<h1>☀️ SolAnimaTurbo</h1>
Golden light, bold colour, and skies that glow, in just 12 steps.
SolAnimaTurbo is SolAnima with the TURBO for ANIMA distillation LoRA baked in. It keeps SolAnima's bright, saturated, sun-soaked anime (crisp cel shading, glossy highlights and luminous skies) but finishes an image in 12 steps at CFG 1 instead of 30 steps at CFG 4. In our tests that makes each image about four times faster.
It still loves sunrise, high noon and golden hour, and short text in quotes still renders well, so posters and signs are fair game.
No trigger words. Use Anima's usual quality tags and describe the scene.
<hr />
<h2>⚡ The recipe</h2>
<ol>
<li>Base: <a rel="ugc" href="https://civarchive.com/models/3016498">SolAnima</a>, a SLERP merge (t = 0.5) of <a rel="ugc" href="https://civarchive.com/models/3003466">Ghostbubblehybrid V3</a> by sbbell8295 and <a rel="ugc" href="https://civarchive.com/models/2936983">Berry Model Raspberry v5.0</a> by irregular_degular</li>
<li>Baked in: <a rel="ugc" href="https://civarchive.com/models/2619830?modelVersionId=3334160">TURBO for ANIMA (less steps) V4</a> by n_Arno, at strength 1.0</li>
</ol>
Raspberry v5.0 is itself a blend of mommymia's <a rel="ugc" href="https://civarchive.com/models/2968947">Marshmallow Milk Anima</a> and <a rel="ugc" href="https://civarchive.com/models/2788672">Dessert Models Anima (Chocolate Fudge)</a>, according to its file metadata, so their style runs through SolAnimaTurbo too.
TURBO for ANIMA is a DMD2-style LoRA: it teaches the model to finish an image in a few steps without classifier-free guidance. Its weights are folded straight into the model, so there is nothing extra to load, and every other weight is SolAnima's, unchanged. It uses the V4 file made for Anima 2B, the model SolAnima is built on. Merged and converted locally with stable-diffusion.cpp.
<h2>☀️ Settings</h2>
<ul>
<li>Steps: 12. Anything from 8 to 16 works; more steps add a little fine detail.</li>
<li>CFG: 1. You can go up to about 1.4 for a bit more punch, with a few extra steps.</li>
<li>Sampler / scheduler: Euler a + simple</li>
<li>Resolution: 832×1216 portrait, 1216×832 landscape</li>
<li>Negative prompt: has no effect at CFG 1. Above CFG 1, use <code>worst quality, low quality, score_1, score_2, score_3, artist name, signature, watermark, blurry, jpeg artifacts, chromatic aberration, bad anatomy, bad hands, extra fingers</code></li>
<li>Text encoder: <code>qwen_3_06b_base.safetensors</code> (Anima's Qwen3 0.6B)</li>
<li>VAE: <code>qwen_image_vae.safetensors</code></li>
<li>No extra speed LoRA: the turbo is already built in, so don't stack another turbo, DMD2 or lightning LoRA on top.</li>
</ul>
The text encoder and VAE are attached to this version as optional downloads, in case you do not have them yet.
<h2>🌤️ Prompting</h2>
Anima understands Danbooru-style tags and natural language, and SolAnimaTurbo likes both together:
<ul>
<li>Start with <code>masterpiece, best quality, score_7, safe</code> (or <code>nsfw</code>), then <code>1girl</code> / <code>1boy</code> / <code>no humans</code>, then tags for the subject, outfit and setting.</li>
<li>Follow the tags with two or more full sentences describing the scene, the light and the style. Very short prompts give unpredictable results.</li>
<li>Tags are lowercase with spaces instead of underscores. Artist tags need an <code>@</code> in front.</li>
<li>Put text you want in the picture in double quotes: lettering across the top reads "SOLAR EXPRESS". Short, common words come out cleanly; long or unusual words can be misspelled.</li>
