This extension for AUTOMATIC1111's Stable Diffusion WebUI enables the creation of pseudo-animations by interpolating between different prompts within the latent space. It targets users looking to generate sequential images with smooth transitions, offering a novel approach to animation and visual storytelling directly within the Stable Diffusion workflow.

How It Works

The core mechanism involves interpolating the conditioning vectors derived from multiple prompts. By gradually shifting these vectors, the extension generates a sequence of images that appear to transition between the input prompts. It supports various interpolation modes like linear interpolation and replacement, with options for token-wise or channel-wise vector manipulation to control the smoothness and nature of the transitions.

Quick Start & Requirements

• Installation: Install via the "Install from URL" tab in the Stable Diffusion WebUI Extensions tab, using https://github.com/Kahsolt/stable-diffusion-webui-prompt-travel.git.
• Prerequisites: Requires AUTOMATIC1111/stable-diffusion-webui v1.5.1 and Mikubill/sd-webui-controlnet v1.1.229. For video generation, a post-processing pipeline with tools like Real-ESRGAN and FFmpeg is recommended, with an optional auto-installer for Windows.
• Documentation: README_ext.md

Highlighted Details

• Supports SDXL v1.0 models.
• Experimental ControlNet integration for interpolating between hint conditions.
• Optional post-processing pipeline for enhanced smoothness using Real-ESRGAN and RIFE.
• Offers various interpolation modes including linear, replace, and "embryo" genesis.

Maintenance & Community

• Active development with frequent updates and bug fixes, keeping pace with Stable Diffusion WebUI changes.
• A QQ chat group is available for suggestions, discussions, and bug reports (ID: 616795645).

Licensing & Compatibility

• The repository is licensed under the MIT License.
• Compatible with AUTOMATIC1111/stable-diffusion-webui and sd-webui-controlnet.

Limitations & Caveats

• May not support schedule syntax like [prompt:prompt:number].
• Likely incompatible with hires.fix due to conceptual conflicts; batch upscaling and img2img are suggested alternatives.
• Video generation requires separate tool installation and can be resource-intensive.