Generated reference · ComfyUI workflow · motion

GIMM-VFI Interpolate

Raise a clip's frame rate by generating intermediate frames with GIMM-VFI. The audio track is carried through GetVideoComponents and re-muxed by CreateVideo, so a clip with sound does not lose it.

motion6nodesGPU H100out unspecified~120sv1.0.0
workflow.json
On this page
6Graph nodes
6Node classes
6Inputs
4Models

Tags: motiongimm-vfiframe-interpolationrunpod-serverlessvolume-backed

Inputs (6)#

The typed parameter surface callers bind when they request this workflow. Enum options and numeric bounds are the values the workflow document declares.

source_videovideorequired
The clip to interpolate. Its audio, if any, is preserved.
source_fpsfloatdefault 16.0min 1.0max 60.0
The frame rate of the source clip. The output rate is this times the multiplier, which is what keeps the clip's real duration unchanged; declaring it wrong changes the speed, not the content.
multiplierenumdefault 2
How many output frames per input interval. 2 inserts one frame between each pair. GIMM-VFI produces 1 + (N-1) x multiplier frames from N inputs, so the output rate is source_fps x multiplier.
options: 234
modelenumdefault gimmvfi_f
Which flow estimator. _f is FlowFormer, _r is RAFT; both are the arbitrary-timestep LPIPS fp32 checkpoints.
options: gimmvfi_fgimmvfi_r
ds_factorfloatdefault 1.0min 0.01max 1.0
Downsample factor for the flow estimate. 1.0 is full resolution; lower trades accuracy for VRAM on a large frame.
seedintegerdefault 0min 0
GIMM-VFI takes a seed; the same inputs and seed give the same frames.

ComfyUI node graph (6)#

The executable ComfyUI prompt graph: 6 nodes across 6 distinct node classes, wired by 6 data dependencies. Nodes tinted green come from a custom node pack this workflow declares; the rest are ComfyUI core / baked-community classes.

flowchart TD n0["1: LoadVideo"] n1["2: GetVideoComponents"] n2["3: DownloadAndLoadGIMMVFIModel"] n3["4: GIMMVFI_interpolate"] n4["5: CreateVideo"] n5["6: SaveVideo"] n0 -->|video| n1 n2 -->|gimmvfi_model| n3 n1 -->|images| n3 n3 -->|images| n4 n1 -->|audio| n4 n4 -->|video| n5
Data-flow DAG — scroll to zoom, drag to pan.

Nodes (6)#

1LoadVideocore
file = {{source_video}} tmpl
outputs: VIDEO
2GetVideoComponentscore
video = ◂ node 1 · out[0]
outputs: IMAGEAUDIOFLOATCOMBOCOMBO
3DownloadAndLoadGIMMVFIModelcore
model = {{model_map[model]}} tmplprecision = fp32
outputs: GIMMVIF_MODEL
4GIMMVFI_interpolatecore
gimmvfi_model = ◂ node 3 · out[0]images = ◂ node 2 · out[0]ds_factor = {{ds_factor}} tmplinterpolation_factor = {{int(multiplier)}} tmplseed = {{seed}} tmpl
outputs: IMAGEIMAGE
5CreateVideocore
images = ◂ node 4 · out[0]fps = {{source_fps * int(multiplier)}} tmplaudio = ◂ node 2 · out[1]
outputs: VIDEO
6SaveVideocore
video = ◂ node 5 · out[0]filename_prefix = gimm-vfiformat = mp4

Parameter banks (2)#

The prompt / configuration lookup tables this workflow keys into from its inputs — the vocabulary that turns a style / palette / preset selection into graph parameters.

model_map (2)#

gimmvfi_f
gimmvfi_f_arb_lpips_fp32.safetensors
gimmvfi_r
gimmvfi_r_arb_lpips_fp32.safetensors

requires_families (1)#

gimm-vfi

Models & dependencies#

Custom node packs (1)#

The non-core ComfyUI node classes this graph requires; the RunPod worker image the workflow runs on must bake or install a pack that provides every one of them.

ComfyUI-GIMM-VFI

Models required (4)#

gimmvfi_f_arb_lpips_fp32.safetensors
gimmvfi_r_arb_lpips_fp32.safetensors
raft-things_fp32.safetensors
flowformer_sintel_fp32.safetensors

Output contract#

What a successful run of this workflow returns.

primary
{"type": "video", "format": "mp4", "codec": "h264", "fps_source": "derived", "audio": true, "alpha": false, "description": "The interpolated clip at source_fps x multiplier, with the source's audio re-muxed."}

Taxonomy & routing#

How the control plane classifies this workflow — from the committed workflow-taxonomy-registry.json. It drives the consistency / control surface the agentic director can exercise over the workflow.

assetFamily
restored-video-master
outputPackageProfile
video-master-profile
controlModalities
model-lockseed-locktemporal-lock
consistencyDimensions
motion
notes
GIMM-VFI frame interpolation. temporal-lock is the only real control: the graph synthesises frames BETWEEN existing ones and changes nothing else, so identity, lighting and environment are inherited from the source rather than held by this workflow. No prompt-template-lock or sampler-scheduler-lock because there is no diffusion and no sampler.