Study Analyzes 6 Million Pixiv AI Images
Formal_Drop526 · reddit · 2026-07-14
A new study titled "Navigating the Open-Source Model Ecosystem" analyzes 6 million AI-tagged images on Pixiv, revealing the true workflow habits of AI art creators.
- Highly Concentrated Model Usage: Despite over 22,000 base models on the platform, just 560 models account for 80% of image generations.
- LoRA as Standard: By 2025, 75% of images use at least one LoRA, and works using LoRAs generally receive more views and bookmarks. However, stacking 3-5 or more LoRAs yields diminishing returns and can even hurt reach.
- Shifting Preferences: As new base models natively support character prompting, character LoRA usage is declining, with users shifting focus to style and concept LoRAs.
- Resistance to Version Upgrades: Creators are highly hesitant to update model versions; 20 weeks after release, only 40% of images use the latest version, primarily because new versions often break existing LoRA compatibility or alter aesthetic styles.
The research team has open-sourced the dataset, including prompts and metadata, on HuggingFace.
Related event: Study Analyzes 6M Pixiv AI Images to Reveal Ecosystem(2 posts)→
More from Multimodal
- TischEins, a large Midjourney SREF library, relaunches with new design after security outage — tisch_eins · 2026-10-26
- MOSS-Image-SFT 4-step distillation tops 16.5K downloads on Hugging Face — _akhaliq · 2026-10-26
- Claude Code can now edit videos: a 10-prompt workflow from raw footage to final cut — TawohAwa · 2026-10-26
- One-sentence 3D worlds turn cinematic with CineMap and TapNow — gizakdag · 2026-10-26
- No Camera Needed: Creator Builds an Entire World in CineMap and Directs It Into Cinema — gizakdag · 2026-10-26
- Kling vs. Seedance: the Chinese AI video model rivalry — DavidmComfort · 2026-10-26