ComfyUI Video Progression

Custom ComfyUI nodes for iterative video storytelling workflows.

This package is built around a simple progression loop:

• generate a first scene
• keep the current video state in Infinite_Video*.mp4
• generate further scenes from the last frame of that video
• optionally force a custom last frame
• choose or auto-pick candidate images
• rebuild the full frame batch for the next video write

Included nodes:

• Story Scene Switch
• Whole Video
• Last Frame Save
• VP Image Chooser
• Prompt Traveling
• Vace Resolution Wizard

Installation

Copy or clone this repository into your ComfyUI custom_nodes folder:

ComfyUI/custom_nodes/Comfyui-Video-Progression

Then:

1. Restart ComfyUI.
2. Hard refresh the browser (Ctrl+F5) so the VP Image Chooser frontend loads.

Requirements

These nodes expect a standard ComfyUI environment with:

• torch
• numpy
• Pillow
• imageio

Video-aware nodes also rely on an imageio backend that can read MP4 files in your environment.

Nodes

Story Scene Switch

Main routing node for first-scene vs further-scene generation.

Inputs:

• scene_mode: First Scene Generation or Further Generations
• user_prompt: Yes or No
• auto_generate: Yes or No
• force_last_frame: Yes or No
• image_input: optional IMAGE
• last_frame_input: optional IMAGE
• prompt: optional input STRING
• trigger: optional FLOAT used to force re-execution when needed

Outputs:

• Image
• Prompt
• SceneMode
• LastFrame
• ForceFlag
• AutoFlag

Behavior:

• auto_generate = Yes
  • forces the node to behave as Further Generations
  • First Scene Generation cannot be auto-generated
  • the SceneMode output reflects the effective mode used by the node
• First Scene Generation
  • removes temporary last-frame files from the ComfyUI output folder
  • uses image_input as the main image when available
  • if force_last_frame is Yes, last_frame_input is exposed through LastFrame
• Further Generations
  • reads the last frame from the latest Infinite_Video*.mp4 in the output folder
  • if force_last_frame is Yes and last_frame_input is connected, that image is used instead
  • falls back to image_input only if the video frame cannot be read

Prompt logic:

• when scene_mode is First Scene Generation, or user_prompt is Yes, the node uses the provided prompt
• otherwise it loads prompt_further.txt from the output folder
• the prompt body is not rewritten
• the node only adds instruction prefix/suffix text

Extra outputs:

• ForceFlag is a BOOLEAN mirror of force_last_frame
• AutoFlag is a BOOLEAN mirror of auto_generate
• LastFrame is only populated when force_last_frame is Yes

Typical uses:

• connect ForceFlag to LazySwitchKJ
• connect AutoFlag to VP Image Chooser.auto_pick
• connect SceneMode to Whole Video.scene_mode

Whole Video

Builds the full frame batch that should be sent to VHS_VideoCombine.

Inputs:

• scene_mode: string from Story Scene Switch
• wait_for: IMAGE batch, typically from the VAE decode output you want to turn into video

Outputs:

• Images
• count

Behavior:

• First Scene Generation
  • if a previous Infinite_Video*.mp4 exists, it is archived as Definitive_Infinite_Video_XXX.mp4
  • only the new wait_for frames are returned
• Further Generations
  • loads frames from the latest Infinite_Video*.mp4
  • deletes that source file
  • appends the new wait_for frames
  • returns the combined batch

Important:

• this node does not write the final video file itself
• it prepares frames for VHS_VideoCombine
• VHS_VideoCombine should then write the new Infinite_Video*.mp4

Last Frame Save

Utility node that writes an input image to:

output/temp_last_frame.png

Inputs:

• image

Outputs:

• Image

Behavior:

• removes the previous temp_last_frame.png if it exists
• saves the new one
• passes the image through unchanged

VP Image Chooser

Standalone image chooser extracted into this repository, with its own frontend.

Inputs:

• mode: Always Pause or Keep Last Selection
• preview_rescale
• images: optional IMAGE batch
• auto_pick: optional BOOLEAN

Outputs:

• image

Behavior:

• when auto_pick is True, the node randomly selects one image and returns it immediately
• when auto_pick is False, the node opens a modal dialog so the user can choose manually
• Keep Last Selection reuses the previous selection when possible

Notes:

• this chooser does not depend on Easy Use
• if you update the chooser frontend, restart ComfyUI and hard refresh the browser

Prompt Traveling

Cycles through up to 20 prompts, one prompt per line.

Inputs:

• prompts: multiline STRING, one prompt per line
• prompt_index: visible slider from 1 to 20

Outputs:

• Prompt

Behavior:

• blank lines are ignored
• only the first 20 non-empty lines are used
• on each run, the node outputs the prompt matching the current slider value
• after the run, the visible slider is automatically incremented by 1
• when the next value exceeds the number of available prompt lines, it wraps back to 1

Notes:

• the node is forced to execute on every queue so the counter can advance even if the text does not change
• after editing the frontend files, restart ComfyUI and hard refresh the browser

Vace Resolution Wizard

Resizes an image to a fixed target format for video generation.

Inputs:

• image
• AUTO
• format: Square, 16/9, or Vertical
• crop_instead_of_pad
• Draft

Outputs:

• Width
• Height
• Resized_Image

Target sizes:

• Square: 720 x 720
• 16/9: 960 x 544
• Vertical: 544 x 720

If Draft is enabled, width and height are halved.

Behavior:

• if AUTO is enabled, the format is inferred from the input aspect ratio
• if crop_instead_of_pad is enabled, the image fills the target and is center-cropped
• otherwise the image is fitted inside the target and padded

Recommended Wiring

Typical progression wiring:

1. Story Scene Switch
2. your generation / edit branch
3. VP Image Chooser
4. Whole Video
5. VHS_VideoCombine

Useful control links:

• Story Scene Switch.AutoFlag -> VP Image Chooser.auto_pick
• Story Scene Switch.ForceFlag -> LazySwitchKJ.switch
• Story Scene Switch.SceneMode -> Whole Video.scene_mode
• decoded frames -> Whole Video.wait_for

Output Folder Conventions

These nodes read and write in the ComfyUI output directory resolved from:

1. folder_paths.get_output_directory()
2. /output
3. ./output

Files used by the workflow:

• Infinite_Video*.mp4
• Definitive_Infinite_Video_XXX.mp4
• prompt_further.txt
• temp_last_frame.png

Troubleshooting

• If VP Image Chooser does not open correctly, restart ComfyUI and hard refresh the browser.
• If further generations use the wrong frame, verify that the latest Infinite_Video*.mp4 exists in the output folder.
• If prompt progression is wrong, verify that prompt_further.txt exists and is updated before the next run.