720p limit is on the reference video you upload, not the output), and the Audio file knob accepts a .wav or .mp3 for audio-driven generation in either mode.

Prompt from the graph. Every model node with a prompt has a Prompt input arrow, and that arrow is the only place its prompt comes from - there is no prompt box on the model node itself, so a prompt lives in exactly one spot. Wire a Prompt node (holds text) or a Prompt Enhancer node (improves it with AI, Regular prose or structured JSON, one credit) into it and its text drives the run. Chain them like Studio: Prompt into Prompt Enhancer into your model node. A prompted node will not generate until a Prompt node is wired in.

Generate a shot

Create a node, write a prompt, and wire your inputs. The Match source aspect ratio checkbox shapes the output to your plate so nothing gets recomposed to a square default.

Click Generate: it sits alone on its own full row under the settings. You confirm the exact credit cost first and are never charged more than the number you approved; prices are fetched live on every generate and are never stored in your script. While it runs, the Status line reports live progress with elapsed time and the model's typical duration. When it finishes, the result drops into your script as a colorspace-tagged Read node and the file lands in a vfxengine folder next to your saved script, named by model, never overwriting an earlier take (the plugin asks you to save your script first, so results always have that one predictable home). Any refund for a failed render shows in your credit history, and if anything goes wrong the Status line says what happened and what to do next in plain words.

Credits

Credits run every generation. Approved beta testers get a monthly allowance of free credits; those refresh each month and expire at month end. Purchased credits are valid for 12 months from purchase. There is no subscription: buy a pack when you need it, here or from the Buy credits button inside Nuke.

credits

$4.99$3.99

$0.080 per credit, valid 12 months

Worked examples

For shots built end to end, see the experiments page: each one has the grey 3D blockout, the rendered shot, the exact prompts, and the full Nuke project to download, with every Read resolving out of the box. Open one next to these docs and you can trace exactly how the nodes were wired.

Troubleshooting

The VFX Engine menu did not appear

Make sure the installer actually ran (it adds one line to ~/.nuke/init.py), then fully quit and reopen Nuke. Confirm you are on Nuke 13 or newer. If you launched Nuke from a terminal with a custom NUKE_PATH, the installer path still wins because it writes to init.py, which Nuke always reads.

On macOS, install.command will not open

macOS blocks downloaded scripts with an "Apple could not verify" message. Fastest fix: open Terminal, type sh followed by a space, drag install.command into the window, and press Enter. Prefer double-clicking? Open System Settings, go to Privacy & Security, scroll to the Security section, click Open Anyway, then open the file again. The installer only needs to run once.

I am on Nuke 13, 14 or 15

Supported. Those versions use a different Qt build than Nuke 16 and up, and the plugin handles both automatically, so there is nothing extra to install or configure. Nuke 12 and older are not supported: they run Python 2, which the plugin cannot use. You will see a friendly dialog there instead of an error. We build and test on Nuke 17, so if anything looks off on an older version, email info@vfxengine.com and we will get on it quickly.

Make sure you are signed in at vfxengine.com in the same browser, approve the prompt that opens, then click Sign in with VFX Engine again in the panel. The panel shows your live balance once it connects.

The node refuses to generate and points at its inputs

Read the Status line - the node explains exactly what to change before anything is charged. The usual one is a Seedance node mixing Ref images with First and Last frames (pick one: refs teach the look across the whole shot, First and Last pin exact frames; they cannot combine). Image nodes are relaxed about their inputs: wire any arrow you like, or none at all and generate from the prompt.

The node says to wire a Prompt node to generate

That model needs a prompt, and prompts come from a Prompt node - there is no prompt box on the model node itself. From the VFX Engine (Beta) menu open Prompt tools and add a Prompt node, type your text, wire it into the model node's Prompt arrow, and click Generate again.

A generation was blocked or I ran out of credits

Buy credits from the Buy credits button on the node or panel, or on this page. Every generation shows its exact cost before it runs, so you always know the number before you confirm.

My video result plays only one frame

That was a bug in plugin versions before v1.1.7: the Read node was created without probing the movie, so its frame range stayed 1-1. Update the plugin (download, run the installer, restart Nuke). For a Read that is already frozen, open its File browser, re-select the same file, and click Open - Nuke then reads the real frame range.

