LingBot Post-Process Settings

The pack's published workflows end with two optional upgrades: RIFE frame interpolation (smoother motion by inserting frames) and FlashVSR spatial upscaling (2× resolution bump). Both are heavy - RIFE loads a flownet, FlashVSR loads a whole model - so you don't want them running when you're just smoke-testing. LingBotPostProcessSettings is the control panel that decides, with two booleans and an FPS math helper, and it hands its decisions to the branch switch (LingBotLazyImageSwitch) so disabled branches don't execute at all.

This is a settings node in the same family as LingBotGenerationSettings: inputs in, scalar outputs out, nothing else happens here. The real work (interpolation/upscaling) happens in nodes from other packs - ComfyUI-VFI for RIFE and ComfyUI-FlashVSR_Ultra_Fast - which is why the install section matters more than usual.

Inputs and what they compute

• source_fps (default 20) - your generation FPS. This is the base the multiplier acts on.
• enable_rife (boolean, default false) - master switch. Off means RIFE is skipped completely, per the tooltip.
• rife_fps_multiplier (default 2.0, 1.0–4.0, step 0.25) - the tooltip example: "2.0 converts 20 FPS to 40 FPS." Target FPS is clamped at 240, so source_fps × multiplier beyond that is a hard error.
• enable_flashvsr (boolean, default false) - the spatial upscale switch.

Outputs: rife_enabled, rife_target_fps (source × multiplier), flashvsr_enabled, and output_fps (the target if RIFE is on, else the source FPS). Those feed two places: the enabled booleans into LingBotLazyImageSwitch, and the FPS values into the VHS video-combine node so the saved MP4 is stamped with the post-processed frame rate.

How the pipeline is ordered

When both are on, order is FlashVSR first, then RIFE - the README is explicit. That ordering makes sense: upscale spatially first, then interpolate temporally, so RIFE isn't asked to invent frames between half-res images. Both default to disabled and execute lazily; FlashVSR additionally needs its model folder (ComfyUI/models/FlashVSR) installed per the upstream README before it can run at all.

Install - this one actually needs three packs

cd ComfyUI/custom_nodes
git clone https://github.com/ALX-CODE/lingbot-video-1.3b-fp8
pip install -r requirements.txt   # LingBot pack itself

But this node's dependencies are separate packs:

• ComfyUI-VideoHelperSuite - MP4 output (needed regardless).
• ComfyUI-FlashVSR_Ultra_Fast - the 2× upscaler, plus its FlashVSR model folder.
• ComfyUI-VFI - RIFE; it downloads flownet.pkl on first use, or you can drop it in ComfyUI/models/rife for offline use.

Install all three via ComfyUI Manager or by cloning into custom_nodes and installing requirements. And here's the trap the README warns about: all three must be installed for the workflow files to even open, even when their toggles are off - otherwise you get missing-node errors before you've generated a single frame.

Gotchas

• output_fps isn't the "final video FPS" if you skip post - it's just source_fps passed through. Don't be surprised.
• RIFE at 4× from 20 FPS gives 80 FPS and 4× the frames - watch VRAM and disk; interpolation is not free even with lazy evaluation once enabled.
• Keep source_fps matched to the actual generation FPS (i.e. what LingBotGenerationSettings outputs), or the frame-rate math in the saved file will be a lie.
• If a post-processed branch errors, read the error before assuming the switch is broken - most often it's a missing FlashVSR model folder or a flownet that couldn't download.