One Grade, Three Deliverables
Leah is delivering a six-minute product film from a Premiere Pro 2024 project. Deliverables: a 3840x2160 HDR10 master for YouTube, a 1920x1080 SDR file for a client's internal broadcast, and a 1080x1920 vertical cutdown. The sequence working colour space is Rec.2100 PQ, the footage is Sony S-Log3 originating in Rec.2020, and the titles are #FFFFFF MOGRTs built in After Effects in Rec.709. On the HDR reference monitor the white titles look dim grey; in the SDR export they clip while the shadows crush. The HDR master, exported as 8-bit H.264, shows banding across the product's gradient background.
1. Why do the white titles look grey in HDR and clipped in SDR, and what would you change?
Read the worked answer
Both symptoms come from one question: what does white mean in HDR?
- In a PQ (Rec.2100) sequence, peak white is 1000+ nits but graphics white is deliberately mapped lower. Premiere exposes this as Graphics White Luminance in the sequence's colour management, defaulting to 203 nits per BT.2408. A #FFFFFF Rec.709 MOGRT therefore lands at 203 nits, which looks grey beside 1000-nit specular highlights. That is correct behaviour, not a bug.
- In the SDR export, PQ values are not tone-mapped by default, so everything above 100 nits clips. Titles blow out and shadows crush because the low end of the PQ curve is squeezed into a narrow SDR range.
Fix: decide the intent first. Titles should sit at graphics white for HDR legibility - do not push them toward peak just to look bright. For SDR, either enable tone mapping in Export Settings and review on a Rec.709 display, or build a separate 709 sequence with a colour-space conversion and re-check the titles there. One preset cannot serve both.
2. Specify the export settings for the HDR10 master and the SDR broadcast file, and address the banding.
Read the worked answer
HDR10 master: HEVC, Main10 profile, 10-bit, Rec.2100 PQ, with HDR10 metadata enabled (mastering display, MaxCLL/MaxFALL). Without the metadata, players fall back to a flat, washed-out render. VBR 2-pass at roughly 60–80 Mbps for 2160p, Maximum Render Quality on, and confirm no tone mapping is applied. A ProRes 422 HQ master uploaded to YouTube is also legitimate and avoids the encode question entirely.
SDR 1080p: H.264 High, 8-bit, Rec.709, tone mapping on, roughly 16–20 Mbps VBR 2-pass.
Banding: an 8-bit H.264 encode of a 10-bit PQ gradient will band - that is the codec, not the grade. Use 10-bit HEVC wherever HDR is retained, raise the bitrate, and if the SDR file still bands, add 2–4% fine grain before export to dither the gradient. If the internal broadcaster accepts ProRes 422, take that route and stop fighting the codec.
3. How would you produce the vertical cutdown, and what are the hidden risks in that workflow?
Read the worked answer
Reframe rather than letterbox. A 16:9 master inside 9:16 wastes most of the frame and shrinks faces.
- Duplicate the sequence and set it to 1080x1920.
- Run Auto Reframe as a first pass, not a final one: Subject tracking for interviews, Slower for locked-off shots, Faster for UI footage. It writes Motion > Position keyframes, and on wide shots it will happily follow the wrong subject.
- Manually keyframe any shot it gets wrong. A cut is always more acceptable than a drifting reframe.
- Re-check safe areas: lower-thirds and titles built for 16:9 sit outside a 9:16 frame, so the MOGRTs need repositioning or a dedicated vertical version.
- Check sharpness honestly. A 9:16 crop of 3840x2160 is 1215x2160 scaled down to 1080x1920 - you are not upscaling, but any shot already pushed past about 110% in the 16:9 cut will soften visibly once reframed. Evaluate a frame at 200% before committing the whole cutdown.