Ran a quick test of the new Generative Expand in Photoshop Beta. It clearly beat Topaz Gigapixel and the older Neural "Super Zoom" filter.
My method:
1. Grab a high rez image with some fine detail.
2. Crop a 4096 x 4096 pixel section and save that as "Original". This because the current Beta Gen Expand has a 4096 limit.
3. Resize that Original by 50% using Bicubic Sharpener, save as "HalfSize"
4. Feed that HalfSize to Topaz and upscale 2X, save as "Topaz".
5. Feed the HalfSize to Beta Gen Expand at 2X. This returns 2 layers. Ignore/delete the top layer. Save the bottom layer as "BetaGen". More about that top layer later.
6. Feed HalfSize to Neural Superzoom, save as "Neural".
7. Load Original, Topaz, BetaGen, and Neural as layers in Photoshop with Original on the bottom.
8. Put the Topaz, BetaGen, and Neural layers in Difference mode.
Now, display your Histogram and toggle the top layers on one at a time. For each, click on the silly Triangle/Exclamation Point icon in the Histogram window and then read the StdDev value. The StdDev is a measure of how far each layer differs from the Original. Smaller is better. I tested 4 images this way and the Beta Gen Expand won each time, beating Topaz Giga by a wide margin. The difference between Topaz Giga and Neural Superzoom is usually too close to call.
Of couse, the Beta Gen Expand is not worth much until they lift the 4096 pixel limit.
My method:
1. Grab a high rez image with some fine detail.
2. Crop a 4096 x 4096 pixel section and save that as "Original". This because the current Beta Gen Expand has a 4096 limit.
3. Resize that Original by 50% using Bicubic Sharpener, save as "HalfSize"
4. Feed that HalfSize to Topaz and upscale 2X, save as "Topaz".
5. Feed the HalfSize to Beta Gen Expand at 2X. This returns 2 layers. Ignore/delete the top layer. Save the bottom layer as "BetaGen". More about that top layer later.
6. Feed HalfSize to Neural Superzoom, save as "Neural".
7. Load Original, Topaz, BetaGen, and Neural as layers in Photoshop with Original on the bottom.
8. Put the Topaz, BetaGen, and Neural layers in Difference mode.
Now, display your Histogram and toggle the top layers on one at a time. For each, click on the silly Triangle/Exclamation Point icon in the Histogram window and then read the StdDev value. The StdDev is a measure of how far each layer differs from the Original. Smaller is better. I tested 4 images this way and the Beta Gen Expand won each time, beating Topaz Giga by a wide margin. The difference between Topaz Giga and Neural Superzoom is usually too close to call.
Of couse, the Beta Gen Expand is not worth much until they lift the 4096 pixel limit.
