These are images upscaled to 8K! Workflow with Hi res — Anima 1.0 Aesthetic + Base V2The V2 is here. Base 1.0 and Aesthetic 1.0 now live in the same workflow, and the eyes got their own spa treatment.### What's new in V2- Dual model: Base 1.0 + Aesthetic 1.0 merged in one workflow <code>ModelMergeSimple</code>, 65% Base / 35% Aesthetic by default) — the raw, faithful look of Base, with a dash of Aesthetic logic. One slider to rule them all: <code>0</code> = 100% Base, <code>1</code> = 100% Aesthetic.- Eyes upgrade (the real star of V2):- Dual paired detection <code>Eyeful_v2-Paired</code>) → grabs both eyes in one shot instead of hunting them one by one- SAM segmentation + Impact detailer on top- DifferentialDiffusion inside the detailer for cleaner inpainting- Auto <code>[CONCAT] detailed eyes</code> — the detailer prompts itself, with minimal style-safe tags so it refines your eyes instead of repainting your style- Guide 768 · max 1024 · denoise 0.35, plus a Tiled Detailer option for big renders- Anti-grain recipe: CFG 3.5 + <code>simple</code> scheduler + the short official prefix <code>masterpiece, best quality, score_7,</code>). No HDR soup, no ultra-detailed tag fight.- Cleaner layout: numbered groups <code>01 MODELES</code> → <code>08 POST &amp; SAVE</code>, plus a Fast Groups Bypasser to toggle options in one click.### Pipeline1. GEN — 35 steps · CFG 3.5 · <code>res_multistep</code> · <code>simple</code> · Base×Aesthetic merge2. 2ND SAMPLER — refine pass, 16 steps, denoise 0.2 (OFF by default)3. HIRES PRE — Remacri 50% (= 2x)4. EYES DETAILER — the good stuff (see above)5. HIRES POST — ExtraSmoother 50% (OFF by default, for strong PCs)6. SAVE### Recommended settings- GEN: 35 steps · CFG 3.5 · <code>res_multistep</code> · <code>simple</code>- MIX ratio: 0.35 (65% Base / 35% Aesthetic) — too smooth? lower it. Too raw? raise it. It's a slider, not a marriage.- Eyes: guide 768 · max 1024 · denoise 0.35 · steps 14 · bbox crop 2- HiRes: 2x · LoRA: optional slot (style LoRAs trained on Base 1.0 work best)### Required models- Anima Base 1.0 + Aesthetic 1.0 (CircleStone Labs)- Qwen 3 0.6B text encoder + Qwen image VAE- <code>Eyeful_v2-Paired.pt</code> · <code>sam_vit_b_01ec64.pth</code>- <code>4x_foolhardy_Remacri</code> (+ <code>Remacri_ExtraSmoother</code> for the optional post pass)### Required custom nodesComfyUI-Easy-Use · rgthree-comfy · ComfyUI-Impact-Pack · ComfyUI-Impact-Subpack · JPS Nodes · RES4LYF (for <code>res_multistep</code> — new in V2, don't skip it or you'll get missing nodes)### ✨ Bonus — the neon umbre99 UI extensionThe glowing Umbreon logo living in the bottom-left corner of my ComfyUI screenshots? That's a tiny custom UI extension I made — and you can have it too. Grab _<code>neon-umbre99.zip</code>__ (attached to this page), extract it into <code>ComfyUI/custom_nodes/</code>, restart ComfyUI, refresh with Ctrl+F5. The light runs along the logo's contours, it glows faster while generating, flashes gold on success, turns red on error, and you can drag it anywhere (double-click = reset). It changes absolutely nothing about generation results — it just makes ComfyUI feel like home.### Tips- Eyes still shy? Raise the detailer denoise to 0.4, or add <code>detailed eyes</code> twice in your prompt (it stacks, don't tell anyone).- Too smooth? Lower the MIX ratio. Too raw? Raise it.- If something breaks, check the model paths first. If everything is correct and it still breaks, blame ComfyUI!! (tradition)### Credits- Anima Base &amp; Aesthetic v1.0 by CircleStone Labs (built on NVIDIA Cosmos-Predict2-2B-Text2Image)- Face Detailer / Impact Pack by ltdrdata- Eyeful detector, upscalers &amp; all the model authors listed above(No Google Translate this time. Growth.)bye ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⢲⣦⣤⣀⡀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠹⣿⣿⣿⠷⠂⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠈⠿⠁⢀⣴⣶⣶⡀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠿⣿⣿⣿⣿⡄⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠈⠛⢿⣿⣿⡀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⣠⣴⣶⣾⡿⠓⠶⣤⡀⠀⠀⠈⣷⣦⣄⡀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⣀⣠⠄⠀⣴⣶⣶⣶⣤⣤⣼⣿⣿⣿⣿⣿⡌⠻⣦⢻⡀⠀⠀⠸⣿⣿⣿⣿⣶⣦⠀⠀⠀⠀⢤⣴⣾⣿⣿⣿⠀⠀⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣷⣶⣾⣿⣄⠀⠀⢻⣿⣿⣿⣿⣿⠀⠀⠰⣄⠀⠀⠉⠛⠛⠿⠧⠀⠘⠛⠛⠋⠉⠈⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⡿⠀⠀⠈⢿⣿⣿⡿⠏⠀⠀⠀⣿⣿⣆⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⣿⣿⣿⣿⣿⣿⣿⣿⣿⡿⠋⠀⠀⠀⠀⠀⠉⠉⠀⠀⠀⢀⣼⣿⣿⣿⣧⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⣿⣿⣿⣿⣿⣿⣿⠋⠁⠀⠀⠀⠀⠀⠀⠀⠀⠀⠲⢶⣶⣿⣿⣿⣿⣿⣿⣧⣴⣶⣶⣦⣤⣄⣀⣀⣠⣾⣿⣿⣿⣿⣿⣿⣿⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠈⠙⠛⠿⠿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⣼⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⡇⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⡇⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⣸⣿⣿⣿⣿⣿⣿⠋⡀⢹⣿⣿⣿⣿⣿⣿⣿⣿⣿⠋⠙⣿⠇⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⣠⣿⣿⣿⣿⣿⣿⠃⣼⡇⣸⠏⠀⠀⢻⣿⣿⣿⣿⡏⣸⡆⡟⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⢰⣿⣿⣿⣿⣿⣿⣿⠀⢋⠔⠁⠀⠀⠀⢸⣿⣿⡿⣿⡇⠛⢠⠁⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠘⣿⣿⠏⢰⣿⣿⡿⠛⠁⠀⠀⠀⠀⠀⢸⣿⣿⡇⢸⣿⣶⡇⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⣿⣿⡇⣼⣿⣿⠃⠀⠀⠀⠀⠀⠀⠀⠀⠻⣿⣷⢸⣿⣿⡇⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⣿⣿⣧⣿⣿⣿⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠙⣿⣿⣿⣿⡇⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠈⠛⠻⠛⢿⣿⣧⣀⡀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠉⢹⣿⣿⣄⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠛⠿⠿⠟⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠈⠉⠉⠀⠀⠀⠀⠀

Workflow with Hi res — Anima 1.0 Aesthetic + Base V2 🌙

Even more precise but more demanding—tested on an RTX 5060, and the quality is top-notchI suppose the rendering is just as good on my latest Loras (mainly tested on my Pokemon-style )

Files with SHA256: 5b290e13e358d889dca3395e4a17ba63dced1588ae77b9df6c9827392610a1a8