Upscaled via Stable Diffusion, Topaz Gigapixel A.I. and REAL-ESRGAN, combined results in Photoshop. WARNING: Upscale used A.I. reimagining, so the details aren’t 1:1 to the original.
So this pretty much a new upscaling workflow given how tiny the original image is, [like really tiny (498×352)](https://imghut.com/i/angies-groves-by-eizin-suzuki-1988-v3.zauCk). This came from the idea I had when I’ve been using Ultimate SD Upscale to upscale my other A.I. stuff. So with the upscaler set to 1024×1024 tiles, 50 for padding, 25 for mask blur and the prompts/config like this:
* Prompt: masterpiece, best quality, scenery, 3 cars parked outside a store, eizin suzuki, american store, blue van on the left, red 60s muscle car in the middle, yellow 60s muscle car on the right, parking, sidewalk in the foreground, blue clear sky
* Negative prompt: lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, 1girl, humans
* Config: Steps: 50, Sampler: DPM++ SDE Karras, CFG scale: 11, Upscaler: R-ESRGAN 4x+ Anime6B
* Model: MIX-Pro-V3
Additionally I tested between setting the denoising strength to 0.35, 0.4 and 0.45, and in short I found 0.35 great at retaining the original details to the best whilst looking sharp (anything lower will probably look blurry). 0.45 is where I started to find reimagined bits advantageous, although destructive, such as detailed highlights, car bits and posters with less garble, as such I selected masked bits I found fitting. 0.4 on the other hand was in a weird in-between where it wasn’t as accurate as 0.35 in terms of retaining original detail whilst not looking as sharp/reimagined as 0.45, as such I just ditched it.
Also fetched some bits from an upscale via Topaz Gigapixel/REAL-ESRGAN that keeps the original details more faithfully, but isn’t as sharp. All composited in Photoshop, color graded to look more like the scan version [from Eizin Suzuki’s site](http://gallery.eizin.co.jp/?p=303).
![Ocean Drive by Eizin Suzuki (1988) [2560x1440]](https://wallpaperdump.cf/wp-content/uploads/2021/03/59qq2zz7ynb31-1240x698.png "Ocean Drive by Eizin Suzuki (1988) [2560x1440]")
![Passing Through California by Eizin Suzuki, 1985 [3840x2160]](https://wallpaperdump.cf/wp-content/uploads/2020/11/e6n29iw69o161-1240x698.png "Passing Through California by Eizin Suzuki, 1985 [3840x2160]")
![Holiday in Southern California by Eizin Suzuki [3840x2160]](https://wallpaperdump.cf/wp-content/uploads/2023/04/i2yop926e3qa1-1240x698.png "Holiday in Southern California by Eizin Suzuki [3840x2160]")
![原画 PINE KNOT MOTEL, since 1982 by Eizin Suzuki [3840x2160]](https://wallpaperdump.cf/wp-content/uploads/2023/04/i5sx05wdv5ma1-1240x698.png "原画 PINE KNOT MOTEL, since 1982 by Eizin Suzuki [3840x2160]")
![Rose Red Roadster by Eizin Suzuki, 1989 [1440x3040]](https://wallpaperdump.cf/wp-content/uploads/2023/04/vyr6hd1azxc91-1240x2618.png "Rose Red Roadster by Eizin Suzuki, 1989 [1440x3040]")
![CALIFORNIA CHARTER, since 1982 by Eizin Suzuki [2560x1440]](https://wallpaperdump.cf/wp-content/uploads/2023/04/pgcsz1wn68191-1240x698.png "CALIFORNIA CHARTER, since 1982 by Eizin Suzuki [2560x1440]")
acoolrocket March 3, 2023
[Original color grade](https://imghut.com/i/angies-groves-by-eizin-suzuki-1988-orig-grade.zaqhn)
You can find [my other upscales of Eizin Suzuki here](https://imghut.com/album/eizin-suzuki-upscales.LNdH) and [wallpaper versions here](https://imghut.com/album/eizin-suzuki-upscales-169-wallpapers.LTyC). Or [Eizin Suzuki’s site](http://gallery.eizin.co.jp/) for the full catalog.
—
Upscaled via Stable Diffusion, Topaz Gigapixel A.I. and REAL-ESRGAN, combined results in Photoshop. WARNING: Upscale used A.I. reimagining, so the details aren’t 1:1 to the original.
So this pretty much a new upscaling workflow given how tiny the original image is, [like really tiny (498×352)](https://imghut.com/i/angies-groves-by-eizin-suzuki-1988-v3.zauCk). This came from the idea I had when I’ve been using Ultimate SD Upscale to upscale my other A.I. stuff. So with the upscaler set to 1024×1024 tiles, 50 for padding, 25 for mask blur and the prompts/config like this:
* Prompt: masterpiece, best quality, scenery, 3 cars parked outside a store, eizin suzuki, american store, blue van on the left, red 60s muscle car in the middle, yellow 60s muscle car on the right, parking, sidewalk in the foreground, blue clear sky
* Negative prompt: lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, 1girl, humans
* Config: Steps: 50, Sampler: DPM++ SDE Karras, CFG scale: 11, Upscaler: R-ESRGAN 4x+ Anime6B
* Model: MIX-Pro-V3
Additionally I tested between setting the denoising strength to 0.35, 0.4 and 0.45, and in short I found 0.35 great at retaining the original details to the best whilst looking sharp (anything lower will probably look blurry). 0.45 is where I started to find reimagined bits advantageous, although destructive, such as detailed highlights, car bits and posters with less garble, as such I selected masked bits I found fitting. 0.4 on the other hand was in a weird in-between where it wasn’t as accurate as 0.35 in terms of retaining original detail whilst not looking as sharp/reimagined as 0.45, as such I just ditched it.
Also fetched some bits from an upscale via Topaz Gigapixel/REAL-ESRGAN that keeps the original details more faithfully, but isn’t as sharp. All composited in Photoshop, color graded to look more like the scan version [from Eizin Suzuki’s site](http://gallery.eizin.co.jp/?p=303).
[All results/generations can be found in this album.](https://imghut.com/album/angies-groves-by-eizin-suzuki-1988.MIZC)