Files
sakura1bgx-ComfyUI_FlipStre…/README.md
T
2024-08-26 00:05:19 +09:00

105 lines
3.8 KiB
Markdown

# ComfyUI_FlipStreamViewer
**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
1. **Run ComfyUI** and load the `workflow.json` file located in the `workflows` folder.
2. Access the viewer at the following URL if ComfyUI is running at `http://localhost:8188`:
`http://localhost:8188/viewer`
3. To respond to viewer input:
- Check 'Extra options', 'Auto Queue', and 'instant' on the ComfyUI control panel.
- Then click the 'Queue Prompt' button.
4. 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 or 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](https://civitai.com/models/395638). 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.