<blockquote><h2 id="what-this-is">What this is</h2>For a year I have been making images on Illustrious with a stack of five style LoRAs and a long, fixed style prompt. It works, but it is heavy: five resources to load, a prompt prefix I paste every time, and a look that only exists as long as I keep that whole rig together.This LoRA is that rig, folded into one file for Anima.It is not a copy. The point was never to reproduce the same pictures — it was to carry the hand over to a base model that understands sentences, composes scenes, and handles more than one person in frame.Trigger word: <code>dwil</code><hr /><h2 id="where-the-training-data-comes-from">Where the training data comes from</h2>337 of my own published images, from 3 March to 25 August 2026. Nothing scraped, nothing borrowed — every image is one I made and posted. 256 of them are 1248×1824, so the model learned from hi-res-fixed output, not from small drafts. That is why it holds up under a second pass.The set is honest about what it is: heavily portrait, mostly single figures, mostly night and interior, and it includes an adult range — the source posts run from Soft through X. A small block of 32 images carries no people at all, deliberately, so that the style would not be welded to "a woman looking at the camera".<hr /><h2 id="the-illustrious-stack-it-comes-from">The Illustrious stack it comes from</h2>Credit where it is due — this look was built on other people's work:0.8 <a target="_blank" rel="ugc" href="https://civarchive.com/models/2100974">Ch4mpi | Shiiro's Styles</a>0.7 <a target="_blank" rel="ugc" href="https://civarchive.com/models/2169329">Modern anime render</a>0.5 <a target="_blank" rel="ugc" href="https://civarchive.com/models/1745229">Cartoon styles — Midjourney S2</a>0.4 <a target="_blank" rel="ugc" href="https://civarchive.com/models/580857">Realistic Skin Texture style</a>0.2 <a target="_blank" rel="ugc" href="https://civarchive.com/models/1324670">Sinozick | Shiiro's Styles | Niji</a>With the main checkpoints been <a target="_blank" rel="ugc" href="https://civarchive.com/models/1629360">IllustriousNXT_XL v2.0</a>, Most of the images at CFG 4, 29 steps, Euler a.And this style prefix, in front of every prompt:<code>Sinozick_illu, gradient, spot color, Carto4on, ch4mpi_illu, modern_anime_render,</code><code>high_detail, glossy_highlights, vivid_colors, dynamic_lighting, atmospheric_depth,</code><code>soft_shading, painterly_texture, clean_lineart, (flat color:2), (no lineart:1),</code><code>(no outline:1), (Flat vector:1), cinematic photography style</code> <code>dwil</code> replaces all of that. One token, one file, on a different base model.<hr /><h2 id="how-it-was-trained">How it was trained</h2><ul><li><code>kohya-ss/sd-scripts</code>, <code>anima_train_network.py</code>, <code>networks.lora_anima</code></li><li>Base: <code>anima-base-v1.0</code></li><li>Rank 16, alpha 16 — rank 32 was tried and made no difference I could see</li><li>4250 steps, 12 epochs, 337 images / 363 views after repeats</li><li>AdamW8bit, LR 1e-4, <code>cosine_with_restarts</code> ×3, bf16, bucketed at 1024</li><li>Captions are booru-style tags, trigger kept in first position</li><li>~4h50 on a single 8 GB card</li></ul><hr /><h2 id="recommended-settings">Recommended settings</h2><ul><li>Base: <code>anima-base-v1.0</code> — weight 1.0</li><li>CFG 4.0, 30 steps, sampler Euler a, schedule Normal</li><li>832×1216 portrait or 1216×832 landscape</li><li>Hi-res fix ×1.5, denoise 0.5 — this is where it looks best, and it is how the training images were made</li></ul>Weight behaves smoothly from 0.6 to 1.2. Lower weight keeps more of the base model's contrast and grain; higher weight flattens further. 1.0 is where I work.<hr /><h2 id="what-it-does-not-do">What it does not do</h2>It flattens. That is the point, but it means this is not a texture or micro-detail LoRA — at full weight it carries less surface grain than the base model, not more. If you want punchy high-contrast output, run it at 0.6 or lighten your prompt.Wide shots flatten more than close-ups. The training set is portrait-heavy and it shows. Group scenes and complex multi-character compositions are not what it was built for, and I have not verified that it helps there.One quirk worth knowing, since it is baked into the source style: my Illustrious prompt asks for <code>clean_lineart</code> and <code>(no lineart:1)</code> at the same time. Those two fight each other, and the flatness you see is partly the outcome of that fight.<hr /><h2 id="measured-not-claimed">Measured, not claimed</h2>Every number above comes from paired measurements — same prompts, same seeds, one factor changed at a time. The gallery includes wide shots, low light, a non-human subject and a six-character scene precisely because those are the hard cases, not because they all succeeded.</blockquote>

V3LOURS

1.0 — first public release.Trained on 337 of my own posts (March–August 2026), rank 16, 4250 steps.Two earlier internal builds did not ship, and what they taught is baked into this one:rank 32 — twice the file size, no visible gain. Reverted to 16.a shorter run — I trained a version at ~14 epochs before this one and could not tell the two apart in blind side-by-side tests. Longer training measurably flattens surface texture, so there is no reason to go further than this.Known limits are listed in the model description. If you find a failure mode I have not named, tell me — that is more useful to me than a compliment.

Files with SHA256: 5dc25d12eb5b4652707e4fde4fb5dacf87e73d58417413200bb64d3cec5c0a7a