Remove Video Objects and Watermarks (Wan)
Upload a video and say what to remove. Florence2 and BiRefNet build the masks, and Wan MiniMax-Remover fills the area across all frames.
- L40S
- Video editing
- Free template
How it works
The video is loaded with VHS_LoadVideo and resized with ImageResizeKJv2 using the selected resolution and aspect ratio. Masks are created with Florence2 (comfyui-florence2) and BiRefNetRMBG, then combined with MaskToImage, SolidMask and ImageCompositeMasked.
Removal uses the ComfyUI-WanVideoWrapper nodes. WanVideoModelLoader works with Wan2_1-MiniMaxRemover_1_3B_fp16, and WanVideoVACEModelSelect adds Wan2_1-VACE_module_1_3B_bf16. WanVideoMiniMaxRemoverEmbeds prepares the source frames and mask. Two LoRAs, Wan2.1_T2V_14B_FusionX and lightx2v cfg step distill, speed up sampling. WanVideoSampler runs with the unipc scheduler, and encoding and decoding use Wan2_1_VAE_bf16 and umt5-xxl-enc-bf16 for text.
Optional helpers include TeaCache, SLG, Enhance-A-Video, block swap, VRAM management and torch compile. The output goes through ColorAdjustment and LTXVFilmGrain, then VHS_VideoCombine writes the MP4. An LLM node (OpenAIProviderNode) can help phrase the removal text.
Tips
- Use short, specific object names, such as the logo in the corner, for the Florence2 step.
- Start with LQ or MQ to check the mask, then switch to HQ for the final render.
- Watermarks that move or fade are harder, so check the full clip.
- The 81-frame default fits Wan models. Longer clips cost more time.
What you can use it for
- Removing watermarks or logos from clips
- Deleting people or objects from footage
- Cleaning up product videos
- Preparing b-roll with distractions removed
Run it in three steps
Open the template
Click Run on ComfyDeploy. The workflow, its custom nodes and models are set up for you on a cloud machine.
Add your inputs
Provide input video and description of what to remove in the playground or the ComfyUI editor.
Run, share or deploy
Run it on an L40S, share it with your team as a playground, or deploy it as an API endpoint.
Frequently asked questions
What GPU does the video object remover need?
The recommended GPU is the L40S. ComfyDeploy provisions it for you, so there is nothing to install.
How does it find what to remove?
It uses DownloadAndLoadFlorence2Model to locate objects from text, and BiRefNetRMBG for segmentation masks. SolidMask and ImageCompositeMasked help build the final mask.
Can I pick the quality?
Yes. The quality input has HQ, MQ and LQ options, and the RESOLUTION input offers I2V480p, I2V720p, T2V1.3B and T2V14B.
