Here is a list of tag sets for specific models.
Notes for each file:
anima-2.9B-preview-V1.csv - (unofficial expanded anima finetune) - uses Anima recommendations with 2026-08-12 cutoff, the correct cutoff (~2026-07-01) is not possible without historical backups (from what I can tell), since gel doesn't expose creation date.
Anima-1.0.csv - uses the Anima recommendations, @ before artist names, underscore formatting cutoff 2025-09-01, gelbooru based.
Anima-preview.csv - uses the Anima recommendations, @ before artist names, underscore formatting cutoff 2025-09-01
Anima-preview-alternate.csv - same as regular list but uses gelbooru as a base. Gel artists definitely work, so this seems like it's more correct than the regular list, but information about the dataset is still a bit cryptic.
NoobAIXL1.1_dash.csv - both danbooru and e621 with a cutoff of 2024-11-03 (dash format)
NoobAIXL1.1_underscore.csv - both danbooru and e621 with a cutoff of 2024-11-03 (underscore format)
ChenkinNoob-XL-V0.1_dash.csv - both danbooru and e621 with a cutoff of 2025-09-01 (dash format)
ChenkinNoob-XL-V0.1_underscore.csv - both danbooru and e621 with a cutoff of 2025-09-01 (underscore format)
ChenkinNoob-XL-V0.2_dash.csv - both danbooru and e621 with a cutoff of 2025-11-01 (dash format)
ChenkinNoob-XL-V0.2_underscore.csv - both danbooru and e621 with a cutoff of 2025-11-01 (underscore format)
ChenkinNoob-XL-V0.3_dash.csv - both danbooru and e621 with a cutoff of 2026-01-01 (dash format)
ChenkinNoob-XL-V0.3_underscore.csv - both danbooru and e621 with a cutoff of 2026-01-01 (underscore format)
IllustriousV0.1_dash.csv - danbooru only with a cutoff of 2024-04-30 (dash format)
IllustriousV0.1_underscore.csv - danbooru only with a cutoff of 2024-04-30 (underscore format)
IllustriousV1.0_dash.csv - danbooru only with a cutoff of 2024-06-30 (dash format)
IllustriousV1.0_underscore.csv - danbooru only with a cutoff of 2024-06-30 (underscore format)
Cutoff date and aliases are not used for the e621 parts because the e621 api is cancer and would require 10+ hours per list just to get all of the aliases.