Here's kimi-k2-thinking with the reasoning block included: https://www.kimi.com/share/19bcfe2e-d9a2-81fe-8000-00002163c...
F1 af toggle "loudnorm=I=-14:LRA=1:tp=-1:linear=false:dual_mono=true"
The I=-14 is the target 'volume', which makes it as loud as YouTube and Spotify, setting it to -23 will be much quieter, but gives you more dynamic range. LRA=1 goes from 1-50, with 1 being the most compressing which is probably what you want in this case. TP=-1 sets the limiter to -1dB, which is plenty of clipping threshold because loudnorm calls the resampler for 192kHz before working in dynamic mode. Linear=false because we use dynamic mode here and cannot even use linear mode. Dual_mono=true makes -14 as loud for mono as stereo sources. F1 af toggle "pan=stereo|c0=FC+LFE+FL+BL+SL|c1=FC+LFE+FR+BR+SR,loudnorm=I=-14:LRA=1:tp=-1:linear=false:dual_mono=true"