Dev builds a character-swap LoRA dataset end-to-end with Codex and GPT Image 2.5
ostrisai · x · 2026-09-23
Developer akatzai built the initial dataset for a character-swap Minimax H3 LoRA by having Codex drive GPT Image 2.5 directly through its imagegen tool — generating scenes, creating character references, and editing the character into each scene without manual artwork.
Next step: training the LoRA on a 4090 rig using ostris's toolkit, with plans to open-source both the LoRA and the dataset on Hugging Face if it works.
More from coding & agent
- Peter Yang's Opus 5.5 hands-on: recreating Disney's Soaring flight and calling Claude's comeback — petergyang · 2026-09-23
- Dev: if your eval costs $20 for 100 runs, the problem set isn't hard enough — pvncher · 2026-09-23
- Porting a dual-screen device to Android 17 with Claude: 110 flashes, 3 burned resets, and eink reverse engineering — PaddleStroke · 2026-09-23
- He used AI to build his 2019 game idea — it worked, but the idea wasn't as good as he thought — BLUECOW009 · 2026-09-23
- Open-source Agent Message Board gives parallel coding agents a shared SQLite forum — airesearch12 · 2026-09-23
- A good eval has a shelf life of maybe 3 months, practitioner argues — pvncher · 2026-09-23