My result saved to a temp folder instead of the vfxengine folder

That was a v1.1.7 bug: on a brand-new, never-saved script the save-first prompt could be skipped, and the file fell back to the operating system temp folder (which the OS purges). Fixed in v1.1.8 - update the plugin. If a result already landed in temp, move it out now, or re-download it from your dashboard on vfxengine.com: the server copy is unaffected.

The result did not show up

Check the Status line at the bottom of the node. Renders take from a few seconds to a couple of minutes depending on the model. Failed renders refund their credits automatically, so you never pay for one that did not complete. And if the Status line ever says the plugin lost sight of a run, open your dashboard on vfxengine.com: a completed result still lands there even when Nuke stops tracking it.

FAQ

Which Nuke versions are supported?

Nuke 13 or newer (Nuke or NukeX) on macOS or Windows, built and tested on Nuke 17. The plugin uses Nuke’s bundled Python and Qt, so there is nothing else to install. Nuke 12 and older run Python 2 and are not supported. Nuke Non-Commercial is not supported either: it blocks third-party plug-ins and MP4 export, which the plugin needs.

How do I update the plugin?

When an update is available the panel says so and shows an "Update now" button (v1.1.8 and newer): one click downloads and installs it in place, then you restart Nuke. You can always update by hand instead: download the newest zip and run its installer, exactly like installing. Either way it replaces the previous version in its permanent home (~/.nuke/vfxengine-toolset) and keeps your sign-in and settings. For critical fixes the plugin may require updating before it will generate again.

Do I need an internet connection?

Yes, a stable one: generation runs in the cloud. The plugin uploads your wired inputs, the model renders, and the result downloads back next to your script, so a dropped connection interrupts that round trip. Brief blips are retried automatically. If the connection drops before a run submits, nothing is charged; if it drops after, the finished result still lands in your dashboard on vfxengine.com, and any refund shows in your credit history.

Do I need to know Python or use a terminal?

No. Unzip the download and double-click the installer for your operating system. There is nothing to pip install and no terminal required.

Are you affiliated with Foundry, and do I need my own Nuke license?

No affiliation. Nuke is a trademark of Foundry, and this plugin is an independent third-party integration, not made, endorsed, or affiliated with Foundry. You need your own valid Nuke license from Foundry to run it. The plugin runs inside a copy of Nuke you already have the right to use and does not include or substitute for a Nuke license.

Why do I need a VFX Engine account?

Because generating spends real credits. Your account shows the exact cost before every render, refunds failed ones automatically, and keeps every result in your dashboard.

How is my account secured? Do I paste an API key?

No keys to paste. You sign in from the panel with a browser approval, the same account you use on vfxengine.com. The plugin stores a token on your machine so you stay signed in, and you can sign out from the panel at any time.

How are credits priced?

Every node shows its exact credit cost before you confirm, and you are never charged more than the number you approved. Prices are always live: the plugin fetches them fresh on every generate. Image nodes are a flat cost; Seedance video is priced per second by resolution, and a video-input run bills the input clip plus the output.

What happens if a render fails?

The credits it used are refunded to your account automatically, so you never pay for a failed generation. Completed renders are billed once and are not refundable.

Where do my results go, and are they stored?

Every finished generation drops into your script as a colorspace-tagged Read node, and the file is saved into a "vfxengine" folder next to your saved Nuke script, named VE_ plus the model and a timestamp (e.g. VE_seedance-20_20260714-171554_cdec58.mp4) so takes never overwrite each other and VFX Engine elements are recognizable anywhere they get copied. The plugin asks you to save your script before generating so that folder always exists. Server-side storage is a convenience, not an archive: we may delete stored generations at any time, so keep the local files (or download from your dashboard).

Can I use my own footage?

Yes. Wire your footage into a Seedance 2.0 node as the reference video and it keeps your camera move and action while restyling the look. See the examples for a full before and after.

How do I get more credits?

Buy them from the Buy credits button in the panel or on any node, or on this page and the credits section of the Nuke page. Purchased credits are valid for 12 months. Approved beta testers also get a monthly allowance of free credits to explore with.

Can I share my account or credits?

Credits are tied to your VFX Engine account. Keep your sign-in to yourself. If your team needs access, each person should have their own account.

Still stuck? Click Feedback on any node or in the panel: it sends your note straight from Nuke with the plugin version and the node's last status attached, so we can act on it immediately. Or email info@vfxengine.com. This plugin is in beta and we read every report.