CWK_Wan2.2_Nodes Introduction
CWK_Wan2.2_Nodes is a set of custom ComfyUI nodes for building Wan 2.2 image-to-video workflows. It helps you prepare a starting image, plan multiple video clips, assign different prompts and LoRAs to each clip, and connect those clips into a longer video.
The package offers two ways to use a multi-clip plan:
- Static workflow: Select a particular clip from your plan and generate it in a non-looping graph.
- Looping workflow: Generate clips one after another, carry information between them, and join their frames into a longer result.
The extension organizes these tasks inside ComfyUI; it does not include a Wan 2.2 model. You need a Wan 2.2 model available in your ComfyUI setup.
How CWK_Wan2.2_Nodes Works
Think of the Prompt Composer as a storyboard. Each block describes one clip: what should happen, how long it should last, which seed to use, and which LoRAs to apply. The composer packages the blocks into a WAN22_PIPELINE—a plan that other CWK nodes can read.
From there, you can choose how to use the plan:
- Pick one clip: The Pipeline Splitter reads a clip at the index you specify and sends its prompt, frame count, seed, and LoRA stacks to the rest of your graph.
- Work through the clips: Loop Open supplies the current clip’s settings to your generation workflow. After that clip is generated, Loop Close adds its frames to the accumulated video and re-queues the workflow for the next clip. The loop finishes when all clips are done.
The Image Prep node can supply a cropped starting image and output settings for downstream nodes. The LoRA Applier applies a clip’s LoRA stack to a model and CLIP pair. Your Wan 2.2 generation nodes still perform the actual video generation; CWK_Wan2.2_Nodes helps coordinate what they receive.
For example, you could make one block for a wide establishing shot and another for a close-up. Give each block its own prompt and seed, then use the looping nodes to generate the clips in sequence rather than manually selecting each one.
CWK_Wan2.2_Nodes Features
Prompt Composer: plan your clips
CWK Wan2.2 Prompt Composer uses a custom canvas interface inside its ComfyUI node. You can add any number of prompt blocks. Each block has:
- Prompt text: The description for that clip.
- Duration in seconds: How long you want the block to last. The composer shows the total planned duration in a live stats bar.
- Seed: A per-clip value you can set individually or randomize across all blocks.
- High-noise and low-noise LoRA stacks: Two independent sets of LoRAs for workflows that use separate high-noise and low-noise stages.
- Enable, disable, and invert controls: Manage whether a block is active. Bulk controls can enable or disable all blocks at once.
The stats bar shows how many blocks are enabled and their total duration. This makes it easier to review your plan before running a longer workflow.
You can save and load named preset collections as .json files through ComfyUI’s userdata system. The Load & Manage Presets panel also lets you rename and delete collections without leaving ComfyUI. A button can activate an animated video-generation preview when ComfyUI-VideoHelperSuite is installed.
Output: pipeline (WAN22_PIPELINE), which connects to the Pipeline Splitter or looping nodes.
Pipeline Splitter: select a clip without looping
CWK Wan2.2 Pipeline Splitter is for a static, non-looping graph. Give it the composer’s pipeline and a clip_index; it extracts the selected clip’s settings. This is useful when you want to test one part of your storyboard before generating a longer sequence.
Inputs: pipeline, clip_index
Outputs: prompt, frame_count, lora_stack_high, lora_stack_low, seed
LoRA Applier: apply a clip’s LoRAs
CWK Wan2.2 LoRA Applier takes a model, a CLIP input, and a LORA_STACK. It applies the LoRAs one at a time using ComfyUI’s standard LoraLoader. It is designed to work safely inside looping graphs, where the requested LoRA stack may change from clip to clip.
Inputs: model, clip, lora_stack
Outputs: model, clip
Loop Open and Loop Close: build a longer sequence
CWK Wan2.2 Loop Open begins each clip’s pass through a looping workflow. It tracks which pipeline clip is current across ComfyUI re-queues and sends that clip’s prompt, frame count, LoRA stacks, and seed to your generation nodes. It also provides the current image and previous latent for workflows that use them.
