CIVITAI / Workflows

FFGO: unlocking the multi-image reference potential of Wan2.2

This workflow uses the lightweight FFGO model as a way to unlock the multi-image reference potential hidden inside Wan2.2. While many reference-based models exist, very few are specifically aligned with the 2.2 architecture, and FFGO doesn’t change the style—it simply makes Wan2.2 more willing to follow multiple reference images simultaneously. By merging several reference elements into a single composite first frame, the model receives a concentrated bundle of structural and visual cues, and Wan2.2 then completes the missing areas based on its own internal capability. The method does not rely on training biases, and it does not enforce a specific direction; if there is any bias at all, it comes from Wan2.2 itself. High-noise and low-noise LoRAs are not recommended, and accelerated LoRAs can noticeably reduce quality, so the workflow remains intentionally restrained to preserve the core strengths of the base model. In practice, the entire pipeline is still just a standard Wan2.2 image-to-video process—the only difference is that the input becomes a fused first frame made from multiple images. This direct fusion avoids the quality loss that would normally come from splitting elements, running them through editing models, and recompressing them before video generation. Wan2.2 performs best at a recommended resolution of 1280×720 (or the reverse) and around 81 frames, producing the clearest and most stable results. The first few frames of the output typically reflect the reference composition rather than the actual motion, so trimming out the beginning is expected. Compared with older Wan2.1 reference methods such as Phantom, MagRef, or BindWeave—which often produced limited motion or reduced similarity—this approach is far more straightforward and reliable. By combining a clean composite first frame with FFGO’s ability to trigger Wan2.2’s internal reference behavior, the workflow offers a practical and effective way to generate videos that keep multiple elements integrated with high similarity and minimal quality loss. 🎥 YouTube Video Tutorial Want to know what this workflow actually does and how to start fast? This video explains what the tool is, how to launch the workflow instantly, and shares my core design logic — no local setup, no complicated environment. Everything starts directly on RunningHub, so you can experience it in action first. 👉 YouTube