Nodes/Difforum/Difforum · Guide Frames
ComfyUI Node

Difforum · Guide Frames

Warp one anchor image along the Director's camera path. Produces a guide video whose motion is exactly the camera you drew, for video models that take control/guide frames (LTX guides, Wan VACE, H3 first/last frame). Revealed areas are filled with neutral gray and marked in the mask, using the convention the target expects.

By chillithebillis·Created 3 months ago·Updated about 7 hours ago· 8
Difforum · Guide Frames
  • anchor_image
  • direction
  • params
  • camera
  • depth
  • guide_frames
  • masks
  • info
◄mask_convention1 = generate (VACE / LTX / H3)►
◄hole_fillgray►
◄translation_scale1.00►
CategoryDifforum/5 · Video model bridges

Inputs (8)

NameTypeDefaultDescription
anchor_imageIMAGEThe image the camera travels over (usually your first frame).
mask_conventionCOMBO1 = generate (VACE / LTX / H3)Which value the mask uses for 'generate': 1 = generate (VACE / LTX / H3 / Fill Reveal) or 1 = keep.
hole_fillCOMBOgrayWhat goes into the area the camera uncovers before AI fills it: gray (neutral, best for Fill Reveal and H3), stretched edge, or black.
directionoptDIFFORUM_DIRECTIONThe Director's wire: camera, energy curve, scene prompts, keys and look in one cable.
paramsoptDIFFORUM_PARAMSFrom Setup: frame count, fps, width and height. Plug the same Setup everywhere.
cameraoptDIFFORUM_CAMERAA camera track from Camera (keys), Camera (expressions) or Camera Import. Overrides the Director's camera.
depthoptIMAGEDepth of the anchor, for real parallax in 3d mode.
translation_scaleoptFLOAT1.000–20Strength of 3D translation vs. the depth range.

Outputs (3)

NameTypeDescription
guide_framesIMAGE—
masksMASK—
infoSTRING—