ComfyUI Warper Nodes
NODES: DWPose Scaler (Warper), Mouth Mask from Pose (Warper), Facial Part Mask from Pose (Warper), Smart Video Batcher (Warper), Get Batch By Index (Warper), Smart Overlapping Batcher (Warper), Flow Visualizer (Warper), Preprocess for Target (Warper), Crop and Restore (Warper), Aspect Ratio Resolution (Warper)
Nodes (17)
Turn 'some image' into a clean 720p-sized frame
Snap any image to 720p or 1080p
One dropdown to size any generation
Glue prompt fragments together without a mangled comma
Undo your padding without losing pixels
Stretch a skeleton's legs and neck without stretching the image
Face, mouth, and eye masks from DWPose landmarks
See the motion in your frames with RAFT
Pull one chunk out of your batched frames
Batch-generate from a folder of txt prompts
One prompt, precisely, from a folder
Lip-sync masks without a single manual paint
Fit any image to a fixed canvas without cropping it
Keep long videos temporally consistent with overlapping chunks
Split your frame stack so the last batch doesn't blow up
Fixed, incremental, or seeded-random access to your prompt files
Know which prompt file you're on, mid-batch
ComfyUI Warper Nodes
Lightweight JPEG saving and previews for ComfyUI. This package contains exactly two nodes, both under Warper Tools/Image.
Nodes
| Node | Behavior |
| --- | --- |
| Save JPEG (Warper) | Saves every image in a batch directly as .jpg in ComfyUI's output directory and displays the saved images in the node. |
| Preview JPEG (Warper) | Displays every image in a batch using .jpg files in ComfyUI's temporary directory. Does not save to the output directory. |
Connect the IMAGE output of VAE Decode (or any image-producing node) to either node. No intermediate PNG is created.
Both nodes have a quality setting from 1 to 100, default 95, and preserve the original resolution. JPEG encoding uses full chroma resolution (4:4:4). Save JPEG also accepts a filename_prefix, including subfolders and %batch_num%.
JPEG is lossy; file size depends on image content and quality. RGBA inputs lose their alpha channel. These nodes do not embed prompt/workflow metadata; save your workflow separately as JSON. Temporary previews may be removed by ComfyUI. JPEG encoding reduces file sizes, not VAE decoding memory use. Remove or bypass any existing PNG Save/Preview nodes if you no longer want their PNG files.
Installation
From your ComfyUI custom_nodes directory:
git clone https://github.com/AIWarper/ComfyUI-WarperNodes.git
Restart ComfyUI, then search for Save JPEG (Warper) or Preview JPEG (Warper). No extra models, node packs, or dependencies are needed beyond ComfyUI's existing NumPy and Pillow.
To update an existing Git installation, run git pull inside custom_nodes/ComfyUI-WarperNodes and restart ComfyUI.
Legacy nodes retired
This is a breaking update: all former pose, mask, optical-flow, batching, resolution/preprocessing, prompt-loading, and text-file iterator nodes have been retired and removed, along with their examples and bundled RAFT/UniMatch code. Only the two JPEG nodes above are registered.
Existing workflows using retired nodes will report missing nodes after updating. If you still need those workflows, keep a separate ComfyUI installation on legacy commit 35e83bb rather than updating it. The legacy implementation remains available in Git history.
Development
Run the CPU tests with ComfyUI's Python environment (or Python with NumPy, Pillow, and Torch installed):
python -m unittest discover -s tests -v