Runway's Ruby Rebuilds SDR Footage Into True HDR for Pro Workflows

Runway has quietly slotted a new model into its lineup that has nothing to do with generating video from a text prompt. Runway Ruby is a dedicated pipeline for taking standard dynamic range footage, whether captured on a camera or produced by another Runway model, and rebuilding it as true high dynamic range video suitable for professional finishing workflows.

The pitch is narrow but useful: feed Ruby an SDR clip up to 30 seconds long, and it returns a version with reconstructed highlights, deeper shadows, and a wider color gamut, exported in the container formats colorists actually work in.

What Ruby actually outputs

Ruby is aimed at the format side of the HDR problem, not just the tone curve. The model is available for Max plans and Enterprise, and can generate or convert any video into 16-bit EXR sequences or 10 and 12-bit ProRes and HEVC, in BT.2020 color with PQ or HLG. Those last two acronyms matter, because PQ (Perceptual Quantizer) and HLG (Hybrid Log-Gamma) are the two transfer functions that HDR delivery pipelines actually accept.

To put the bit-depth jump in context, SDR holds 6 to 10 stops of dynamic range, uses 8 bits per channel (24 bits per pixel), and peaks at around a few hundred nits. HDR formats stretch that to 12 to 17 stops and up to 10,000 nits of theoretical peak brightness. Ruby is trying to synthesize the missing information at the top and bottom of that curve rather than just remap what is already there.

Pricing and access

Ruby is exposed both in the Runway app and through the developer API. Ruby (ruby) converts SDR video to true HDR on POST /v1/video_to_hdr. Billing is straightforward and per-second:

• Credits can be purchased for $0.01 per credit in the developer portal for an organization.
• Ruby costs 20 credits per second of output at source resolution, which works out to $0.20 per second, or roughly $6 for a 30-second clip.
• Sources larger than 4 megapixels (roughly 4K) jump to 40 credits per second.
• Billing is based on output duration at the source's own resolution, so there is no upscaling premium baked in.

For comparison, Runway's flagship generative models like Gen-4.5 run 12 credits per second and Aleph 2.0 runs 28 credits per second, so Ruby sits in the middle of the pack cost-wise despite doing something structurally different.

Where the professional formats fit

Ruby is not the only way to get HDR out of Runway. These formats add a per-second surcharge on top of the model's rate. Available on Gen-4.5 and Aleph 2.0. When you generate directly with those models, you can request outputs like hdr_prores, hdr_png_sequence, or hdr_exr_sequence for an extra 20 credits per second (40 above 4MP). Ruby is the standalone conversion path for footage that already exists, whether it came from a camera or from a Runway generation that was rendered in SDR.

The reconstruction problem

Simple SDR-to-HDR conversion has existed for years in the form of inverse tone mapping: stretch the 8-bit signal into a 10-bit container, boost the highlights, and call it HDR. The results usually look wrong because the information was never there to begin with. SDR to HDR processes video with a wider dynamic range, restoring information that was lost in the original SDR format. Instead of stretching existing values, the model rebuilds the missing light data. The result is footage that holds up in editing, grading, and delivery.

Recent academic work frames this as a generative problem rather than a signal-processing one. A recent paper on generative HDR video describes the two failure modes such a model has to handle: a sunset river scene whose sky is fully clipped in SDR where the model synthesizes photorealistic warm-sky highlight detail, and an under-lit cave interior where the SDR signal is quantized and the human silhouette is lost against the background, where the model recovers wall texture and figure-ground separation from the quantized shadows. Clipped highlights and crushed shadows are exactly what a big generative video prior is good at hallucinating back in, because the model has seen millions of examples of what those regions probably should look like.

Why this matters for AI-generated pipelines

The more interesting use case is not upgrading legacy footage. It is closing the loop on AI-generated video. Nearly every text-to-video and image-to-video model on the market outputs 8-bit SDR, which is a dead end for anyone trying to intercut generated shots with real camera footage graded in HDR for streaming delivery. Ruby is Runway's answer to that problem, and the fact that it targets EXR sequences specifically signals the intent, because EXR is the format VFX and compositing pipelines use, not a consumer delivery format.

The broader landscape includes Topaz Labs' Hyperion, the dedicated AI model for SDR-to-HDR conversion that performs the inverse tone mapping that increases color depth, contrast, and peak brightness, and LTX Studio's own SDR to HDR feature. What Runway adds is native integration with a generative video stack, so a single-vendor pipeline from prompt to HDR master is now possible.

Practical considerations

Before dropping Ruby into a workflow, a few things are worth knowing:

• The 30-second maximum per generation means longer sequences need to be chunked, which raises the question of whether the model produces temporally consistent output across chunks.
• Output is at the source resolution, so upscaling has to happen as a separate step through Runway's video upscale endpoints or another tool.
• The API returns a single job via POST /v1/video_to_hdr, making it easy to slot into an existing render orchestration setup.
• Access is currently gated to Max and Enterprise plans on the app side, which limits hobbyist experimentation but reflects the professional target.

For teams building anything that needs to look at home on a modern HDR display, whether that is commercial spots, streaming content, or synthetic footage destined for a real production timeline, Ruby collapses what used to be a multi-tool, multi-vendor conversion step into a single API call. Whether the reconstructed highlights hold up under a colorist's scrutiny is the question that will determine if this becomes a standard step in AI video pipelines or a niche convenience.