Recreating the GPT semi-realistic 3D anime look locally with LoRAs, workflow and prompts
AI-Make-NSFW-Stuff · reddit · 2026-10-05
A Redditor shared a full local setup for reproducing the GPT image style where the character looks anime but the environment reads as a real photo:
- LoRAs: GPT-Image-2 Anime Illustration Style LoRA and a Japanese Semi-Realistic Anime (Krea 2 style) LoRA, both on Civitai
- Workflow: a complete ComfyUI workflow posted to Pastebin, ready to import
- Prompt template: multiple fully-written prompts (tennis, beach volleyball, interior portrait) structured as photorealistic scene + SUBJECT/POSE/OUTFIT/SETTING/LIGHTING/STYLE blocks — the key trick is keeping the environment fully photographic while only the character carries anime features
With the LoRA + workflow + prompt trio you can replicate the style locally today.
More from Multimodal
- Cute "Claude hits the beat" animation made with Fable 5.5 — ring_hyacinth · 2026-10-05
- Realistic consistent AI characters on $10-20/month: Flux dev + LoRA on RunPod — SomeCartographer4601 · 2026-10-05
- AI agents given one prompt and $1 each to produce 30s videos: Opus 5.5 vs GPT-6 Astra — no3us · 2026-10-05
- Redditor shares video made with Google Flow on free credits — clooneyge · 2026-10-05
- Solo dev open-sources MIR MEDIA LABS, a local AI media studio for NVIDIA GPUs — Da_QbanBeast · 2026-10-05
- WIP: using Claude to reconstruct structured Blender models from photos, beyond 3DGS — jwt0625 · 2026-10-05