Another ‘Poser to AI’ style success. A while back I made some Poser renders with Nursoda’s Chull figure, to test with Flux Kontext and style transfer. It was a failure. However we now have Flux2 Klein 4B, made in Germany, which is dual-use as it also has a wonderful Edit mode. There are only a few style LoRAs for 4B, but tonight I dug out my Chull renders and I find that the model is so flexible that one can just prompt for a style and get it consistently.
Here we see Chull being given a style makeover by Klein with just a short instructional prompt and a fixed seed. Klein’s Edit mode is working much like an old-school Photoshop filter. The result is a little washed out, but the registration between the original Poser render and the Klein is fairly exact… which means that Photoshop can easily correct that. Just lay the Poser source render on top of the Klein output, and set the ‘Multiply’ layer-blending mode.
The end result is quite appealing. At least when seen at a large size. Like all artwork with very fine lines, it’s not graphic enough to work well at small sizes. Such as in a small comic panel. But for a full-page children’s storybook illustration, it has a lot of appeal.
Problems are that Klein retains the 3D ‘ribbon style’ hair, and that the gaze direction has shifted. Gaze direction was prompted for, asking for it to remain fixed. But in the full-length Klein output we still get the dreaded “must stare at the camera” problem. In the close-up Klein output one of the eye pupils has been squished a little. Possibly Klein doesn’t understand ‘gaze’, though.
Still, not bad for a first proper attempt at a LoRA-less style makeover with Klein.