I am trying to find out why our (cinelerra-gg) ffmpeg integration refuses to load filtergraph < and crashes and burns! > if I set hw decoding to any value (it works for sw decoding). So far I found this example (not accepted) on ffmpeg-devel, but I am still not sure how to apply ot for our case (we use av_buffersink_get_frame() ) https://ffmpeg.org/pipermail/ffmpeg-devel/2018-August/233555.html In general, I hope to get hw decoding (now via vulkan ) + libplacebo doing tonemapping purely on GPU *before* frames ingested into cingg for further (mostly software/host) manipulation, display and encode. But I guess for this I need to pass around hw_device_context created for decoder for this media/input file? (our filtering options are either global or per-file) filter_graph->filters[i]->hw_device_ctx = av_buffer_ref(hw_device_ctx); Paul, your lavfi-preview does not have hw filters previews yet, or I read sources too diagonally?