Its settings include start_image, which supplies the initial image; overlap controls, which determine how clips meet; and force_reset. The loop automatically resets when the pipeline changes or when all clips are complete.
Loop Open inputs: pipeline, start_image, overlap, overlap_mode, overlap_side, force_reset
Loop Open outputs: current_image, prev_latent, prompt, frame_count, lora_stack_high, lora_stack_low, clip_seed, loop_state
CWK Wan2.2 Loop Close receives the newly generated images and latent, adds the frames to the running accumulation, and either re-queues for the next clip or returns the completed result.
You can choose how overlapping frames are joined:
blend_linear(default): A standard cross-fade between clips.blend_sqrt: A softer, square-root-weighted cross-fade.replace: A hard cut with no blending.
The final_only setting controls when images are passed onward. Leave it at True (default) to suppress intermediate clips until the loop is complete, or set it to False to forward intermediate clips to a downstream video-combine node.
Loop Close inputs: loop_state, new_images, new_latent, final_only
Loop Close outputs: images, final_latent
Image Prep: frame your starting image
CWK Wan2.2 Image Prep lets you upload an image from your browser, so you do not need a separate LoadImage node for that step. On the node itself, you can drag and resize a crop frame to choose what stays in view. The crop frame remains locked to your selected output aspect ratio.
Available resolution presets are:
| Aspect ratio | Resolution |
|---|---|
| 16:9 | 832 × 480 |
| 16:9 | 1280 × 720 |
| 9:16 | 480 × 832 |
| 9:16 | 720 × 1280 |
| 1:1 | 1024 × 1024 |
For example, you might choose 9:16 for a portrait-oriented composition or 16:9 for a landscape-oriented one. You can also configure the upscale method, frame rate, sampler, scheduler, steps, and CFG scale. The node passes these settings as outputs for downstream sampling nodes; Image Prep itself does not generate the video.
Outputs: image, width, height, frame_rate, scheduler, sampler_name, total_steps, split_steps, cfg_scale
CWK_Wan2.2_Nodes Models
CWK_Wan2.2_Nodes does not provide separate downloadable models. It is a set of workflow nodes intended for use with a Wan 2.2 image-to-video model already loaded in ComfyUI. Its prompt blocks and LoRA stacks let you vary the settings sent to that model between clips; the model and your connected generation workflow determine the generated video.
Troubleshooting CWK_Wan2.2_Nodes
A clip is not receiving the settings you expected
- In the Prompt Composer, check the block’s prompt, duration, seed, and LoRA stacks.
- Check whether the block is enabled, and review the enabled-block count in the stats bar.
- If you are using the Pipeline Splitter, check its
clip_indexto make sure you selected the intended clip. - If you changed a looping pipeline and want to restart its progress, use Loop Open’s
force_resetcontrol.
The animated preview button does not provide a preview
- Check whether ComfyUI-VideoHelperSuite is installed.
- If it is not installed, use the rest of the workflow without that optional preview feature.
Clip transitions are not what you wanted
- Check the overlap settings on Loop Open.
- Try
blend_linearfor a standard cross-fade orblend_sqrtfor a softer one. - Choose
replaceif you want a hard cut rather than a blend.
Intermediate clips are not reaching your video-combine node
Check final_only on Loop Close. Its default value, True, holds back intermediate images until the loop finishes. Set it to False if you want those intermediate clips forwarded.
Frequently asked questions
Can you test one clip without running the whole loop?
Yes. Connect the Prompt Composer to the Pipeline Splitter and select a clip_index in a static graph.
Do you need a separate LoadImage node for Image Prep?
No. Image Prep supports uploading an image directly from your browser.
Does the extension supply the Wan 2.2 model?
No. You need a Wan 2.2 model in your ComfyUI setup.
Can each clip have different LoRAs?
Yes. Every prompt block has its own high-noise and low-noise LoRA stacks.
Learn More about CWK_Wan2.2_Nodes
- CWK_Wan2.2_Nodes repository: The extension’s source and project information.
- LICENSE.txt: The extension’s MIT license.
