Innovate Futures @ Benji

(Workflow) Skyreel-V1 Img2Vid Ver.20250221 Update

Added 2025-02-21 12:58:43 +0000 UTC

Related Video:

https://www.patreon.com/posts/122771424

Summary of Workflow with Second Pass Sampling for Better Quality

The workflow begins by setting up the ComfyUI interface and loading the necessary models, including the Hanyuan video model fine-tuned for image-to-video generation. The first step involves applying a smoothening Lora to ensure smoother transitions between frames.

Two samplers are used in this process. The first sampler generates a base image that tends to be pixelated and low-quality but establishes the initial structure and motion. This output is then passed to the second sampler, which refines the quality, enhances details, and smooths out animations, resulting in a more polished video.

For the sampling methods, DPM PP2M is preferred in the first sampler, while Eruller with a simple scheduler is used in the second sampler to focus on refining video quality without introducing too much creativity. Both sampling groups are set to 20 steps, although users can experiment with higher step counts for potentially better results.

After the second pass, sound effects are added using MM audio. The entire process allows for step-by-step refinement, ensuring each part of the generation meets acceptable standards before moving forward. This method prevents wasting time on multiple full generations and allows for targeted adjustments.

Finally, the workflow emphasizes that while ComfyUI is essential, it's part of a larger system that includes AI agents for scriptwriting, character creation, and scene generation, all working autonomously to create complete video stories.