Upscale Images While Preserving Fine Detail
Enlarge photos and compare the recovered detail against the original. FideSR was added in CNAPS Studio v1.2.2. Uses one-step diffusion targeting detail preservation and faithful reconstruction. Runs on CNAPS infrastructure; no external API key required.
What it does
FideSR is an image super-resolution option added in CNAPS Studio v1.2.2. The upstream project, styled FiDeSR, targets detail preservation and faithful reconstruction with one-step diffusion. Use it when you want an enlarged image and need to inspect whether fine structures remain consistent with the source. It runs on CNAPS infrastructure in cloud Studio; no local installation is needed.
Problem it solves
- Larger deliverables — Prepare a small source for a larger layout or output.
- A second restoration option — Compare the result against another upscaler on the same input.
- Consistent processing — Reuse the same flow across related images.
- Detail review — Inspect texture and edges before applying the model to a whole batch.
Input/Output
- Input: An image from Image Loader or an upstream image-processing node.
- Output: An upscaled image for Image Viewer, comparison, or final export.
Configuring the Node
Enlargement — Choose from the scale or output-size options exposed by the node.
Enhancement controls — Start with defaults and change one exposed control at a time.
Comparison — Send the original and processed images to Image Compare to inspect the same region.
How to Use
- Connect Image Loader to FideSR.
- Choose an available enlargement setting and keep the remaining defaults.
- Connect the result to Image Viewer and run a single image.
- Compare edges, faces, repeated patterns, and text with the source.
- Apply the approved settings to additional inputs or a Batch Run.
Example Workflow
Flow: Image Loader → FideSR → Image Viewer.
Comparison: Route the original and upscaled result into Image Compare. Review a face, a textured area, and any printed text before deciding which version to keep.
Accuracy & Speed
Execution | CNAPS-hosted; no external provider API key required. |
Processing time | Depends on input dimensions, output size, settings, and GPU availability. |
Detail fidelity | Diffusion-based restoration can change fine detail; compare with the source. |
Batch planning | Measure one representative image before estimating a large batch. |
How it compares
Compare with PiSA-SR, Swin2SR, SwinIR, SMFANet+, and the other new upscaler using the same source and comparable output dimensions. Choose by visible results on your images rather than assuming one model is best for every input.
Technical Details (for engineers)
The official FiDeSR repository accompanies a CVPR 2026 paper. The method uses detail-aware training, frequency-dependent enhancement, and latent residual refinement. The upstream example exposes high- and low-frequency controls; those command-line controls should not be assumed to exist in the Studio node.
Model Source
- FideSR — official repository: FideSR — official repository
- CNAPS integration: v1.2.2 release note
Compliance & Provenance
Provider | FiDeSR research team |
License / service terms | The upstream README states Apache 2.0; base models and dependencies retain their own terms. |
Output provenance | For how CNAPS labels processed outputs and tracks upstream generation, see the AI Policy. |
For platform classification, acceptable use, and output marking, see the CNAPS AI Policy.
Run it now
Open CNAPS Studio, find FideSR in the Component panel, and start with the example workflow above.