Bienvenue
😃 Bases
💼 Applications de base
🧙‍♂️ Intermédiaire
🤖 Agents
⚖️ Fiabilité
🖼️ Prompting d'images
🔓 Hacking de prompts
🔨 Outillage
💪 Calibrage de prompts
🎲 Divers
📙 Référence de vocabulaire
📚 Bibliographie
📦 Prompted Products
🛸 Ressources supplémentaires
🔥 Sujets Brûlants
✨ Générique
🖼️ Prompting d'images🟢 Fix Deformed Generations

Fix Deformed Generations

🟢 This article is rated easy
Reading Time: 1 minute
Last updated on August 7, 2024

Sander Schulhoff

Deformed generations, particularly on human body parts (e.g. hands, feet), are a common issue with many models. This can be dealt with to some extent with good negative prompts. The following example is adapted from this Reddit post.

Example

Using Stable Diffusion v1.5 and the following prompt, we generate a nice image of Brad Pitt, except for his hands of course!

Astronaut

Prompt


studio medium portrait of Brad Pitt waving his hands, detailed, film, studio lighting, 90mm lens, by Martin Schoeller:6

Using a robust negative prompt, we can generate much more convincing hands.

Astronaut

Prompt


studio medium portrait of Brad Pitt waving his hands, detailed, film, studio lighting, 90mm lens, by Martin Schoeller:6 | disfigured, deformed hands, blurry, grainy, broken, cross-eyed, undead, photoshopped, overexposed, underexposed, lowres, bad anatomy, bad hands, extra digits, fewer digits, bad digit, bad ears, bad eyes, bad face, cropped: -5

Using a similar negative prompt can help with other body parts as well. Unfortunately, this technique is not consistent, so you may need to attempt multiple generations before getting a good result. In the future, this type of prompting should be unnecessary since models will improve. However, currently it is a very useful technique.

Notes

Improved models such as Protogen are often better with hands, feet, etc.

Sander Schulhoff

Sander Schulhoff is the Founder of Learn Prompting and an ML Researcher at the University of Maryland. He created the first open-source Prompt Engineering guide, reaching 3M+ people and teaching them to use tools like ChatGPT. Sander also led a team behind Prompt Report, the most comprehensive study of prompting ever done, co-authored with researchers from the University of Maryland, OpenAI, Microsoft, Google, Princeton, Stanford, and other leading institutions. This 76-page survey analyzed 1,500+ academic papers and covered 200+ prompting techniques.

Footnotes

  1. Blake. (2022). With the right prompt, Stable Diffusion 2.0 can do hands. https://www.reddit.com/r/StableDiffusion/comments/z7salo/with_the_right_prompt_stable_diffusion_20_can_do/