<pre><code>Sampler: dpmpp_2m
 Scheduler: Beta
 23-28 steps </code></pre><pre><code>No lora has been used in any image.</code></pre><a target="_blank" rel="ugc" href="https://mikogen.com">Mikogen.com </a>(Create comics and/or generate images) <img src="https://e.deviantart.net/emoticons/d/dummy.gif" alt=":dummy:" width="21" height="15" /> (Try it for free with daily free credits).<h3 id="to-achieve-quality-results-a-base-resolution-of-10241536-is-recommended-(before-any-kind-of-hi-res-fix-the-shorter-side-should-be-1024-px-or-higher).">To achieve quality results, a base resolution of 1024×1536 is recommended (before any kind of hi-res fix; the shorter side should be 1024 px or higher).</h3><h2 id="v1.0-balance-mix">v1.0 Balance Mix</h2><h3 id="this-version-is-a-middle-ground-between-the-&quot;super-detailed&quot;-and-the-&quot;extreme-mix&quot;-versions.">This version is a middle ground between the “super-detailed” and the “Extreme Mix” versions.</h3>It is not a merge of the two models, but a new model with improvements taken from both plus new ones (mainly for better hands and feet).The model is still very good for backgrounds and lighting, yet the most impressive results in terms of realism combined with anime faces remain somewhat exclusive to the “Extreme Mix” model.Instead, this one is much more “friendly” to anime styles (anime screencap, flat colors, anime illustration, no lineart, etc.).Usually, placing any of those tags at the beginning is more than enough and the outputs look less overloaded.Backgrounds stay highly detailed, but sometimes, especially when using the “intricate” tag, they can become too abstract.This model is easier to prompt than the Extreme version. Overall, it is probably a much more versatile model that is easier to prompt for a wider range of styles and environments.<h2 id="v1.0-extreme-mix">v1.0 Extreme Mix</h2>This “Extreme Mix” version combines every improvement from the previous two models with a much more intense boost in detail level, background quality, and overall fidelity.Because of that, the model can be a bit harder to prompt—especially for flat anime styles—but it’s perfectly doable.For a flat anime style with this model, the key is to include these tags:(anime screencap:1.25), (flat colors:1.4)both at the beginning and at the end of the prompt. (anime coloring) is also an option.(These are ComfyUI weights; the values should be much lower depending on the system and the length of the prompt. If the prompt is short, the weights should be reduced to avoid oversaturation.)By default the output may lean quite “2.5D” or look more adult.To make characters appear younger you can add the tag “(young:1.35)” (or others like “big eyes”, “big head” that the model links to younger characters without distorting proportions).For shiny materials, reflections, particles, dark scenes and high detail, the results can be straight-up sensational with the right prompting.<h2 id="v2.0-superdetailed">v2.0 Superdetailed</h2>This version is much more refined and generates more consistent and stylized results as standard.It supports all types of styles with the appropriate prompting while maintaining high quality and detail. Version 2.0 performs better in dark scenes. The style is slightly more “moe” than version 1.0.The maximum possible detail in backgrounds may still be slightly higher in version 1.0.==================<h2 id="v2.0-anime">v2.0 Anime</h2>This version improves the basic detail even with heavy weights for anime such as “anime screen” or “anime style.” This version attempts to approximate the “Nijijourney” style.Apart from general improvements, the style tends slightly toward a more moe or adorable style (but remains just as versatile for all types of styles, from neutral to anime or semi-realistic). ==================v1.0 Anime and v1.0 Superdetailed: Initial release. About: A merge centered on anime style but highly adaptable, especially to dark lighting, detailed backgrounds, or even semi realistic styles.All sample images are generated as follows:Steps: 23 CFG: 7.5 Sampler: dmpp_2m Scheduler: betaBase resolution: 1024x1536Latent upscale x1.29Second KSampler (for hiresfix) with the same settings as the first, using a denoise value of 0.45-0.52 Prompting is very important.Use style reinforcement tags (anime, flat colors, anime screencap, etc.) at the beginning OR end of the prompt. (In the end, it is necessary to repeat it if the prompt is very long. With this model, there will be no degradation for doing so, nor will there be any “burn-in.”)For backgrounds, it's best to add various background quality tags at the end of the prompt.A merge centered on anime style but highly adaptable, especially to dark lighting, detailed backgrounds, or even semi realistic styles.

Mikomic Anime

This model creates extreme detail, but it is generally very dependent on the second hires-fix pass. If you want quality results like the examples, use 1024 or more for the shorter side of the image and a hires-fix pass (x1.29 upscale with 10-12 steps is what was used in the samples for hires-fix).For anime style, you need to repeat several times with tags such as (anime screencap:1.25) and (flat colors:1.4), usually at the beginning and end of the prompt. (Comfyui weights)

Files with SHA256: 5c64308c8bcc2a61332322a9be20cf9e8bc066d87e4f7363d5c50e97b03d6a35