T-DIG diversity policies zero-shot transfer to GPT Image 2 and Gemini
simi_97k · x · 2026-10-28
Thread wrap-up of T-DIG: policies trained on open-source image models boost diversity on proprietary models via zero-shot transfer — TPIPS gains of +0.27 on GPT Image 2 and +0.20 on Gemini 3.1 Flash Image (Qwen-Image policy), beating prompting GPT-5.5 for diverse prompts (+0.19). Limitations: results bounded by the base image model (some camera-angle requests fail), the strength slider scales LoRA weights unevenly, and optimizing TPIPS may bias the policy.
Related event: T-DIG Policy Transfers Zero-Shot to Boost GPT Image 2 and Gemini Diversity(2 posts)→
More from Multimodal
- Zero-AI-at-runtime design generator hits run 8, new aesthetics expected by run 10 — Afinetheorem · 2026-10-28
- Painting with EmbeddingGemma 2: ~30 images/sec on Apple Neural Engine after 12h A100 training — GlennCameronjr · 2026-10-28
- Krea2 BBOX models released: 40k-image fullrank finetune with multi-character LoRA support — Amazing_Painter_7692 · 2026-10-28
- One-line Tekken-style prompt in HeyGen Video yields ad-grade 2K fight footage — aziz4ai · 2026-10-28
- Stunning AI-generated concept art casts goddess Sekhmet walking a post-apocalyptic wasteland — DionysianAgent · 2026-10-28
- Custom node pack brings looping H3 video generation to Comfy Cloud — PurzBeats · 2026-10-28