Stable Diffusion 3 churns out anatomically incorrect images
Stable Diffusion 3 ( SD3) has been criticised for its poor performance in rendering human figures, particularly limbs like hands and feet, resulting in ridicule and concerns about user safety. Compared to other AI image models like Midjourney and DALL-E 3 , SD3 has been found to generate anatomically incorrect visual aberrations, leading to what some describe as "AI-generated body horror". Users on Reddit ridiculed SD3's attempts at depicting humans, with threads highlighting issues in rendering entire human bodies - problems that are particularly noticeable when trying to generate images of people in certain poses or situations, like "girls lying on the grass". Some in the AI community attributed these anatomical failures to Stability AI's decision to filter out adult content (NSFW) from the SD3 training data. There are concerns that the NSFW filter used during pre-training may have been too strict, inadvertently removing non-offensive images and depriving the model of important human depictions. The controversy highlights the ongoing challenges in developing AI image generation models that can accurately depict human anatomy while adhering to content guidelines. System 🤖 Stable Diffusion Operator: Developer: Stability AI Country: USA Sector: Media/entertainment/sports/arts Purpose: Generat e images Technology: Text-to-image; Machine learning Issue: Accuracy/reliability; Robustness; Safety
- Date it happened
- 2024-06-01
- Product, system or model
- Stable Diffusion 3
This incident was imported from AIAAIC and is used under CC BY-SA 4.0. Our additions to it — the structured fields, the translation, the checks against other reports — are published under the same licence.
This is a record of what was reported, not a finding that anyone broke the law. If it names your organisation and you believe it is wrong, the corrections process is free and open to everyone.