Same face prompt across AI image models: KREA beats paid rivals

BBCL2026_546 · reddit · 2026-09-18

A Redditor ran the same detailed portrait prompt (an Indian woman with a red veil) across multiple text-to-image models and posted a comparison grid:

• KREA was the most consistently appealing, beating paid models in the author's view; Grok and ChatGPT keep applying a default "cute face" and ignore half the details
• SeeDream and Gemini were the surprises, adding details on their own
• Moderation differences: Grok is now locked down hard; SeeDream allows topless generations but genitalia often come out deformed
• The author misses Grok Imagine 1.5's wild variety versus today's homogenized output

Full prompt included for reproduction.

Original post →

More from Multimodal

• Qwen Image 2.1 support coming soon to ComfyUI — Time-Teaching1926 · 2026-09-20
• Krea2 Turbo cinematic recipe: 12 steps, dual LoRAs and film-grain prompting — cloutcobain1996 · 2026-09-20
• Creator makes full AI video with Seedance 2.5 using plain prompts, only two generations — LudovicCreator · 2026-09-19
• Long-video consistency trick: saving 5MB latents with MiniMax H3 — reeight · 2026-09-19
• Fully open-source Diffusion Studio generated a 450K-view launch video without touching the timeline — _AustinCalvert_ · 2026-09-19
• Over half of AI videos will be rendered from code, predicts Diffusion Studio demo — _AustinCalvert_ · 2026-09-19

Multimodal channel →