Prompt: Detect all forms of censorship in the image — including mosaic, blur, black or white bar censor, pixelation, filled areas, and identity censor — and accurately restore the censored regions. Prioritize faithful reconstruction of the original uncensored content based on logical structure and visual context. Ensure that the restored areas retain the intended texture, shape, and detail of the uncensored version.

※ This LoRA is an early experimental work, so it may not necessarily perform better than other methods.

The reconstruction quality is not very high. Please treat this as an experiment to demonstrate feasibility. Inpainting with other models may yield higher accuracy.

■ If you are a new user of Kontext, please try installing it by following the instructions at the URL below.

It has detailed documentation and is easy to set up.

https://docs.comfy.org/tutorials/flux/flux-1-kontext-dev

■ Reconstructing censored regions.

Kontext already has the capability to detect and restore such masks, but since it may not always work perfectly, this LoRA was created to support this process.

It improves detection capabilities for mosaics, fills, and lines in sample images.

Since copyrighted material cannot be used, I used AI-generated fruit images as samples.

The sample images are easy to detect and reconstruct due to clear masks and context, but real masked images are usually more complex and harder to process.

■ The prompt below has high mask detection capabilities and serves as a good starting point.

It was also used as the caption during LoRA training.

Detect all forms of censorship in the image — including mosaic, blur, black or white bar censor, pixelation, filled areas, and identity censor — and accurately restore the censored regions. Prioritize faithful reconstruction of the original uncensored content based on logical structure and visual context. Ensure that the restored areas retain the intended texture, shape, and detail of the uncensored version.

■ It works well on its own, but feel free to adjust the prompt to better match your expected results. However, adjusting the prompt alone has limited effect on improving detection and reconstruction quality.

■ High reconstruction quality is not guaranteed.

Thin masks like lines are easier to infer from the surrounding context, but fully covered areas may fail to reconstruct properly due to a lack of detail or texture cues.

■ If there are too many masked areas, the model may fail to recognize what needs to be restored.

In such cases, splitting the image and simplifying the masked regions can help improve results.

Try cropping only the area around the part you want to remove.

Color correction can also help improve detection beforehand.

For example, scanned images may be faded or low-contrast, so adjusting contrast is effective.

Masks filled with dark gray or indistinct tones are also easier to detect if adjusting contrast turns them into solid black.

If print dots are visible, they may interfere with detection. In some cases, redrawing clean black mask lines yourself can improve clarity.

■ Additionally, as this LoRA is an early experiment, it may restore general shapes but struggle to recover realistic textures or fine details.

We apologize if the results fall short—please understand its limitations.

For better results, using inpainting or 0.25–0.5 denoising i2i with another model may be more effective.

SD1.5 is lightweight, performs well even at high resolutions, and is an excellent choice for i2i tasks.

It works as a great refiner and can sometimes produce better results than expected.

If you have enough VRAM, SDXL is also a good option.