Glownet3 · Production-grade fidelity engine for Netflix Studio productions
Glownet expands color gamut, dynamic range, and bit depth — and optionally upscales — in a single pass. It is built for productions that need to bring lower-fidelity material (AI-generated frames, 8-bit sRGB, heavily compressed plates) into a production-grade ACES color-managed pipeline.
What Glownet does
Given an input image, sequence, or video, Glownet:
- Cleans compression artifacts
- Upsamples bit depth (8 → 16 bit), recovering tonal information from quantized inputs
- Upscales resolution up to 4× (e.g., 1K → 4K)
- Conservatively augments input details
- Expands dynamic range by up to 5 stops
- Expands color gamut from Rec.709 to ACES AP0 (the entire visible spectrum)
- Outputs in scene-referred ACES space (EXR sequences)
Glownet is built around an ACES-native training pipeline. Most generative models are trained in sRGB and bolt color management on afterward; Glownet does color management inside the model. The result is Delta E 1.6 color accuracy — close to the perceptual identity threshold of 1.0.
When to use Glownet
- AI-generated material that needs to enter a production ACES pipeline
- 8-bit source material (archive, internet finds) that must live alongside camera-original footage
- Heavily compressed footage where artifacts are visible in the grade
- Shots that need both upscale and fidelity work in one pass (do not chain NetVSR → Glownet — Glownet does both)
- VFX final shot work in an ACEScct or ACEScg pipeline
When NOT to use Glownet
- You only need a pure upscale and want to keep your existing color space → use NetVSR(2) or NetVSRLite
- You need a quick PNG or MP4 to share or review → use NetVSR with the General preset
- You need to push beyond 4× input resolution → use NetVSR2, up to 4K output resolution
Inputs, Outputs, and Cautions
This is the reference panel shown at the top of the Glownet tool page.
Inputs
- 8-bit sRGB or Rec.709
- Image, Video, or Sequence[Image]
- Not working yet: DPX
Outputs
- ACES2065-1, ACEScg, or ACEScct
- Sequence[EXR]
Cautions
- Keep inputs/outputs below 16K resolution (slow and OOM risk above)
- Avoid very large or long jobs — e.g. 10GB+ input files or 1000s of frames
- A typical shot (5–10 seconds, ≤500 frames) is fine
- Long jobs hog the queue and may fail
Before you start
Required permissions
You need at least write permissions on a Content Hub title before you can use Glownet. The app operates scoped to that title; outputs land in that title's Workspace.
- Glownet is only available as a limited release to a handful of production during the June - August 2026 Pilot.
- If you are working on a real production, your VFX supervisor or production coordinator will already have a movie ID. Make sure you are onboarded to that title.
Required tools
- Aspera must be installed for Content Hub. Required for secure upload and download.
- DaVinci Resolve, Nuke, or another VFX-grade EXR viewer. Do NOT use macOS Preview to view EXR output — it misrenders the colors and you will think the tool broke your image.
- A modern web browser.
Job sizing guidelines
- A typical shot (5–10 seconds with up to ~500 frames) is fine.
- Avoid very large files (10GB+) or very long sequences (thousands of frames). They tend to hog the queue and may fail.
- If you have a 30-minute clip, split it before submitting. Glownet works great on lots of small jobs, less well on one giant job.
- Maximum upscale factor is 4×. You cannot reach 16K from an 800×600 source.
Step-by-step walkthrough
Step 1: Open the app and select a title
Go to the Tools Library inside of the Content Hub.
If the URL already has a movieId parameter, the title is preselected. Otherwise, pick a title from the dropdown. You need write permissions on the title in Content Hub.
If you cannot see your title: You probably are not onboarded to it yet. Ask your VFX supervisor or production coordinator to request glownet access inside the Tools Library.
Step 2: Upload your asset
Two ways to provide an input. Use whichever fits your workflow.
From Workspace (already in Content Hub):
- Click the From Workspace tab.
- Browse to the file in your title's CDrive workspace.
- Select it.
File Upload (from your local machine):
- Click the File Upload tab.
- Drag and drop your file or click to browse.
- Accepted file types: .jpg, .jpeg, .png, .tiff, .tif, .mp4, .mov.
- Click the Upload button. The file uploads to CDrive and is selected automatically.
Don't forget to click Upload. After dragging your file in, you must click the Upload button. Without that, the file isn't actually sent to CDrive and the tool won't see it.
Step 3: Set Optional Parameters
Once your file is selected, the Optional Parameters and Export Settings sections become editable.
- Output File Name — defaults to a date-time stamp. Override it to give the output a meaningful name.
- Output Folder Name — defaults to an alphanumeric folder name. Override it to organize your outputs.
Step 4: Choose Export Settings
| Setting | Options | What it means |
| Preset | VFX, Interchange | VFX gives production-grade defaults. Interchange is good for handing off between tools. General Purpose is most permissive. |
| Colorspace | ACEScct, ACEScg, ACES2065-1 | See the color space guide below. |
| File Format | EXR | Glownet only writes EXR. PNG/MP4 are not options by design — they would discard the fidelity Glownet just added. |
| Bit Depth | half (16-bit), full float (32-bit) | Half is standard for VFX and produces smaller files. Full float is for cases where you cannot afford any rounding. |
Which ACES variant should you pick?
