Prompt Scheduling [Latents] 🎭🅐🅓

Displayed as Prompt Scheduling [Latents], and the pack's own node description says exactly what it does: "Encode a schedule of prompts with automatic interpolation, its length matching passed-in latent count." That last part is the useful bit - instead of you manually figuring out how many frames you have and hand-writing a schedule keyed to that number, this node reads your latent batch, sees how many frames it actually is, and interpolates your prompt schedule to fit automatically.

This is the built-in way to do prompt travel - starting on one description and drifting toward another over the course of your clip - that the README points to before mentioning the alternative, BatchPromptSchedule from the separate ComfyUI_FizzNodes pack. If you want your subject's outfit to shift, the scene to transition, or the style to drift across a generated animation, this is the node.

How it works

You write multiple prompts in the prompts field, keyed to points across your sequence, and the node interpolates between them as it encodes conditioning - smoothly blending from one prompt's influence to the next rather than hard-cutting. Because it takes your actual latent batch as an input, the schedule automatically scales to however many frames that batch contains, so you don't have to keep a frame-count number in sync with your prompt schedule by hand when you change batch size.

The inputs and outputs that matter

Required: prompts (multiline string - your scheduled prompt text), clip (your CLIP model, same as any text encode node), and latent (the batch this schedule gets matched against - this is what makes the schedule "automatic" rather than manually counted).

Optional: prepend_text and append_text (multiline strings added before/after every scheduled prompt - handy for a consistent style tag or quality suffix you don't want to retype into every scheduled entry), values_replace (a substitution mechanism for the schedule), print_schedule (boolean - logs the resolved schedule to console, useful for confirming your interpolation did what you expected before committing to a full render), and tensor_interp (a dropdown for the interpolation method between prompts). Outputs are CONDITIONING (wire to your KSampler) and LATENT (a passthrough of your input latent, sized to match - convenient for chaining without a separate wire back to your original latent source).

How to install it

Standard for the pack - ComfyUI Manager, search AnimateDiff Evolved by Kosinkadink, or:

cd ComfyUI/custom_nodes && git clone https://github.com/Kosinkadink/ComfyUI-AnimateDiff-Evolved

No model download tied to this node - it's a conditioning/scheduling utility that sits alongside whatever motion setup you've already got.

Common issues & troubleshooting

Not sure the schedule is being read correctly. Turn on print_schedule before committing to a full render - it logs exactly what the node resolved your prompt schedule to, which is the fastest way to catch a typo or misunderstood syntax before spending generation time on it.

Transitions between prompts feel abrupt rather than smooth. That's controlled by tensor_interp - try a different interpolation method from its dropdown if the default isn't blending the way you expect.

Changed your batch size and the schedule seemed to shift unexpectedly. That's expected behavior, not a bug - the whole point of this node is that the schedule re-fits to whatever latent batch size you feed it, so a batch-size change will re-interpolate your prompt points across the new frame count automatically.