返回工作流

LTX 2.3 - EditAnything - Video Editing via Prompt

This experimental workflow uses an IC LoRA to attempt to Add, Remove, Change, or Convert elements of a video. For example, you can add a dog to a scene, or change the art style.

下载工作流在 Cloud 中试用

Input

What to add, remove, replace or restyle, and what to keep as it is.

The clip to edit.

How many frames of the clip to edit (77 at 24 fps is about 3 s).

Default: 77

Target frame; the clip is padded to this aspect ratio.

Default: 1,080

Target frame; the clip is padded to this aspect ratio.

Default: 1,920

Advanced settings

What to avoid.

Leave empty for a new result each run. Enter a number to repeat one.

Run

Output
00:00
Example

试试一个示例

关于此工作流

LTX 2.3 - EditAnything is an experimental ComfyUI workflow that performs text-driven edits on existing video clips. Using an in‑context (IC) LoRA, it attempts to add, remove, replace, or restyle objects directly inside each frame without manual masking. The pipeline is simple: VHS_LoadVideo decodes your source clip into frames, a custom IC‑LoRA edit node (UUID 2e37ecc9-770a-4ac2-972c-e953901f49d3) applies the text instruction to those frames, and VHS_VideoCombine stitches the edited frames back into a video. A MarkdownNote node in the canvas provides author guidance and caveats.

Technically, the IC LoRA conditions the diffusion/edit process with your prompt so the model can infer where and how to apply changes across frames. Because the edit is prompt-only and experimental, temporal consistency will vary; shorter clips, clearer prompts, and conservative edit strength typically preserve motion and identity better. VHS_LoadVideo and VHS_VideoCombine (from VideoHelperSuite) handle robust frame I/O and encoding, so you can iterate quickly by tweaking the edit node’s prompt and strength, then re-queuing to compare results side by side.

常见问答