<h1 id="knk-helioblend-bt7psc20b">KNK HelioBlend</h1>None of the images in the previews were upscaled, detailed, or treated beyond adjusting the sampling steps. Additionally wasn't used any quality or negative tags on them.This model was created using Illustrious v0.1 as a base, along with synthetic data generated using my previous model, <a target="_blank" rel="ugc" href="https://civarchive.com/models/621361?modelVersionId=694629">KNK CieloBlend (Ponyv6),</a> pushing it to achieve the most aesthetically pleasing results possible.The result is a model capable of producing high-quality generations without the need for extensive negative or quality tags, while offering the flexibility to blend the default style with its internal ones.<h2 id="version-3-(trained):-akzlk1mip">Version 3 (trained):</h2>I leaved apart the lora merging to train this model directly using fine tunning. I've to unprune all the tags from the dataset and were removed several images. Now due the realism reduction, the model needs more tries to get perfect hands but the general performance is still good. Now the general style can blends much better with native and lora styles. They performs much better than before.<h2 id="version-2-(merge):-wu37vrrex">Version 2 (merge):</h2>I noticed several issues with prominent lips and low compatibility with anime styles, which caused various eye distortions. I updated the dataset by including more anime-like characters, and now the model generates better characters without forcing them to have unnatural lips or eye deformations.<h2 id="recommendations-yfzvelz4l">Recommendations</h2><ul><li>Prompt: The majority of the dataset used sunlight settings.</li><li>Negatives: (These aren’t really necessary; I recommend using only those you genuinely need, like "half-closed eyes" when using "embarrassed," for example.)<code>worst quality, comic, multiple views, bad quality, low quality, lowres, bad anatomy, bad hands, scan artifacts, twitter username, jpeg artifacts, 2koma, 4koma, extra digits, fewer digits</code></li><li>Recommended Resolutions for XL:<ul><li>(1024, 1024)</li><li>(1152, 896)</li><li>(896, 1152)</li><li>(1216, 832)</li><li>(832, 1216)</li><li>(1344, 768)</li><li>(768, 1344)</li><li>(1536, 640)</li><li>(640, 1536)</li></ul></li></ul>For more information about the model and additional recommendations, you can check the repository on <a target="_blank" rel="ugc" href="https://huggingface.co/OnomaAIResearch/Illustrious-xl-early-release-v0">Hugging Face</a>.

KNK HelioBlend (Illustrious v0.1)

knkHelioblend_v30_full_fp16.safetensors

knkHelioblend_v30_full_bf16.safetensors

v3.0

v2.0

KNK HelioBlendNone of the images in the previews were upscaled, detailed, or treated beyond adjusting the sampling steps. Additionally wasn't used any quality or negative tags on them.This model was created using Illustrious v0.1 as a base, along with synthetic data generated using my previous model, <a href="https://civitai.com/models/621361?modelVersionId=694629" rel="ugc noopener noreferrer nofollow" target="_blank">KNK CieloBlend (Ponyv6),</a> pushing it to achieve the most aesthetically pleasing results possible.The result is a model capable of producing high-quality generations without the need for extensive negative or quality tags, while offering the flexibility to blend the default style with its internal ones.Version 3 (trained):I leaved apart the lora merging to train this model directly using fine tunning. I've to unprune all the tags from the dataset and were removed several images. Now due the realism reduction, the model needs more tries to get perfect hands but the general performance is still good. Now the general style can blends much better with native and lora styles. They performs much better than before.Version 2 (merge):I noticed several issues with prominent lips and low compatibility with anime styles, which caused various eye distortions. I updated the dataset by including more anime-like characters, and now the model generates better characters without forcing them to have unnatural lips or eye deformations.Recommendations<ul><li>Prompt: The majority of the dataset used sunlight settings.</li><li>Negatives: (These aren’t really necessary; I recommend using only those you genuinely need, like "half-closed eyes" when using "embarrassed," for example.)worst quality, comic, multiple views, bad quality, low quality, lowres, bad anatomy, bad hands, scan artifacts, twitter username, jpeg artifacts, 2koma, 4koma, extra digits, fewer digits</li><li>Recommended Resolutions for XL:<ul><li>(1024, 1024)</li><li>(1152, 896)</li><li>(896, 1152)</li><li>(1216, 832)</li><li>(832, 1216)</li><li>(1344, 768)</li><li>(768, 1344)</li><li>(1536, 640)</li><li>(640, 1536)</li></ul></li></ul>For more information about the model and additional recommendations, you can check the repository on <a href="https://huggingface.co/OnomaAIResearch/Illustrious-xl-early-release-v0" rel="ugc noopener noreferrer nofollow" target="_blank">Hugging Face</a>.

Files with SHA256: d7577810c327461be5441e0e376741926f84691eff53e33c7fe213456699655c