Reference settings:
context_overlap: 1 128
context_size: 1 128,(1.1 model use max 72)
Total video frames > context_size
decode_chunk_size: 1 200
Video image dimensions must be the same

Video and image are scaled to the ratio: 576*1024

Tip:
Local test 1:
Reference parameters: Test Context Size 72, model 1 1, overlay 6, 576*1024, total video frames 90
VRAM usage: Sampling phase 35167MiB, Decoding phase 48639MiB
The generated quality is indeed high! VRAM usage is indeed large
Test device: 4090 48G fp32
Time consumption approximately 12 minutes

Website online test 2:
Fine-tune changed to BF16 4090 24g, frame extraction 2, total frames 73 (meaning video length >= 73*2/30)
Generated video frame interpolation, frame rate 30
Online runtime: 04:43

Local test 3:
Reference above, frame extraction changed to 1
(Local VRAM usage reference - Sampling phase: 18049MiB, Decoding phase: 33889MiB, Frame interpolation phase: 48641MiB, generation time close to 9 minutes)

During testing, Keep model load was enabled. To reduce VRAM usage, it is recommended to disable it during use and clear memory nodes after sampling.