Provide ref_pad_image when your ComfyUI workflow uses WanImageToVideo with a start_image and you want the padding after the start image to carry a reference instead of neutral gray. It is a specific conditioning-input choice, not a general reference-strength control or a guarantee against visual drift. If start_image is absent, the official node documentation says ref_pad_image has no effect.
Decide and connect the inputs
The node places the supplied start image at the beginning of the frame sequence. According to the WanImageToVideo input documentation, remaining frames use neutral gray values unless a padding reference is supplied. With both inputs connected, the first image in the ref_pad_image batch replaces that gray padding. The documentation describes this as SVI-style anti-drift padding used by models such as ID-V2V; it does not claim that every model or scene will become more consistent.
Use this sequence to make the choice:
- Confirm that the workflow actually runs the ComfyUI
WanImageToVideonode. Do not assume a similarly named field in another interface has identical behavior. - Check whether an image is connected to
start_image. If not, leaveref_pad_imageout of the decision: by the documented behavior, it cannot affect the node output on its own. - Decide what you want the padding frames to carry. Keep the optional input unconnected if neutral gray padding is intended. Connect a suitable reference if you want the padding to carry image content.
- Inspect the actual type labels and image batch presented to
ref_pad_image. The node uses only the first image in that batch, so select or order the intended reference first rather than assuming every batch item participates. - Before generating, record the expected padding source as “neutral gray” or “first reference image.” This checks the input structure, not the eventual visual quality of a generated clip.
The node description specifies resizing to the configured width and height. It does not give a general rule for every aspect-ratio mismatch, batch interaction, or output appearance, so do not infer those details from the input names alone.
Hypothetical input-check worksheet
The following values and observations are illustrative, not a test result or a promise about generated video. The table helps distinguish a missing prerequisite from an intentional gray-padding choice.
| Hypothetical workflow state | Input observation before generation | Expected padding source from the documented behavior | Decision |
|---|---|---|---|
| Start image connected; no padding reference | start_image has one image; ref_pad_image is empty | Neutral gray | Leave the optional input empty if that is the intended conditioning structure |
| Start image connected; reference batch contains a scene image followed by a detail image | Both inputs are connected; the scene image is first in the padding batch | First batch image, the scene image | Keep this order only if the scene image is the intended padding reference |
| Start image connected; intended reference appears second in the batch | Two reference images are visible; the desired one is not first | First batch image, not the desired second image | Reorder or select the intended first image, then recheck the type label |
| Start image missing; padding reference connected | ref_pad_image has images, but start_image is empty | No effect from ref_pad_image | Connect a start image only if the workflow calls for one; otherwise do not expect this input to act alone |
This is an input-acceptance worksheet based on the node description, not a comparison of generated results. It records the proposed inputs and the documented padding source; it cannot establish that reference padding improves a particular shot.
Limits and what to do next
Treat the padding reference as part of the node’s conditioning structure, not as a numeric slider for stronger image adherence. The documentation says the first reference image is resized and written into the padding frames when start_image is also present. It does not specify a tunable strength, promise exact preservation of image details, or guarantee that later frames will avoid drift.
If the intended reference is not being used, first check the start_image connection and then the first item in the reference batch. Confirm the actual input type labels; node color can be a clue, but the type label is the reliable check. If the wiring and order match the documented behavior yet the generated clip still diverges, record that as an output-quality issue rather than assuming the padding input failed. The supplied documentation does not establish a universal fix or a particular follow-up setting. A controlled comparison between gray padding and reference padding may help you evaluate your own workflow, but that comparison is an author-recommended review method, not an official node guarantee.
Common questions
Does `ref_pad_image` replace `start_image`?
No. The documented behavior requires start_image for the padding reference to have an effect. The start image still initializes the beginning of the sequence.
Does the node use every image in the reference batch?
No. The documentation specifies that it uses the first image of the batch. Do not treat later batch images as additional references for this input.
Is reference padding guaranteed to stop drift?
No. The documentation calls the behavior SVI-style anti-drift padding, but does not promise a result for a particular model, prompt, or scene. Treat it as a conditional input choice and assess generated output separately.
Sources and next step
This guide interprets public documentation and uses explicitly hypothetical acceptance examples, not application tests. Controls and availability can change; verify the cited manual against your installed version.
For work created in Imagild, carry the input requirements and acceptance checklist into the film workspace. The external software controls discussed above belong to their respective products.