# ComfyUI-FrameIO

![preview](preview.png)

High-performance frame input/output nodes for **AI video pipelines** in ComfyUI.

ComfyUI-FrameIO focuses on **efficient frame storage, loading, and reuse** for long video generation workflows, solving common problems like massive PNG sizes, duplicate frames, and fragile filename-based loading.

---

## ✨ Features

### 🧩 Frame Saving
- Save batch `IMAGE` tensors as **WebP**
- **Lossy or Lossless WebP**
- **Skip identical frames** using hash-based deduplication
- Massive disk space reduction vs PNG
- Returns **STRING_LIST** (ComfyUI list-native)

### 📥 Frame Loading
- Pattern-based frame loading
- **STRING_LIST → IMAGE loader** (no filename guessing)
- **Frame range selector** (`start`, `end`, `step`)
- **Auto-detect frame count** (set `frame_count=0`)
- **Execution Trigger** node for flow control
- **List Video Files** node to scan input directory
- Safe with duplicate paths (from deduplication)

### ⚡ Optimized for AI Video
- **Multi-threaded IO** for high-speed saving & loading
- **Direct PyTorch normalization** (avoids slow CPU/NumPy conversions)
- Designed for **long sequences**
- Works with WebP / PNG / JPG
- Preserves exact frame ordering
- **Security restricted** (saves only to ComfyUI output directory)

---

## 📦 Included Nodes

### 🔹 Batch Save Image Sequence (WebP)
Save an image batch efficiently to disk.

**Inputs**
- `images` (IMAGE)
- `path_pattern` (STRING)
- `start_index` (INT)
- `lossless` (true / false)
- `quality` (INT)
- `skip_identical` (true / false)
- `overwrite` (true / false)

**Outputs**
- `paths` (STRING_LIST)
- `count` (INT)

---

### 🔹 Batch Load Image Sequence
Load images from disk using a filename pattern.

**Inputs**
- `path_pattern` (STRING) - e.g., `./frame{:06d}.webp`
- `start_index` (INT)
- `frame_count` (INT) - **Set to 0 to auto-detect all available frames**
- `ignore_missing_images` (true / false)

**Outputs**
- `images` (IMAGE)
- `count` (INT)

---

### 🔹 Batch Load Image Sequence (Trigger)
Same as the standard sequence loader, but includes an `optional trigger` input. Useful for forcing ComfyUI to wait for a previous node (like a video render) to finish before loading.

---

### 🔹 Batch Load Image List (Advanced)
Load images directly from a `STRING_LIST`.

**Inputs**
- `paths` (STRING_LIST)
- `start` (INT, optional)
- `end` (INT, optional)
- `step` (INT, optional)

**Outputs**
- `images` (IMAGE)
- `count` (INT)

> This is the **recommended loader** when using frame deduplication.

---

### 🔹 List Video Files (Input Dir)
Scans your ComfyUI `input` folder for video files.

**Inputs**
- `subfolder` (STRING) - Subfolder within the input directory.
- `filter` (STRING) - Wildcard filter (default `*`).
- `load_cap` (INT) - Maximum number of files to return (default `100`, 0 for no limit).
- `deep_search` (true / false) - Search subdirectories recursively (default `false`).

**Outputs**
- `paths` (STRING_LIST) - List of absolute paths.
- `count` (INT) - Number of files found.

---

## 🧠 Why ComfyUI-FrameIO?

| Problem | Solution |
|------|------|
| PNG sequences too large | WebP (5–10× smaller) |
| Duplicate frames | Hash-based skip |
| Filename guessing | STRING_LIST loader |
| Partial re-render | Frame range selector |
| Long videos | Optimized IO |

---

## 🎬 Recommended Workflows

### 🔥 Best Practice (Long Videos)
```

KSampler
↓
Batch Save Image Sequence (WebP)
↓ STRING_LIST
Batch Load Image List (Advanced)
↓ IMAGE
Video Combine / FFmpeg

````

### 🎯 Partial Reload
- `start = 300`
- `end = 900`
- `step = 2`

Perfect for re-encoding or post-processing sections.

---

## ⚙️ Recommended Settings

| Use Case | Settings |
|-------|---------|
| Final video | Quality 95, Skip Identical = true |
| Iteration | Quality 90 |
| Dataset / archival | Lossless = true |
| Static shots | Skip Identical = true |

---

## 📥 Installation

```bash
cd ComfyUI/custom_nodes
git clone https://github.com/room3dev/ComfyUI-FrameIO.git
````

Restart ComfyUI.

Nodes will appear under:

```
FrameIO
```

---

## 📄 License

MIT