Image Upscale
This app upscales your image using the best open source upscaling model, SeedVR2.
例を試す
このワークフローについて
This ComfyUI workflow performs high-quality single-image super-resolution using the open-source SeedVR2 upscaling model. It’s a minimal, production-friendly pipeline: LoadImage ingests your source image, the SeedVR2 Upscale node (c35dd4ba-0fe3-47d6-afce-1c56d9fe5beb) increases resolution while reconstructing fine detail, and SaveImage writes the result to disk. Because the graph focuses solely on upscaling, there are no prompts, samplers, or diffusion steps involved—making it fast, repeatable, and easy to tune.
Under the hood, SeedVR2 applies a learned mapping that predicts high-frequency detail at a higher resolution, preserving the original composition and color while improving sharpness and texture. In practice, you load your image, choose the scale factor exposed by the SeedVR2 node (commonly 2x or 4x, depending on how the node is configured), queue the job, and collect the output. This makes the workflow ideal for preparing assets for print, upscaling renders, or polishing AI-generated images without altering their content.
よくある質問
No. It uses a learned upscaler (SeedVR2) to reconstruct detail at a higher resolution without prompts or diffusion steps. Composition and colors are preserved, though mild sharpening and texture recovery can occur.
The available scale factors depend on how the SeedVR2 Upscale node is configured in this workflow (commonly 2x or 4x). Open the node to see the options. If you need very large outputs, consider running 2x twice.
SaveImage writes files to your ComfyUI/output directory using the filename prefix you set in the node. Check that folder after you click Queue Prompt.
Try a lower scale factor, close other GPU-intensive apps, or upscale in two passes (e.g., 2x then 2x). A GPU is recommended; CPU execution can work but is significantly slower.