Vertical to Horizontal Video Reframe

Vertical to Horizontal Video Outpainting Workflow (Wan 2.1)

This workflow converts vertical videos into horizontal cinematic videos by intelligently expanding the frame using Wan 2.1 VACE video generation and AI outpainting. It preserves the original content while generating new visual areas on the sides to match a wider aspect ratio.

The process begins by loading the source video using the VHS Load Video FFmpeg node, which extracts frames and retrieves video metadata such as frame count, FPS, width, and height. These values are analyzed to determine the original aspect ratio and ensure proper resizing and processing of the input frames.

The workflow then resizes the frames while maintaining proportions, targeting a horizontal format (typically 1280×720). Mathematical nodes calculate the padding required to center the original vertical content within the new horizontal canvas.

To generate the missing side regions, the workflow performs outpainting. A mask is created for the empty padded areas, and these masked regions are filled using the Wan 2.1 diffusion model. The system uses:

• Wan 2.1 VACE 14B model for high-quality generation

• FusionX LoRA to improve image-to-video consistency

• UMT5 text encoder for prompt conditioning

• Wan 2.1 VAE for decoding latent frames

The workflow optionally accepts prompts to guide how the expanded regions should look, while negative prompts help suppress artifacts such as blur or noise.

A KSampler performs the diffusion process to generate new visual content for the masked areas across the full frame sequence. The generated latent frames are then decoded into images using the VAE.

Finally, the processed frames are assembled back into a video using the CreateVideo node while preserving the original FPS, producing a smooth horizontal video with expanded scenery.

Key Features

• Converts vertical video to horizontal format

• Uses AI outpainting instead of cropping

• Maintains original frame timing and motion

• Supports prompt-based scene extension

• Optimized with Wan 2.1 + LoRA acceleration

Output

The result is a fully reconstructed horizontal video where the original vertical content remains centered and the newly generated areas blend naturally with the scene.