GPT Image Stealth Models Slammed for Low-Taste Output; Users Want Finetuning
flowersslop · x · 2026-10-23
Leaks suggest OpenAI is stealth-testing new GPT Image models under codenames "shooting-star-green-alpha" and "shooting-star-red-alpha" (unverified); early tester feedback says accuracy is "not so accurate."
The quoting poster goes harder: if v3 still can't beat the year-old nano-banana pro and keeps churning high-CTR engagement slop, OpenAI should conclude it can't do tasteful imagegen, give up, and focus on ASI. Practical ask: even basic LoRA finetuning support would fix most of the taste problems.
More from Models
- Claim: Anthropic kills the REA, Claude now refuses to do it — kingslyj · 2026-10-23
- Google reportedly building a Decision model with new createDecision method in Gemini API — lyraxana · 2026-10-23
- DeepSeek V4.1 Flash test: official API hits 99% cache vs 92-96% on OpenCode Go — teortaxesTex · 2026-10-23
- 1017 Reportedly Destroys Opus 5.5, Astra and Sol at Building a Complex Machine — teortaxesTex · 2026-10-23
- Professor: Keep Your Unique Voice — Experts Spot LLM Writing Instantly — plopesresearch · 2026-10-23
- Kimi K3.1 rumored to drop imminently, screenshot shows it live in Kimi Code — 141_1337 · 2026-10-23