ComfyUI_FlipStreamViewer is a tool that provides a viewer interface for flipping images with frame interpolation, allowing you to create high-fidelity pseudo-videos without needing AnimateDiff.

Required Custom Nodes

Please install the following nodes via ComfyUI-Manager:

ComfyUI Impact Pack
ComfyUI Frame Interpolation
ComfyUI WD 1.4 Tagger
FizzNodes
LoRA Tag Loader for ComfyUI

Getting Started

Run ComfyUI and load the workflow.json file located in the workflows folder.
Access the viewer at the following URL if ComfyUI is running at http://localhost:8188:

http://localhost:8188/viewer

To respond to viewer input:

Check 'Extra options', 'Auto Queue', and 'instant'.
Click the 'Queue Prompt' button on the ComfyUI control panel.

The 'preset' folder may be created in the current directory for saving and loading preset files. Sample presets can be found in the presetsamples folder.

Prompt Format

Use the following format for input prompts. The 'pre text' and 'append text' sections apply to all frames. The separator ---- should consist of four hyphens, and each line of 'frameN text' applies to individual frames.

''' pre text,

frame1 text
frame2 text

append text '''

Limitations

Excessively long text may cause exceptions.
The keywords BREAK, AND, and <lora:...> cannot be used in the prompt.

Parameters

You can adjust the following parameters on the viewer:

Seed: Random seed for KSampler.
Iv: Interval in seconds for auto-changing the random seed.
Nstp: Steps for KSampler.
Ncfg: CFG for KSampler.
Nsta: Start step for the second KSampler; increasing this value will stabilize the output between frames.
Nfrm: Number of frames; each frame undergoes x8 interpolation in the sample workflow.
Nspf: Seconds per frame displayed in the viewer stream.

Inputs

Provide the following text inputs in the viewer:

Prompt Input: Located at the top of the left pane (refer to the Prompt Format section above).
Negative Prompt Input: Located below the Prompt Input; this applies to all frames.
LoRA Input: Located in the middle of the right pane; input LoRA such as <lora:startsfilename:1>.

Controls

Left Pane Buttons:

Update: Apply input data to the workflow.
Change: Change the seed and update.
K: Keep the seed to search for another good seed.
B: Go back to the previous seed.
M: Move the checkpoint file.
L: Load a preset.
P: Load a preset prompt only.
S: Save the current preset.
D: Delete a preset.

Right Pane Buttons:

T: Toggle LoRA enable/disable.
M: Move the LoRA file.
T: Toggle tag enable/disable at the Prompt Input.
R: Switch to random tags at the Prompt Input.
U: Same as Update.
W: Apply WD14Tagger to the first frame; outputs tags to LoRA input.
C: Capture an image from another browser tab or desktop and set it as the first frame; useful with W.

Additional Features

LoRA Preview: At the bottom of the right pane; click to jump to the Civitai LoRA page if found.
Auto-hide: Click to hide controls and images; the viewer auto-hides after 5 minutes of inactivity.

Quality and Speed

Use more higher-quality checkpoint models.
Recommended speed setting: 4stp 1cfg 0sta dpmpp_2m,sgm_uniform with <lora:TCD-SDXL:1>. You can download the TCD LoRA from Civitai. Rename the file to TCD-SDXL.safetensors.

Customizing Workflow

You can customize the workflow for specific needs:

FlipStreamLoader: Use the 'mode' parameter to switch between sub-directories for checkpoints and LoRAs, useful for managing models like sd15, pony, etc.