Standards & Conformance¶
Transduce is built to slot into an existing professional pipeline, not to ask a facility to trust a black box. This page is the short version of that case, for a VFX supervisor, colorist, or post-production lead deciding whether Transduce belongs in a real delivery chain: which industry standards and conventions Transduce follows, why each one matters to your pipeline specifically, and where to go for the full technical detail — real numbers measured against the shipped code, not marketing claims.
Every claim below links to a page written in that same spirit: specific standards bodies and specification numbers, exact conversion matrices or code paths, and where possible, actual measured deviation figures from running the shipped encoder — not a description of what it's supposed to do.
Colour management — ACES, not a proprietary space¶
Every clip is converted to ACEScg — the Academy Color Encoding System's own scene-linear working space (SMPTE ST 2065-1) — before any grading, effects, or AI processing touches it, and converted back out to whatever a given delivery needs. This is the same working space already in use across professional VFX and colour pipelines, so footage round-trips cleanly with Nuke, Resolve, Flame, or any other OCIO-aware tool a facility already runs — Transduce isn't introducing a colour space anyone downstream has to learn or work around.
A full camera-original ACES 2065-1 production EXR, round-tripped through decode → ACEScg → re-encode → re-decode, was measured directly against its own original pixel data: maximum absolute deviation 0.000244, mean absolute deviation 7.4 × 10⁻⁷ — the resolution limit of 16-bit half-float storage itself, not accumulated error.
→ The Colour Science Behind Transduce — full conversion matrices, the 8-step processing pipeline, and the complete round-trip measurement.
Embedded alpha — the convention professional tools actually assume¶
Alpha-carrying formats (OpenEXR, CineForm 444 Alpha, Uncompressed 4:4:4:4 Alpha) default to premultiplied alpha — the convention OpenEXR's own documentation calls "the norm," and what Nuke, Resolve/Fusion, and Flame all assume with zero per-clip setup. Straight alpha stays available as an explicit choice for garbage/holdout-matte work, where the convention genuinely is different — Transduce doesn't force one convention onto a use case it doesn't fit.
Measured directly against the shipped OpenEXR encoder: Premultiplied
output matches true_RGB × alpha to within 0.000195 (16-bit
half-float precision limit); Straight-mode RGB is verified
alpha-independent — zero variance across a full alpha sweep from 0 to
1, not just a claim about which code path runs.
→ Alpha Channels — Straight vs. Premultiplied — the standards behind the convention, why a straight-alpha file can look "cut out" in a compositor even when nothing is wrong with it, and the full measured validation.
Resize & upscale — deterministic, and honest about the AI path¶
The default resize path (Fast) is entirely deterministic math — documented resampling kernels, exact anti-alias formulas, full float32/negative-ACEScg precision preserved end to end — the same kind of predictable, repeatable behavior a DI pipeline depends on. The optional AI path (Enhanced, SPAN super-resolution) is opt-in, never silently substitutes for Fast if its model file is missing, and round-trips through the exact same ACEScg⇄Rec.709 matrices documented on the colour science page rather than a separate, undocumented conversion.
→ How Transduce Resizes (and Upscales) Your Footage — the full geometry, filter-by-filter behavior, and how the SPAN cascade actually works.
Delivery formats — conventions a facility already expects¶
- OpenEXR — floating-point, no baked-in curve, the VFX/compositing
and archival standard; optional ACES 2065-1 delivery writes the
acesImageContainerFlag(SMPTE ST 2065-4) so ACES-aware tools auto-detect it correctly. - DPX — Cineon log over ACES 2065-1 by default (the standard, expected DI/VFX encoding), correctly packed per the SMPTE bit-depth spec. High-bit-depth mattes are delivered the real industry-standard way — a second DPX sequence, not a channel most tools won't reliably read at that depth.
- CineForm / Uncompressed — post-house delivery tiers with the same Alpha Type conformance as EXR where they carry alpha.
→ Export Formats for the full list, or the Quick Reference if you just need to know which one to pick for a given handoff.
Open-source transparency¶
Every third-party dependency Transduce ships — PySide6, NumPy, OpenEXR, colour-science, and the rest — is listed with its exact licence and the notice text that appears in the app's own Credits screen. Nothing is undisclosed.
What this page doesn't claim¶
In the same spirit as the pages it links to: Transduce implements ACEScg as a linear working and interchange space with matrix-based conversion on ingest and delivery, but does not currently apply the ACES Reference Rendering Transform (RRT) or an Academy-defined Output Device Transform chain — there's no filmic view-transform curve in the loop today. Current scope targets SDR delivery without a creative rendering transform; a full RRT+ODT pipeline for VFX and scene-linear workflows is planned as a subsequent addition, not assumed to already exist. If a specific standard or workflow you rely on isn't covered above, that's worth telling us directly rather than guessing at fit.