<h2 id="homochi-xl-v2-quick-faq">Homochi XL v2 – Quick FAQ</h2> Homochi XL is thesuccessor to the MelloMochi SD1.5 model, optimized for anime, bara, and yaoi‑style men with strong anatomy, clean lines, and stable style control. It uses a hybrid of Florence‑2 natural‑language style prompts and WD14‑style tags, letting you switch between slender Ikemen and high‑detail Bara builds without losing consistency.What style range does it cover?<ul><li>90s retro anime</li><li>2.5D / modern anime</li><li>Clean, high‑detail portraits and full‑body shots</li><li>Works with both SFW and tasteful suggestive setups</li></ul>What samplers and settings work best?<ul><li>Preferred sampler: Restart (cleanest lines and detail)</li><li>Alternatives: DPM++ 3M SDE Karras (cinematic), Euler a (lightweight)</li><li>Steps: 30–35 (up to 50 with Restart)</li><li>CFG: 5.0–7.0</li><li>Clip Skip: 2</li><li>See V2 Workflows below for resolution and upscale guidance.</li></ul>What’s the difference vs v1.0? v2 is tuned for cleaner anatomy, smoother style shifts, and easier upscaling via Ultimate SD Upscale + 4x_NMKD‑UltraYandere. v1.0 still works but behaves more like the older MelloMochi flow; v2 is the current “mainline” branch.<hr /><h2 id="supported-resolutions">Supported Resolutions</h2><pre><code>Square: 1024 × 1024 
Portrait (Tall): 896 × 1152 (3:4) 832 × 1216 (2:3) 768 × 1344 (9:16) Landscape (Wide): 1152 × 896 (4:3) 1216 × 832 (3:2) 1344 × 768 (16:9)</code></pre>Model trains best on 2:3 / 3:2 and 9:16 / 16:9 but other ratios can work with better prompting.<hr /><h2 id="v2-workflows">V2 Workflows</h2><h2 id="workflow-a-highvram-(recommended)">Workflow A – High‑VRAM (Recommended)</h2><pre><code>1. Base Generation (txt2img) 
Sampler: Restart 
Steps: 30–35 (up to 50 with Restart) 
Resolution: Recommended Resolutions Above 

2. High‑Res Upscale (Ultimate SD Upscale) 
Script: Ultimate SD Upscale 
Denoising: 0.25–0.45 
Scale: 2× 
Upscaler: 4x_NMKD‑UltraYandere_300k 
Type: Chess - Tile size: 1024×1024 

3. Final Inpainting (fix eyes, hands, etc.) 
Inpaint Area: Only Masked 
Resize to: 1024×1024 
Only Masked Padding: 32–64 
Denoising: 0.45–0.60 
Prompt: region‑focused, e.g., "close‑up of anime eyes", "clean masculine hands"</code></pre>This preserves fine details and keeps anime lines sharp through the 4K–ish pass.<h2 id="workflow-b-lowervram-(8-gb-friendly)">Workflow B – Lower‑VRAM (8 GB–friendly)</h2><pre><code>Base Generation (same resolutions, same samplers) 
Samplers: Euler a or DPM++ 3M SDE 

2. Inpaint First (at base resolution) 
Inpaint Area: Only Masked 
Resize to: 1024×1024 
Denoising: 0.40–0.60 
Fix anatomy, eyes, hands first 

3. Optimized Upscale 
Denoising: 0.25 
Script: Ultimate SD Upscale 
Scale: 1.5× or 2× 
Upscaler: 4x_NMKD‑UltraYandere_300k 
Type: Chess - Tile size: 512×512</code></pre>This minimizes crashes while keeping edits intact.<hr /><h2 id="comfyui-workflow">ComfyUI Workflow</h2>Alternate prompting method: Homochi XL also works well with a subject + style prompt split in SDXL workflows. Put the character, pose, clothing, and environment in one prompt, and keep the art style, lighting, and finish in a second prompt block for better control over composition and rendering.Recommended workflow: Use this premade ComfyUI workflow for batch selection and refinement: <a target="_blank" rel="ugc" href="https://civarchive.com/articles/14582/comfyui-workflow-sdxl-facedetailer-upscaler-and-batch-selection">https://civarchive.com/articles/14582/comfyui-workflow-sdxl-facedetailer-upscaler-and-batch-selection</a>Suggested settings:<ul><li>Sampler: Restart (res_multistep)</li><li>Scheduler: Karras</li></ul>Subject Prompt:<pre><code>1man, solo, handsome male mage with long white-blonde hair, violet eyes, slender build, floating in a meditative pose, holding a glowing crystal orb, layered robe, cosmic background, slightly above-eye-level close-up, SFW</code></pre>Style Prompt Prompt:<pre><code>masterpiece, best quality, very awa, ultra-detailed, absurdres, highly detailed, in the surreal watercolor-infused illustration style of Yoshitaka Amano</code></pre><hr /><h2 id="positive-prompt-template:">Positive Prompt Template:</h2>Tag Based Prompt:<pre><code>masterpiece, best quality, very awa, ultra-detailed, absurdres, highly detailed, SFW,
[ART STYLE], [Optional: ARTIST/SERIES],
[SUBJECTS], SOLO/DUO/GROUP, [character details],
[pose/action],
[clothing/style],
[environment/lighting]</code></pre>Example:<pre><code>masterpiece, best quality, very awa, ultra-detailed, absurdres, highly detailed, SFW,
Final Fantasy concept art, Yoshitaka Amano,
1boy, solo, handsome male mage, long white-blonde hair, violet eyes, slender build,
floating in a meditative pose, holding a glowing crystal orb,
layered robe with translucent folds and ornate patterns,
cosmic background, soft magical glow</code></pre>Hybrid Prompt:<pre><code>masterpiece, best quality, very awa, ultra-detailed, absurdres, highly detailed, SFW, [ART STYLE], [Optional: ARTIST/SERIES]
[SUBJECTS], SOLO/DUO/GROUP, [your subject/character, e.g., long hair, blue eyes],
[pose/action, e.g., standing, dynamic pose],
[clothing/style, e.g., detailed dress, intricate patterns],
[environment/lighting, e.g., cinematic lighting, soft glow]</code></pre>Example:<pre><code>masterpiece, best quality, very awa, ultra-detailed, absurdres, highly detailed, SFW, Final Fantasy concept art, Yoshitaka Amano
1boy, solo, handsome male mage, long white-blonde hair, violet eyes, slender build,
floating in a meditative pose, holding a glowing crystal orb,
layered robe with translucent folds and ornate patterns,
cosmic background, soft magical glow</code></pre><hr /><h2 id="negative-prompts-(suggested-v2-defaults)">Negative Prompts (Suggested V2 Defaults)</h2>Style‑Focused Negative (v2‑friendly):<pre><code>(image drawn by an amateur:1.3), simple details, incohesive lighting, poor art direction, (lack of unique point of view), lowres, bad anatomy, bad hands, missing fingers, extra digit, fewer digits, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, simple background, flat lighting, deformed, poorly drawn,</code></pre>Universal Negative (works across schedulers):<pre><code>lowres, bad anatomy, bad hands, missing fingers, extra digit, fewer digits, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, simple background, flat lighting, deformed, poorly drawn</code></pre>Homochi XL responds well to both Florence‑2 descriptive style and WD14 tagging, so you can mix or simplify these as needed.

Files with SHA256: 0f84ec5bb5bb6cd31340c7d2e79bcd57499d84fbe78c9f548d87c5e625805e93