How to vary prompts across frames (keep previous prompt when no new one is defined)

[kaitoyonde]

Hi,

I’m working on a morphing image generation loop in ComfyUI, and I’m trying to figure out a clean way to handle changing prompts across frames, while keeping the previous prompt active when there’s no new one defined.

Here’s my setup:

• standard SDXL stuff, KSampler, LoRA, CLIP Prompt, SDTurboScheduler, set the denoiser to around 0.51 - 0.75, and steps between 1 - 5
• I’m doing loopback generation, where the output image is fed back in as reference input for the next frame. i am using custom node from ImpactPack of FrameSender and FrameReceiver
• I use Canny ControlNet to preserve object structure while letting the image evolve gradually.
• I Rescale, Cropped the image effectively zooming the image, and use that as the next frame input
• The goal is to create smooth morphing effects, and then stitch all the frames into a video.

My current approach

I use an INT node to track the frame number, with “Control After Generate” set to Increment, so the frame value automatically increases after each generation.

For each frame, I want to apply a different positive prompt depending on the frame number. For example:

{
    0: "a cat on a chair",
    5: "a cat jumping",
    10: "a cat flying"
}

That means I want this behavior:

frames 0–4  → "a cat on a chair"
frames 5–9  → "a cat jumping"
frame 10    → "a cat flying"

Right now, I’m testing this using a custom Dict node from the Basic Data Handling plugin, where each key = frame number and value = prompt.
But since not all frame numbers exist in the dict, I need a way to use the previous value for missing frames.

(Workflow Embedded)

What I need

I want something that, for a given frame, returns the prompt for:

max(key) where key <= frame

Basically: “use the most recent defined prompt up to this frame.”\

Example result

Here’s an example of what I’ve achieved so far with this setup (loopback + ControlNet + Canny) but i literally directed (change it on the go, using 'Run after change') the prompt manually across.
Youtube

You can see how it morphs smoothly when the prompts align.
Any advice, node examples, or alternative workflows are very welcome

[hasan-aghayev]

The lookup you need is a previous-keyframe lookup, not an exact Dict lookup: choose the largest numeric key that is less than or equal to the current frame. With keys 0, 5, and 10, frame 7 should return the entry for key 5. Add key 0 as a default so frames before the first change have a value. If your Dict node only supports exact keys, the simple workaround is to expand the schedule to one entry per frame; the cleaner option is a small helper node that sorts numeric keys and returns the last key at or before the frame. Make sure the keys and frame counter are both numbers, not strings, and read the counter before the current iteration generates its frame.