<li>Negative prompts do nothing at CFG 1, so steer with the positive prompt: describe what you want to see.</li>
<li>SolAnimaTurbo defaults to vivid and sunny. If you want muted or moody, ask for it.</li>
</ul>
<h2>🌞 Which file?</h2>
<ul>
<li>BF16 · <code>solanimaturbo-v1.0-bf16.safetensors</code> · 4.2 GB: the full-precision master. Every showcase image was rendered with this file.</li>
<li>FP8 (scaled) · <code>solanimaturbo-v1.0-fp8_scaled.safetensors</code> · 2.3 GB: a lighter safetensors file for ComfyUI.</li>
<li>Q8_0 · <code>solanimaturbo-v1.0-q8_0.gguf</code> · 2.2 GB: near-lossless GGUF.</li>
<li>Q5_K · <code>solanimaturbo-v1.0-q5_K.gguf</code> · 1.5 GB: for tighter VRAM, at a small quality cost.</li>
<li>Q4_K · <code>solanimaturbo-v1.0-q4_K.gguf</code> · 1.2 GB: the smallest file; some fine detail is lost.</li>
<li>INT8 · <code>solanimaturbo-v1.0-int8.safetensors</code> · 2.3 GB: for ComfyUI INT8 (convrot) workflows. It also runs natively in recent stable-diffusion.cpp builds.</li>
<li>W4A4 · <code>solanimaturbo-v1.0-int4_convrot.safetensors</code> · 1.8 GB: ComfyUI INT4 W4A4. Experimental, and it does not run in stable-diffusion.cpp yet.</li>
</ul>
Every file is converted directly from the BF16 master, never from another quant. GGUF files need ComfyUI-GGUF, Forge or stable-diffusion.cpp.
<h2>🌇 Content note</h2>
Like SolAnima, SolAnimaTurbo handles mature content when you prompt for it with <code>nsfw</code>. The last showcase image contains nudity; everything else is SFW. All characters are adults. Please follow CivitAI's rules and the laws where you live.
<h2>🌆 Cover prompt</h2>
<blockquote>
masterpiece, best quality, score_7, safe, 1girl, solo, adult, long hair, golden hair, amber eyes, orange kimono, sun print, obi, holding umbrella, oil-paper umbrella, rooftop, tiled roof, east asian architecture, sunrise, cloud, wind, floating hair, full body, from below. A young woman in her twenties with long golden hair and warm amber eyes stands on the tiled rooftop of an old Japanese inn at sunrise, holding an open red paper parasol over her shoulder. She wears an orange kimono patterned with small golden suns, and her loose obi ripples in the morning breeze. Behind her an enormous rising sun fills half the sky, painting the clouds pink and gold, and soft warm light rims her hair.
</blockquote>
12 steps · CFG 1 · Euler a / simple · 832×1216 · seed 1170421
Every showcase image uses the same prompt as the matching image on SolAnima's page, and all but two (the hammock and the fennec fox) use the same seed, so you can compare the two models side by side. Each image keeps its full prompt and settings, so you can remix any of them.
<h2>🌃 Credits</h2>
Thank you to n_Arno (TURBO for ANIMA), sbbell8295 (Ghostbubblehybrid), irregular_degular (Berry Model), mommymia (Marshmallow Milk and Dessert Models), and Circlestone Labs for Anima. SolAnimaTurbo would not exist without their work. 🧡

SolAnimaTurbo

v1.0: solar flare ⚡
<ul>
<li>SolAnima v1.0 with TURBO for ANIMA V4 (n_Arno) baked in at strength 1.0</li>
<li>Recommended: 12 steps · CFG 1 · Euler a / simple · 832×1216</li>
<li>Seven formats: BF16, FP8 (scaled), Q8_0, Q5_K, Q4_K, INT8 and W4A4 (INT4 convrot), all converted from the BF16 master</li>
<li>The text encoder (Qwen3 0.6B) and VAE are attached as optional components</li>
</ul>

Files with SHA256: 6a40fa1e5e5a2a79280abd0d8ae193002ab9d3aa19ef459560480908b5cfd767