ComfyUI Node
Morpheus Dream 🌙
One dream of a Morpheus storyboard. Chain several (optional 'shots' input) and feed the chain to 'Morpheus (Video Sampler)'. Keyframes are optional: with link = continue a shot starts from the previous shot's last generated frame.
Morpheus Dream 🌙
- shots
- start_frame
- end_frame
- shots
- info
◄prompt—►
◄duration5.2►
◄linkcontinue►
◄seed_offset0►
◄keyframe_strength0.999►
◄refineauto►
CategoryKinburg-Nodes/Bestiary/Morpheus
Inputs (9)
| Name | Type | Default | Description |
|---|---|---|---|
| prompt | STRING | What happens in THIS shot. With both keyframes wired, describe the transition between them. | |
| duration | FLOAT | 5.20.2–15.1 | Shot length in seconds. Snapped UP to the model's frame grid — 17k+5 frames at 24 fps, i.e. steps of 0.71 s: … 4.46, 5.17, 5.88, 6.58 … s. The trained range is 124-362 frames = 5.17-15.08 s; anything shorter still runs but is out of distribution. The 'info' output shows the length you actually get. |
| link | COMBO | continue | How this shot starts when NO 'start_frame' is wired: • continue — inherit the last generated frame of the previous shot (a continuous take; motion still resets slightly at the seam, since the model only sees a still frame). • cut — start from nothing: text-to-video. A hard montage cut. A wired 'start_frame' always wins over both. |
| seed_offset | INT | 00–4294967295 | Added to this shot's seed. Bump it to re-roll THIS shot only — it changes the shot's cache key, so every earlier shot still replays from cache (later ones re-sample, because their first frame changes). |
| shotsopt | KINBURG_MORPHEUS_SHOT | Wire the PREVIOUS 'Morpheus Dream' here to chain them. Left→right = shot order. Leave empty for the first shot. | |
| start_frameopt | IMAGE | First frame of this shot. Overrides 'link'. NOTE: it is STRETCHED to the canvas (no crop), so its aspect ratio should match — the sampler's 'canvas' = auto takes the canvas from this image. | |
| end_frameopt | IMAGE | Last frame of this shot — the target the shot moves toward (cover-cropped to the canvas). The next shot still inherits the GENERATED tail frame, not this image, so the two stay visually identical. | |
| keyframe_strengthopt | FLOAT | 0.9990.9–1 | How hard the model is held to its keyframes (the model's visual_cond_noise_aug). 1.0 = exact; lower mixes noise into the keyframe latent and lets the shot drift from it. Worth lowering a hair on a long chain, where re-encoded handoff frames accumulate colour/detail drift. Advanced — leave at 0.999 unless you are chasing that. |
| refineopt | COMBO | auto | Whether the sampler's own vision LLM (its optional 'llm_config' input) reworks this shot just before it is sampled, when the real first frame finally exists. • auto — rework the OPENING when this shot's first frame is one the writer never saw (an inherited tail: its text was written against a forecast); write the whole shot when there is no proper prompt at all; otherwise leave it alone. • off — hands off. Use this when you pasted a finished prompt yourself. • opening — always rewrite [Scene Overview] and the first beat from the real frame. • full — always rewrite the shot from the frames, keeping the style and negative sections. Ignored entirely when no 'llm_config' is wired into the sampler. |
Outputs (2)
| Name | Type | Description |
|---|---|---|
| shots | KINBURG_MORPHEUS_SHOT | — |
| info | STRING | — |