NeoTokyo — Vintage Cel-Shaded Anime Series (Illustrious-XL & Anima)
NeoTokyo is a series of custom checkpoints built on top of Illustrious-XL (IL-XL) and Anima, merging base anime foundations with fine-tuned retro cel-shading LoRAs directly into the UNet.
The project brings back the authentic, hand-painted aesthetic of golden-era Japanese animation without relying on post-processing hacks or bulky prompts. It is split into two specialized editions to fit different vintage aesthetics:
📦 Which Edition Should You Use?
📼 90s Edition (Classic):
Aesthetic: Gritty late 80s / 90s OVA, dark cyber-noir, moody urban alleys, and rain-soaked nights (Cowboy Bebop, Cyber City Oedo 808, Wicked City).
Style: Heavier inking, high contrast, dramatic shadows, bold cross-hatching, and rough traditional line art.
📺 80s Edition (City Pop):
Aesthetic: Golden 80s TV anime, retro sci-fi, warm summer countryside, and City Pop daylight (Kimagure Orange Road, Dirty Pair, City Hunter, You're Under Arrest).
Style: Clean two-tone cel shading, bright saturated palettes, softer line work, large expressive 80s eyes, and balanced anatomical rendering.
✨ More editions coming soon!
Key Strengths & Features
Multi-Engine Compatibility: Available in both Illustrious-XL (full Danbooru & LoRA ecosystem) and Anima (superior lighting and spatial coherence).
100% Passive Cel-Shading: Direct prompt rendering with bold ink outlines, cel shading, and subtle analog film grain straight out of the box.
Pure 2D Finish: Eliminates modern plastic/3D gloss in favor of classic animation cel textures.
Rock-Solid Anatomy: Clean facial features and reliable hand/finger rendering on complex poses.
Recommended Settings For Illustrious-XL Editions:
Sampling Method:
DPM++ 2M| Schedule:KarrasSteps: 30 | CFG Scale: 4.5 – 5.5 | Clip Skip: 2
Resolution: 896 × 1152
Hires. fix:
R-ESRGAN 4x+ Anime6B(1.5x upscale, 15–20 steps, Denoising0.35 – 0.40)
For Anima Editions:
Sampling Method:
euler| Schedule:simple(ornormal)Steps: 25 – 30 | CFG Scale: 3.5
Resolution: 896 × 1152 (or 1024 × 1024)
Anima Editions (ComfyUI Diffusion Model Only)
⚠️ Installation: This file contains the Diffusion Model (UNET) weights only (no baked CLIP or VAE). Load it via the Load Diffusion Model node alongside the base Anima files (available on HuggingFace):
This model ➔
ComfyUI/models/diffusion_models/qwen_3_06b_base.safetensors ➔
ComfyUI/models/text_encoders/qwen_image_vae.safetensors ➔
ComfyUI/models/vae/(Qwen-Image VAE)
Recommended Prompt Boosters (Optional):
📼 90s Edition:
classic cel shading, 1990s anime visual, retro anime, vintage film grain
📺 80s Edition:
classic cel shading, 1980s anime visual, city pop, vintage cel
Negative Prompt:
worst quality, low quality, 3d, render, CGI, digital illustration, photo, deformed, blurry, bad anatomy, bad hands, extra fingers, missing fingers, glowing skin, plastic skin, smooth gradients
Description
📺 90s Edition - (Anima Architecture)
This is a DiT diffusion model merged directly with anima-aesthetic-v1.1. It captures the authentic 1990s retro anime aesthetic (traditional cel shading, dark film grain, hand-drawn ink lines, and vintage cinematic lighting) without needing an external LoRA.
⚠️ ComfyUI Installation Note: This file contains the Diffusion Model (UNET) weights only (no baked CLIP or VAE). Load it using the Load Diffusion Modelnode (not Load Checkpoint) alongside the separate components below.
Recommended Settings:
Architecture: DiT (Anima 2B)
Sampler:
euler(Avoid SDE samplers to prevent oversaturation)Scheduler:
simple/normalCFG Scale:
3.0–4.0(Do not exceed 4.5)Steps: 25 – 30
Resolution: 1024x1024 (or standard portrait/landscape ratios)
Required Components (ComfyUI):
Diffusion Model:
ComfyUI/models/diffusion_models/(This file)Text Encoder:
qwen_3_06b_base.safetensors➔ComfyUI/models/text_encoders/(Type:lumina2/qwen)VAE:
qwen_image_vae.safetensors➔ComfyUI/models/vae/
(💡 More details, installation instructions, and download links can be found in the main model description below.)



