Get Consistent Character and Styles with Dalle-3
Entertainment
Get Consistent Character and Styles with Dalle-3
Hey everyone, welcome back to another show! Recently, I did a live session where I explored using variables to get consistent characters in DALL-E 3. I revisited this method and made some revisions based on a new approach. I stumbled upon a video by Gilberry, who has an excellent YouTube channel, and I'll link to his video for reference. He discussed using custom instructions for DALL-E 3, which inspired me to incorporate a similar strategy but tailored for comic art.
Using Custom Instructions
Custom instructions enable you to set up background parameters and output descriptions, essentially guiding ChatGPT like a set of rules. This method was quite different from my usual approach, but I found it highly effective for comic style illustrations.
Initially, I adapted Gilberry’s instructions, tweaking them to fit a comic art style rather than focusing on images or photography. After making necessary edits, I ended up with custom instructions that aligned well with my needs.
Describing Characters
One critical realization was that overly descriptive character definitions could limit versatility. Instead, providing broader descriptions allowed for better flexibility in depicting characters in various positions.
For example, when describing the Sentinel’s suit, I was more specific to ensure a particular look. However, I was less detailed with Blake to maintain versatility.
Crafting the Story
I created a straightforward story prompt: Blake starts his day in a coffee shop until an army appears, and he transforms into the Sentinel to save the day. The goal was to keep it concise to avoid going off tangent. Including "write in the style of an action comic book" in the prompt resulted in a well-structured panel narrative, which was a pleasant surprise.
Generating Images
Using DALL-E 3, I generated images in batches of four panels. Initially, I faced some challenges where DALL-E 3 didn’t render certain elements due to content restrictions. By simplifying the language to avoid triggering these restrictions, I managed to get consistent results.
Fine-Tuning and Completion
Despite some minor inconsistencies (e.g., character outfits changing between scenes), the resulting images were quite impressive. The next step involved converting the panel script into a narration using ChatGPT, which could then be voiced using 11 Labs.
In summary, using custom instructions in DALL-E 3 to maintain a "North Star" for character consistency and art style yields effective results. While it’s not perfect, it achieves about 80% of what I aimed for, and minor edits can be made using tools like Photoshop or in-painting software. I'll be posting the final product soon with voiceover.
Thanks for tuning in, and keep being creative!
Keywords
- DALL-E 3
- Custom Instructions
- Comic Art Style
- Character Consistency
- ChatGPT
- Panel Narrative
- Illustration Fine-Tuning
FAQ
Q: What are custom instructions in DALL-E 3? A: Custom instructions allow you to set up background parameters and output descriptions, providing rules to guide the AI's output.
Q: Why should you be less descriptive when defining characters? A: Being less descriptive offers more versatility in depicting characters in various positions and scenarios.
Q: What inspired the revised approach in using custom instructions? A: I was inspired by a video from Gilberry’s YouTube channel which discussed using custom instructions for DALL-E 3.
Q: How did you handle content restrictions in DALL-E 3? A: Simplifying language and avoiding certain trigger words helped bypass content restrictions.
Q: Are there any tools recommended for fine-tuning images? A: Yes, tools like Photoshop and in-painting software are useful for minor edits and fine-tuning the images.
Q: What's the next step after generating comic panels? A: Convert the panel script into a narration and use a tool like 11 Labs for voiceover work.