Skip to content

Latest commit

 

History

History
54 lines (49 loc) · 3.57 KB

File metadata and controls

54 lines (49 loc) · 3.57 KB

Migrating from Difforum 0.x

Difforum 1.0 goes from 40 nodes to 33 and makes the Director timeline the main way to drive a shot. Nothing breaks on load: every 0.x node is still registered, hidden from the node search and moved to Difforum/legacy, so old workflows open and run. Legacy nodes will be removed in 2.0.

Node map

0.x 1.0
Anim Setup, Anim Setup+ Setup (seconds or frames, aspect, target model grid)
Film Director Director (timeline): separate Scenes / Camera / Energy tracks. It outputs the camera, energy and prompts directly (one direction wire) instead of text for other nodes to parse
Camera Keys Camera (keys), same syntax, and loop_mode included
Camera Shots Camera (keys) with transition 0 (hard cuts)
Camera Move (presets) Camera (keys) with a single key, or one Director block
Seamless Camera Camera (keys) with loop_mode
Camera (advanced) Camera (expressions), where each axis also accepts a schedule socket (e.g. an Audio Curve)
Camera Path Preview, Warp (2D/3D), Storyboard Storyboard (sheet + frames + camera path, same engine as the sampler), or Animatic for a previz video
Schedule / Sample Schedule Schedule
Schedule Info / Schedule Plot Schedule Plot (image + text)
Audio Analyzer Audio Analyzer (+ offset_seconds, a bands plot)
Audio Schedule (reactive) Audio Curve, or a per-block audio reaction in the Director
Prompt Schedule / Prompt Scenes Prompt Travel (2.5s: timestamps allowed) or the Director's Scenes track
Prompt Batch (→ AnimateDiff) Prompt Travel with build_batched
Feedback Sampler Feedback Sampler (see "behaviour changes" below)
Live Sampler, Live Step Live Sampler
Guide Builder (Wan VACE) Guide Frames (mask convention and hole fill selectable)
Loop Take, Loop Blend, Ping-Pong Loop (method)
Symmetry, Echo Trails, Detail Guard, Flow Stabilize same names, under Post
Model Profile, Model Catalog MODELS.md: recommendations that go stale belong in docs, not in nodes
Load Video, Save Video ComfyUI core Load Video / Create Video + Save Video (these keep audio)
VJ Look, Colour Grade, Glow, Glitch, Datamosh a dedicated grading / glitch pack; out of scope for a direction tool

Behaviour changes in the Feedback Sampler

  • 3D without depth used to freeze dolly / orbit moves silently. It now runs a pseudo-3D fallback and reports it.
  • 3D lateral moves are measured in pixels at the reference depth, the same unit as 2D pans. A 0.x 3D graph with translation_x will pan less; raise translation_scale if you relied on the old magnitude.
  • The depth map follows the image (depth_tracking = follow). Set static for the 0.x behaviour.
  • Cadence in-betweens are crossfaded between keys instead of being plain warps.
  • Steps scale with the energy (step_scaling = by energy): a frame at denoise 0.5 runs half the steps, like Deforum. Set fixed on Render Options for the 0.x cost.
  • Colour anchor defaults to scene. Set anchor_mode = first for 0.x.
  • The camera pose accumulates in view order (P_f = d_f · P_{f-1}). Poses from combined rotation + translation moves differ slightly from 0.x; per-frame deltas are unchanged.
  • Inputs: strength_schedule → strength, cfg_schedule → cfg_curve, positive_schedule → prompts. Colour, detail, symmetry, depth calibration and chunking moved to Render Options; without it the Director's style sets the look.