Here’s my second attempt at an Illustrious merge. This time, I wanted to see how merging LoRA and LoCon models into the checkpoint would work. I think the results turned out pretty good. It responds well to the artist tags I’ve tried, but I haven’t tested all of them.Let me know if there are any artist/style tags you’d like to use that don’t work as well as others! I’m happy to fine-tune this checkpoint after more testing and hearing your feedback.<h3 id="i-value-your-feedback-both-positive-and-constructive.-i'd-really-appreciate-it-if-you-could-be-as-honest-as-possible.-c3972uema">I value your feedback, both positive and constructive. I’d really appreciate it if you could be as honest as possible.</h3>Personally I’ve been experimenting with <a rel="ugc" href="https://github.com/CFGpp-diffusion/CFGpp">CFG++</a> samplers. I like going Euler A CFG++ with a scale of 0.5 to 0.7 and around 35 steps.With non-CFG++ samplers, I’d run a scale of 3.5 to 7 and between 20 and 50 steps.<h3 id="prompt-the-same-way-as-other-illustrious-models-2u5usi3se">Prompt the same way as other Illustrious models;</h3><pre><code>(subject, {1boy/1girl/2boys/2girls/etc.}, artist/style).

Describe image with natural language. {A portrait of ...}.

(tags, quality tags).</code></pre>This is technically an EPS model, but the results are practically identical with and without EPS sampling so it's not needed.

Files with SHA256: f56d68fe4af1c809c9a9f74293405e555258821a99630949c43e20bacd80585d