Midlibrary: Library of 4000 Midjourney artistic styles
A catalog of 4,000 Midjourney “artistic styles” prompts debate over what it means to imitate or “steal” an artist’s style in the age of generative image models. Commenters weigh the creative empowerment and lower costs these tools offer—especially for non-artists and low-budget projects—against concerns about training on unlicensed work, devaluing professional illustrators, and flooding cultural spaces with derivative visuals. Technical points around Stable Diffusion vs. SDXL, fine-tuning costs, and LoRA models highlight that high-quality stylistic replication often demands significant compute and expertise, even as ethical and legal norms remain unsettled.
Site and project
- The site is intermittently down, likely due to traffic (“hug of death”), with some using Wayback snapshots instead.
- Front page serves many images; commenters wonder about bandwidth and hosting costs.
- Some confusion about the name; clarified as “midlibrary,” a fan-made style guide for Midjourney users, not an official product.
Model capabilities, training, and tools
- Comparisons between Stable Diffusion 1.5 and SDXL: SD 1.5 seen as fast and “good enough” for many use cases; SDXL can be better but is slower and harder to fine-tune.
- High-quality SDXL fine-tuning (e.g., for anime) is compute-heavy, reportedly on the order of hundreds of high-end GPU hours; community projects struggle with this.
- LoRA vs full fine-tuning: LoRA is viewed as good for adding specific characters/styles; extensive domain shifts need full fine-tuning.
- Several links to anime/furry models, style cheat sheets, and artist-tag lists for SD/SDXL.
Ethics, copyright, and “style theft”
- Some argue the site catalogs styles that were effectively “stolen” from artists; others say style isn’t copyrightable and the guide simply names styles for prompting.
- Disagreement over whether using artist names alongside AI output is misleading vs clearly “in the style of.”
- Concern over models generating Disney-like characters; question whether and when corporations will act.
- Broader clash: one side calls training on copyrighted images “stolen property”; the other likens it to humans learning from prior art and says the distinction is unclear.
Impact on artists and the future of work
- Many express sadness for artists whose styles can be cheaply mimicked, fearing loss of livelihoods and reduced incentive to share work.
- Others argue artists motivated by passion will continue creating; some liken AI to industrialization: contract/production artists get hit hardest, “high art” less so.
- Counterpoint: being “immortalized” in a model doesn’t pay bills; economic harm remains.
Practical uses and limitations
- Numerous examples where AI is a tool, not the goal: comics, card games, music videos, children’s stories, posters, web backgrounds, NPC avatars, personal stickers.
- Often used where no human artist would have been hired, or where budgets/time are very limited.
- Persistent issues: style/character consistency, bland or generic look, and “AI tells” in images.
- Some see AI art as good for references, moodboards, or replacing crude doodles; others argue it still lacks intention, deep control, and reliable copyright status.