ComfyUI anime renders look flat: indie dev shares full setup and asks what to try next
Conscious-Season-607 · reddit · 2026-11-20
An indie game artist rendering anime characters locally in ComfyUI on an RTX 3060 Laptop (6GB) asks how to reach the depth in cloth, hair and lighting seen in reference images.
Setup: Nova Anime XL IL v19.0 (plus SteinCustom V15, Hassaku XL v3.4), Fine Anime Screencap XL LoRA at 0.7, clip skip 2, Euler a, 30 steps, CFG 4.0, 1024x1344 with a RealESRGAN hires pass (denoise 0.2), and IP-Adapter plus-sdxl-vit-h for character consistency.
Tried and failed: img2img at 0.3 keeps design but stays flat; 0.6 adds depth but shifts details; 0.7+ on large canvas duplicates faces and stretches figures; "2.5D" prompting barely helps; extra LoRAs (Detail Tweaker, People's Works, lighting slider) do almost nothing. He asks about tile ControlNet, smaller canvas, or prompt/lighting changes.
More from Multimodal
- Open-source repo offers 500+ AI-generated low-poly 3D landmarks for maps under CC0 — tomjohndesign · 2026-11-20
- CCORE releases 11,000+ Three.js scenes and GLB assets, training locally runnable 4B 3D models — ConfidentDinner6648 · 2026-11-20
- One video-gen prompt blends dynamic camera moves, slow motion and action in a single shot — umesh_ai · 2026-11-20
- A Civitai model the author quietly deleted leaves a fan wondering whether to pick up the torch — Ximena-WD · 2026-11-20
- Regenerate-IC beats SeedVR2 on quality at 150% of the runtime, users find it worth it — Anitabcfan · 2026-11-20
- A short film made entirely by code with Fable 5.2 — no Blender, only Edge TTS for voiceover — OfirPress · 2026-11-20