- ACES2065-1 — master interchange. Use this when handing off between tools and stages.
- ACEScct — working space for color grading. Use this if you are about to grade in Resolve.
- ACEScg — working space for compositing. Use this if you are going into Nuke for comp.
Why Glownet won't write a PNG: PNG is 8-bit and would clip the wide gamut and dynamic range that Glownet just added to your input. The tool blocks this by design.
Step 5: Generate
Click the Generate button in the bottom right corner. The first job in a while may take a few minutes to start because the GPU has to warm up; subsequent jobs are faster.
Why the first job is slow: Glownet runs on cloud GPUs that aren't always warm (this is cost-controlled). The first request after a 5-minute idle window pays a cold start of 1–12 minutes while the GPU is found, acquired, and set up. Once warm, subsequent requests are fast. If your job seems stuck for less than a minute, just wait.
Step 6: Track your job
Click Jobs in the top right to open the Jobs Tracker. Status colors:
- Yellow — running
- Green — succeeded
- Red — failed
Job history is persisted per title. You can refresh the page, close the tab, come back later, and the tracker still shows all your runs.
If your job is broken, click the rightmost icon next to the job to copy the job id. Share that with Support when reporting the failure.
Step 7: Retrieve your output
Three ways to grab your output. Pick whichever fits your workflow.
- From the Jobs Tracker: click View in ContentHub or View in CDrive.
- From the bottom of the tool page: expand the Generated Outputs accordion and download from there.
- From your title's Workspace in Content Hub directly: outputs live in the folder you named in Optional Parameters.
Open EXR files in a VFX-grade viewer. EXR files contain wide gamut and high dynamic range. macOS Preview misrenders them, so the colors will look wrong. Open EXR output in DaVinci Resolve, Nuke, or your usual VFX viewer.
Frequently asked questions
Why is my first job taking so long?
The cloud GPU is cold-starting. After ~5 minutes of idle, the GPU spins down to save cost. Your next job pays a 1–12 minute cold start. Subsequent jobs within the keep-alive window are fast.
My EXR looks wrong when I open it.
Are you opening it in macOS Preview? Preview misrenders EXR. Open it in DaVinci Resolve, Nuke, or another VFX-grade viewer.
If you believe that the EXR output is missing color tags or has the wrong metadata when viewed in Nuke etc, please report this bug to Support!
Why does Glownet refuse to give me a PNG?
Because PNG is 8-bit and would clip the wide gamut and dynamic range Glownet just produced. The tool blocks PNG by design.
What's the maximum upscale factor?
4×. So 800×600 cannot reach 16KHow many jobs can I run at once?
Glownet runs one concurrent request per GPU due to memory constraints. If you submit multiple jobs, they queue FIFO and may scale up to a few GPUs depending on configuration. Heavy back-to-back jobs may hog resources for everyone, so please be considerate with big jobs.
My job failed. Now what?
Open the Jobs panel and find the failed run. If you are a @netflix.com user, click the rightmost icon to copy the Studio Orchestrator link. Send that link plus a brief description of what you were trying to do to the support channel. If you submitted a very large job (more than ~500 frames), try splitting it before resubmitting.
Where does my output file live?
In your title's Content Hub Workspace, in the folder you named under Optional Parameters. You can also retrieve it from the Jobs Tracker or from the Generated Outputs accordion at the bottom of the tool page.
Is this tool free? Yes during the pilot. There is no per-seat licensing cost to productions.
Common pitfalls
Viewing EXR output in macOS Preview.
Colors look wrong, washed out, or oversaturated. Fix: open the file in DaVinci Resolve, Nuke, or another VFX-grade viewer.
Expecting Glownet to write PNG.
PNG is 8-bit; Glownet's output is 16-bit ACES. The tool refuses by design. Use NetVSR with the General preset if you need PNG.
Uploading a 2-hour clip.
Jobs fail partway through, often during OCIO conversion. Glownet is sized for typical shots — split before submitting.
Hard-clipped highlights.
Glownet cannot invent content that was clipped at capture. The model softens clipped areas but cannot reconstruct missing data.
Output looks correct in Resolve but wrong in Nuke.
Check color management config in both software. Most common: Nuke is expecting ACEScct and your output is ACES2065-1. Pick the export format that matches your downstream tool.
First job feels slow.
Cold start. Wait — the next job within 5 minutes will be fast.
Upload didn’t get to Cdrive
Did you forget to click Upload in the UI? Remember to click the Upload button explicitly.
Trying to upscale beyond 4×.
Glownet3’s builtin upscaler is capped at 4× input resolution. It cannot produce output larger than 4 times your input resolution.
Getting Support
The fastest path to a fix
- Open the Jobs Tracker and find the failing job.
- Copy the Job ID
- Click on the Support button
- Submit the form for support including your job id