Mosaic with Narrowband Images of the Veil Nebula

I planned a 3x2 mosaic, which spreads about 2.3° x 3.2° - an area of more than 30 full moon areas. I am using the Quattro with a 0.86 reducer/flatterner, resulting in a quite fast telescope - thus I could plan for only one clear night per panel with about 3 hours of integration time. The camera is a ZWO ASI 294 MC pro @ -16°C and for the Ha / OIII lines I am using an Optolong L-eXtreme dual-band filter.

After the image acquisition and stacking with SiriL, I tried the approach I used for the Elephant's trunk (see here) with one major change: I wanted to implement BlurXterminator in the workflow. My standard workflow with PixInsight for narrowband images is currently the following:

1. Crop Ha, transfer the crop to OIII
2. Clone both images
3. Use BlurXterminator on Ha / OIII only non-stellar for without sharpening of the stars (as the next step might sometimes fail)
4. Remove the stars with Starnet2
5. If necessary, do some noise reduction, e.g. with MLT.
6. If necessary, do some background flattening, e.g. with DBE.
7. Make images non-linear
8. Combine Ha/OIII to an RGB image with PixelMath, using the technique described here.
9. Apply BlurXterminator to the Ha/OIII clones and sharpen the stars.
10. Create a star mask from the OIII clone using Starnet2
11. Combine the clones using PixelMath for a natural star color (requires some trial and error).
12. Use the star mask to combine the nebula image with the stars
13. Use curve transform to boost the stars and adjust the colors of the nebula.

When using this process for a mosaic, a couple of additional steps are required:

• Plate solve the panels using ImageSolver. This can fail on the Ha/OIII images, so I combined them with PixelMath to a color image, which works for the image solver, and then transferred the results to the Ha/OIII images using the script CopyCoordinates.
• I tried to first combine the Ha and OIII images with MosaicByCoordinates and GradientMergeMosaic, resulting in two large Ha / OIII mosaics, and combine them with the process described above. However, this approach failed, when I applied BlurXterminator to these mosaics - the image was full of artifacts, noise, strange stars. Seemed like the BlurXterminator AI has issues with the overlapping areas of the mosaic.
• So I tried a different approach and applied BlurXterminator to the Ha / OIII images of each panel separately.
• From there, the process continued with MosaicByCoordinates and GradientMergeMosaic, resulting in an Ha and an OIII mosaic image, which could then be processed as described above (skipping steps 3 and 9).

Below: Resulting mosaics of the H-alpha (left) and OIII images (right):

The star removal with Starnet2 revealed some issues in the H-alpha image: A strange background in the upper right corner which are more typical for the OIII signal in e.g. Bortle 4 areas and some issues in the average-merge of the GradientMergeMosaic script.

I was able to remove the background carefully with DBE and merge them with PixelMath. Finally, I added the stars as described above.

The resulting image still had some blue tint in the background, which I removed in Photoshop PS2 and did some clone stamp to remove areas where the image stitching showed blurry results in the star background.