CIVITAI / Workflows

EverAnimate Long-Video Consistency Editing Workflow

Watch the full video first if you want to understand how this EverAnimate long-video consistency editing workflow works in practice. The video shows how an existing video can be edited with reference guidance, pose control, background preservation, character masking, and long-form continuation while keeping the edited result more stable across multiple segments. This ComfyUI workflow is designed for EverAnimate long-video consistency editing. Its main purpose is not only to generate an animated character, but to edit an existing video while preserving the important motion, body structure, background relationship, and temporal continuity. Compared with a simple one-shot video edit, this workflow is built for longer clips where character identity, pose rhythm, mask boundaries, and visual consistency often break after the first segment. The workflow starts from a driving video. VHS video loading and VHS_VideoInfo read the selected FPS, frame count, width, height, and duration. This gives the workflow a stable timeline before generation begins. The source frames are then processed through multiple pose and detection routes, including ViTPose, YOLO, SDPose, PoseAndFaceDetection, DrawViTPose, BBoxYOLO, and SDPoseKeypointExtractor. These preprocessing stages turn the original video into usable pose guidance, body structure information, and motion control signals. The editing part is where this workflow differs from the pure generation version. It includes a character mask path, mask expansion, and block-style mask processing. GrowMaskWithBlur expands and softens the mask, while BlockifyMask converts it into a more usable character-editing mask. This mask is then sent into ComfyEverAnimate as character_mask, together with background_video and pose_video. This allows the workflow to focus the edit on the character area while keeping the background or surrounding structure more stable. The core generation node is ComfyEverAnimate. In the first local editing segment, it receives the reference image, pose video, background video, character mask, positive and negative conditioning, width, height, length, pose strength, face strength, and motion handoff settings. After the first segment is generated, TrimVideoLatent and ComfyEverAnimateTrimImages remove redundant anchor latent frames and duplicate image frames. The workflow then uses continue_motion to pass motion inf