Testing 93 art styles on FLUX.2 with one seed: style names alone do little; technique descriptions placed first win
Glad_Captain8190 · reddit · 2026-10-11
The author rendered 93 art styles with FLUX.2 [klein] 4B (GGUF in ComfyUI, 4 steps, guidance 1) on the same six scenes and one seed — 564 images on one comparison page, with a no-style control row on top.
Key findings:
- About half the style names did nothing on their own — phrases like "a pop art picture of" or "ukiyo-e" barely changed output; the 70-word scene description won every time
- Describing the technique (materials, palette, texture, what to exclude) and placing it before the scene fixed pop art, ukiyo-e, cyberpunk and low poly
- Material styles (paper cut, crochet, needle felt, stained glass, wood carving) worked first try
- Pixel art and double exposure never got there; artist names carry far less weight than in SDXL
Every style has its copyable prompt sentence; more can be requested.
More from Multimodal
- Meshy offers Hermes users a month of Pro with API access for $1 — Teknium · 2026-10-23
- Customuse launches Smart Mesh AI with geometry-aware adaptive character clothing tool — SimplyAnnisa · 2026-10-23
- AI-generated movie poster 'The Chase' crams a wolf, a plane and muscle cars into one chase — Aggravating-Sir-115 · 2026-10-23
- SteerTTS-edge: an 11.84MB steerable TTS with unchanged WER running 19x realtime on CPU — Working_Hat5120 · 2026-10-23
- ElevenLabs' Chimera pairs Eleven v4 voice with Seedance video to make character images speak — AIwithGhotai · 2026-10-23
- Open-Source Tennis Models Plus Claude Turn Video Into a World-Class Coach — tweetsatpreet · 2026-10-23