Sign In

SpectraSynth_SDXL_TCD Reviews

Model does what it promises and delivers results in 2 to 5 steps. It is very sensitive to the chosen sampler settings (understandably with so few steps). And it can quickly result in too much contrast, too busy backgrounds or something that almost looks like sensor noise on cheap cameras. I think a model like this shines when you quickly want to experiment with different prompts or just get a lot of ideas out, but for general use I find it a little too unstable. That said it is one of the few models that really allows acceptable results in 2 steps. So for that it gets a positive review.

Really cool model that manages incredibly with only 2 steps and even better with more. The uploads here are all made with dpm++_2s_a sgm_uniform in 2 steps and cfg of 1-1.1 with a 2-3 pass workflow depending on the image. With 2 passes I upscale once by 2x, with 3 passes I upscale each time by 1.5x

The ones containing characters I sent through a bunch of ADetailers as well, all set to 2 steps cfg 1 dpm++_2s_a sgm_uniform.

The prompt adherence is great for the speed it works at and details can turn out great with multi-pass workflows. In some cases I played around with a sampler switch mid workflow as dpm++_2s_a is adding a lot of detail and contrast and just looks like a higher cfg, whereas euler_a is a lot smoother and uniform, basically the opposite, which can be used to adjust the smoothness when using multiple passes.

I'm looking forward to trying out some more settings with this and to future developments. Thank you very much for creating this!