Batch vs Manual: Reddit user seeks optimal pipeline for large character gen
Running overnight batches yields variety but bad hands and weird anatomy...
A Reddit user on a generative AI subreddit is grappling with a common scaling dilemma in their workflow. They use wildcards and dynamic prompts to cycle through a large library of character styles and fandom concepts, generating batches of images. When they let massive scripts run overnight, they get high variety but poor consistency—bad hands, weird anatomy, and low-quality outputs dominate, forcing hours of manual cherry-picking to salvage the good ones.
On the other hand, if they micromanage every prompt and seed, the process becomes painfully slow and defeats the purpose of scaling. The user asks the community for advice on optimizing their pipeline: rely heavily on automated scoring/filtering tools (like CLIP-based aesthetic scoring or face detection) or accept the manual curation grind as a necessary evil. They're looking for sanity-preserving strategies to manage high-volume output without sacrificing quality.
- Wildcards and dynamic prompts generate high variety but often produce bad hands and weird anatomy.
- Overnight batch runs save time upfront but create hours of manual curation work.
- Automated scoring/filtering tools (e.g., aesthetic models, face detectors) may reduce the curation burden.
Why It Matters
As AI image generation scales for creators, efficient curation pipelines become critical for productivity and sanity.