This model is a merge of ComicCraft-LCM and DreamShaper-LCM.
After doing some tests with this new DreamShaper-LCM, I noticed two things:
the model works great, resulting images can be very clear/sharp at very low sampling steps.
however, at lower sampling steps, some images tend to be very dark (due to the noiseoffset), almost a fully black image. In those cases it will need around 15 sampling steps to generate more content in the image.
Since I had the experimental ComicCraft-LCM, I wondered how a merge of both LCM models would look like and after some tests and taking parts of each model, I made this version that I think can be interesting to use.
To describe the model in comparison to the original ones:
Use CFG between 1 and 2. Images are clear from 6 sampling steps, but more steps will make images less dark and more detailed.
The base style is less realistic than DreamShaper's, but it's pretty good at semi-realism (maybe also in the direction of 3D render). By using TIs or LoRAs, it can become more realistic.
In general it's less colorful than the ComicCraft models, since the "comics" influence is lower and also due to the strength of the noiseoffset. It can still be generated with prompting and higher sampling steps.
Like ComicCraft-LCM, it works at low sampling steps (6) and with most samplers (the same comments from ComicCraft-LCM apply in this case) . However, the best results will be with the LCM samplers. When using Automatic1111, the AnimateDiff extension is still needed, because it bundles the LCM sampler (it may not work correctly yet on the DirectML version of A1111).
While results will be less dark at low sampling steps (6) compared to DreamShaper-LCM, they can still be darker than intended. For these cases, increasing the steps to around 8-12 steps will help.
It works pretty well with !Adetailer, HiRes fix and Ultimate SD Upscale.
