A Proposed Workflow - MidJourney - Photoshop - Magnific
Note: this is a work in progress, and I’m learning more each day, so this information may change regularly.
I’ve been testing a workflow that lets you preview at 5 tokens and then build once at a larger size. I’d appreciate any feedback or comments.
I start with an initial Midjourney output from the prompt, “a single photographer wearing a red jacket holding a camera shooting the sea while standing on a cliff top in a storm --style raw --v 6.0”
In Midjourney, I opened the 4-up view, viewed the output in a browser, and downloaded the 4-up image. So no upscaling in Midjourney.
I’ve extracted the V4 image from the 4-up, and it's 1024 x 1024 px, which means it falls into the 1.29 mp range that consumes 5 tokens on upscale.
I’ve loaded the Midjourney image into Magnific and set the parameters for the upscale as per the image below.
Note: the final size will be 2048 x 2048 px and the cost = 5 tokens
The upscaled image is downloaded and is shown below.
This compares the 1024 x 1024 Midjourney image (upscaled in Photoshop to match the Magnific image size) to the 2048 x 2048 Magnific upscaled image.
There is more detail in the photographer’s clothing, camera and the sea and rocks are more defined.
Note: at this point the largest upscale I can do on this image in Magnific is 4x or 8192 x 8192 px and still remain under the 25mp ceiling. But what if I want a larger version, the maximum 10,000 x 10,000 px?
Looking at the Token Usage chart above, the only sizes able to scale in one go to the maximum size, are the 25 mp and the 0.39 mp versions.
To achieve the maximum size, I will open the 2048 x 2048 px file in Photoshop and downscale it in 2 steps to preserve detail. The first is 1024 x 1024 px, and the second is 625 x 625 px.
I will save the 625 x 625 px and open it into Magnific and set an upscale of 16x for a cost of 90 tokens.
Note: this time I want to reduce the incidence of change and keep it as close as possible to the original. So I’ve changed the parameters for upscale as shown in the image below.
The result is shown below. Once again, the detail in the sea, rocks and the photographer are great, but the sky is showing a number of artifacts, and the camera is just wrong.
Note: Magnific has a tendancy to add weird little artifacts from time to time. Can you spot the faces in the rocks? Try the full resolution version and find all three faces.
To fix it I hop back into Midjourney and generate some new images with the prompt “man, wearing gloves, holding a canon camera, taking a photograph, side view, close up --v 6.0 --style raw --s 50”
From the 4-up output, I’ll use V3 and strip it into the 10,000 x 10,000 px Magnific upscale version. For the sky, I’ll take the sky from the 2048 x 2048 px image and scale it to fit. Oh, and I’ll remove the hidden faces from the rocks.
Here is the final image.
If you have any comments you can leave them on this page or contact me at tony@tonyredhead.com