MageTrail 2.8B booru finetune costs $593 so far, hits limits of 41k-image dataset
Turbulent-Bass-649 · reddit · 2026-09-30
The author updates MageTrail, a Danbooru/E621 proof-of-concept full finetune of Microsoft's 4B MageFlow T2I model at 2.8B parameters. Versions V0.1-V0.3 cost $593 total (425 H100 hours) on a condensed 41k-image dataset to instill booru-tag prompting and illustration skills.
Wins: the model learned many concepts; code changes now allow training directly with Flux2VAE (best open-source VAE) without costly realignment; Muinez's independent $200-300 test suggests MageFlow could become a strong 1B-4B base for illustration if trained on 10M+ images.
Lessons: 41k images are far too few — a generalizable, stable base needs 1M+ images, and the booru-essence approach is no shortcut. Out of money, the author is pausing to curate a 200k+ artist-style dataset for the next phase.
More from Multimodal
- VideoLoop rewrites bounded working memory, hits 88.3% on VideoMME long video — Jinfa Huang · 2026-09-30
- I used Codex to make a 15-second animation — and still had to give it editing notes — naridubs · 2026-09-30
- Flux 3 nails four-way split-screen images of one event from four angles — umesh_ai · 2026-09-30
- Filmmakers jam with AI video generation wait times to shoot a music duet with Luma — mrjonfinger · 2026-09-30
- Meta's LSRM Wins ECCV 2026 Honorable Mention, Beats 3D Reconstruction SOTA by 2.4 dB — rsasaki0109 · 2026-09-30
- Adobe Research Shows Adversarial Post-Training Restores Missing High-Frequency Detail in Pixel Diffusion — adobe-research · 2026-09-30