Runway Ruby converts any SDR video to 16-bit HDR ProRes or EXR, and it works on everything you have already generated
Runway's new Ruby model upgrades standard-dynamic-range footage to 16-bit HDR, exporting in ProRes or EXR formats, and it accepts any video already uploaded to the platform or generated there — up to 30 seconds long.
Runway has quietly shipped something that belongs at the colour-grading stage rather than the generation stage, and that distinction matters. Ruby is a new model that takes ordinary SDR footage and converts it to 16-bit HDR, delivering the result as either a ProRes or an EXR sequence. It is available now, and it works on any video you have already uploaded to Runway or generated there, provided the clip is 30 seconds or shorter.
For anyone trying to get AI-generated footage into a professional finishing pipeline, this is a meaningful step. The persistent complaint about AI video has not just been resolution or motion quality — it has been that the output lives in an SDR container that flattens badly on HDR displays and falls apart the moment a colourist tries to push the grade. ProRes and EXR are the languages that finishing suites speak. Ruby answers in those languages.
What Ruby actually delivers
The output formats are the headline. ProRes is the standard delivery format for editorial and colour work in DaVinci Resolve, Avid, and Final Cut Pro. EXR sequences are the currency of compositing: if you are taking AI footage into Nuke, After Effects, or Flame for any serious visual effects work, EXR is what you need. The fact that Ruby produces both means the footage can travel further down a professional pipeline without conversion steps that eat dynamic range before the work even begins.
The 16-bit depth is also significant. Most AI video generation happens in 8-bit, which gives 256 steps per channel. Sixteen-bit gives 65,536 steps per channel, which is what you need when a colourist wants to recover highlights or push shadows without banding. Whether the model is genuinely reconstructing scene-referred luminance data or performing a learned tone-mapping is not something Runway has specified in their announcement, and that distinction matters for professional use.
What has not been said
Runway has not detailed the underlying approach. There is a meaningful difference between a model that hallucinates plausible HDR headroom from an 8-bit clip and one that recovers information the generation model implicitly encoded. The former is useful for display and delivery; the latter would be genuinely novel. No independent benchmarks exist yet, and Runway's own announcement gives no quality metrics or comparison frames.
Pricing for Ruby has not been announced as a separate line item. It is not clear whether conversions consume standard Runway credits, count against generation limits, or are priced per second of output. For studios running high volumes of clips through a finishing pipeline, that number matters considerably.
The 30-second cap is the other constraint to note. Many AI generation models top out at around 10 seconds per clip, so for individual shots this is not a problem. But if you are stitching longer sequences and hoping to run the finished assembly through Ruby in one pass, you cannot. You will need to process shot by shot, which adds pipeline steps.
How this sits against the alternatives
Topaz Labs' Hyperion 2.5 was covered here recently and serves a related purpose: converting AI footage to true HDR, now also available via fal for API integration. The key difference is delivery path. Hyperion is a standalone tool and a fal API endpoint, meaning it can sit anywhere in a custom pipeline regardless of which generation platform produced the footage. Ruby is native to Runway and applies only to content that lives inside Runway's ecosystem. If your workflow spans multiple generation tools, Hyperion's platform-agnostic position may be more useful. If you live inside Runway already, Ruby removes a round-trip export step.
Black Forest Labs' FLUX Video Upscale, also recently covered, addresses resolution rather than dynamic range. These tools are solving adjacent but distinct problems. A complete finishing chain for AI footage is starting to look like: generate, upscale to 2K or 4K, convert to HDR, export to ProRes or EXR. Ruby collapses the last two steps for Runway users.
What to test first
If you have Runway access, the most useful first test is not a showreel shot but a worst-case one: a clip with a bright sky and deep shadow in the same frame. Run it through Ruby and bring the ProRes into your colour suite. Pull the highlights. Push the shadows. How much headroom did the model create, and does it hold up under a grade, or does it posterise? That answer will tell you whether Ruby is a delivery tool or a genuine creative tool. The difference determines where it earns a permanent place in the pipeline.