feat(wanvideo): add manual start reference support for WanAnimate loop #1840
Conversation
|
Do I need to skip the validation here? context_options are not compatible or necessary with WanAnim looping, since it creates the video in a loop. |
|
No need. This modification doesn't require context_options.
…---- Replied Message ----
| From | ***@***.***> |
| Date | 01/01/2026 14:54 |
| To | kijai/ComfyUI-WanVideoWrapper ***@***.***> |
| Cc | 小六妞儿 ***@***.***>,
Author ***@***.***> |
| Subject | Re: [kijai/ComfyUI-WanVideoWrapper] feat(wanvideo): add manual start reference support for WanAnimate loop (PR #1840) |
kslz left a comment (kijai/ComfyUI-WanVideoWrapper#1840)
Do I need to skip the validation here?
image.png (view on web)
context_options are not compatible or necessary with WanAnim looping, since it creates the video in a loop.
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you authored the thread.Message ID: ***@***.***>
|
但是我在运行的时候报了这个提示 |
|
稍等 可能是我改的不对 我再试一下 |
|
需要把context_options节点断开或删去，使用wananimateembeds来控制循环，我这个改动也是在wananimateembeds这里添加参考帧的
…---- Replied Message ----
| From | ***@***.***> |
| Date | 01/01/2026 15:03 |
| To | kijai/ComfyUI-WanVideoWrapper ***@***.***> |
| Cc | 小六妞儿 ***@***.***>,
Author ***@***.***> |
| Subject | Re: [kijai/ComfyUI-WanVideoWrapper] feat(wanvideo): add manual start reference support for WanAnimate loop (PR #1840) |
kslz left a comment (kijai/ComfyUI-WanVideoWrapper#1840)
稍等 可能是我改的不对 我再试一下
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you authored the thread.Message ID: ***@***.***>
|
|
hi, im using wan animate as replacing mode. i put reference image to ref_image and start_ref_image both but there is still degradation exist. |
ref_image and start_ref_image cannot use the same image. The function of start_ref_image is identical to the original auto-looping feature. For example, when the total frame count is 305 = 77 * 4 - 3, WanAnimateEmbeds will automatically divide the sequence into 4 segments. Each segment uses the last frame of the preceding segment as its start_ref_image. However, due to memory constraints, videos lasting several minutes may exceed thousands of frames. This necessitates manually splitting the video into multiple segments for separate generation. Here, start_ref_image serves its purpose: inputting the last frame from the previous generation ensures consistency (matching the automatic segmentation effect of WanAnimateEmbeds). |
|
The function of ref_image remains unchanged; simply input the character or clothing you wish to replace as usual. |
比如说我在生成了1-60帧后想继续生成，就可以输入第60帧，然后选择生成61-120帧 对吗？ |
|
对的
…---- Replied Message ----
| From | ***@***.***> |
| Date | 01/02/2026 00:42 |
| To | kijai/ComfyUI-WanVideoWrapper ***@***.***> |
| Cc | 小六妞儿 ***@***.***>,
Author ***@***.***> |
| Subject | Re: [kijai/ComfyUI-WanVideoWrapper] feat(wanvideo): add manual start reference support for WanAnimate loop (PR #1840) |
kslz left a comment (kijai/ComfyUI-WanVideoWrapper#1840)
hi, im using wan animate as replacing mode. what is proper usage of your changes? these days im finding solution for degrading quality that masked part in video generation when video length is long.
i put reference image to ref_image and start_ref_image both but there is still degradation exist.
ref_image and start_ref_image cannot use the same image. The function of start_ref_image is identical to the original auto-looping feature. For example, when the total frame count is 305 = 77 * 4 - 3, WanAnimateEmbeds will automatically divide the sequence into 4 segments. Each segment uses the last frame of the preceding segment as its start_ref_image. However, due to memory constraints, videos lasting several minutes may exceed thousands of frames. This necessitates manually splitting the video into multiple segments for separate generation. Here, start_ref_image serves its purpose: inputting the last frame from the previous generation ensures consistency (matching the automatic segmentation effect of WanAnimateEmbeds).
比如说我在生成了160帧后想继续生成，就可以输入第60帧，然后选择生成61120帧 对吗？
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you authored the thread.Message ID: ***@***.***>
|
非常感谢，我实现了我想要的效果 |
|
我用WanVideo Animate Embeds控制循环的时候 画质会逐渐劣化，如果我用你的这个改动每次手动上传最后一帧来重新生成，劣化的问题会有改善吗 |
这个实际只是扩展了WanVideo Animate Embeds的循环，效果与WanVideo Animate Embeds的自动循环是一致的，只是为了保持手动分段后的一致性，劣化问题无法避免 |
抱歉再问下 对于这个劣化问题 你目前有什么解决方案吗 |
我平时生成的基本不超3分钟，我感觉3分钟内基本没劣化，因此没怎么研究过劣化问题。代码里这个参考帧是写死只参考一帧的，也许改成每次都参考5帧能改善劣化问题 |
再次感谢 |
|
i usually generate 20sec~1mins video not using start_ref, but only native wananimate embed and comfy loop. |
3 participants
My device has 64GB of memory, and running more than four windows causes memory exhaustion. To bypass this limitation and generate videos indefinitely while maintaining consistency, I added the
start_ref_imageparameter to manually provide a reference image—typically the last frame from the previous video generation. I tested consistency, and the results are nearly identical to those achieved with automatic window segmentation.