One thing most people who have worked with AI drawing have probably experienced is this annoying part: The character in the first picture is perfectly good, but as soon as you change the expression or action, the face collapses into another character.
🤷♂️ Character Consistency has traditionally been a tough problem in the AI area. However, for today, I have a "ultra-stable" workflow. With the capabilities of Nano Banana, we don't require any Complex ControlNet for our image control. All we really require is 3 steps in structured prompts for your character to freely move between crying, laughing, angry, and scolding poses with 100% locked facial features.
Now, let us move on to the nitty-gritty. It is advised you save this first.
Just copy these prompt and sent to AI, remember give AI your origin picture:
Character Sticker Generation Workflow
Step 1: Visual Feature Analysis (Internal Only)
Carefully analyze the uploaded photo and extract the following visual features (do NOT output this analysis to the user; use it only for image generation):
- Gender & Age Impression: Male / Female, Child / Young Adult / Middle-aged
- Hairstyle (Most Critical Feature):
- Bangs style: middle-part, side-part, blunt bangs
- Hair length: short / medium / long
- Hair texture: straight / wavy / curly
- Hair color
- Facial Details:
- Glasses (round / square / black frame)
- Facial hair
- Distinctive moles or marks
- Eye shape and size
- Clothing Features:
- Top color
- Clothing type (hoodie / T-shirt / shirt / sweater)
- Patterns, stripes, or logos
Step 2: Prompt Construction (English Image Prompt)
Using the extracted features, construct a single detailed English prompt for image generation. The prompt must strictly follow the style and structure below.
1. Style Definition (Nanobanana Style Emulation)
- Art Style: Flat vector illustration, Chibi style, thick and clean black outlines, cute and expressive proportions.
- Coloring: Flat colors, minimal shading, bright, friendly, and playful palette.
- Layout: A sticker sheet grid containing 9–12 different poses and facial expressions of the same character.
- Background: White background or very light grid-paper texture.
- Consistency: Same character design across all stickers.
2. Required Expression & Action Set
The sticker sheet must include a randomized selection of 9–12 expressions from the list below:
- Positive & Friendly: Greeting (Waving), Approval (Thumbs up), Victory (Peace sign ✌️), Love (Heart eyes).
- Daily Life: Watching Drama (Popcorn), Lazy (Lying flat), Procrastinating, Eating food.
- Funny & Emotional: Speechless ("..."), Shocked (Jaw dropped), Crying (Cartoon tears), Cool (Sunglasses).
3. Prompt Output Requirements
- Output only one complete English prompt.
- No explanations, no analysis, no extra text.
- The prompt must be ready for direct use in an image generation model.
Step 3: Image Generation
Call the image generation tool using the constructed prompt and generate the final sticker sheet.
By now, you have learned the most complex form of "face control technique" for AI characters.
The technique not only applies to making emojis but can be used to create four-panel strips, IP peripherals, as well as served as your social media avatar.
The key of any technology should be serving creativity. Those who are unable to draw need not worry. With the help of the tool above, your imagination will simply be the only limit.
Top comments (0)