Driving MidJourney Prompts with Silhouettes
I was struggling to set up a scene where a woman and a man were engaged in a heated conversation. Then, I decided to use a silhouette of two people arguing, and it turned out to be exactly what I needed.
This technique is highly useful, though not foolproof. While it may not work 100% of the time, it will undoubtedly reduce the number of images required for a given prompt.
Essentially, you need to gather silhouettes of the poses or actions you want the character to execute. Just write the prompt and select a silhouette image as prompt mode and hit enter.
POC
I’ve been using silhouettes since then and it’s working for me like a charm. Let’s dive into some examples.
Prompt:
cinematic shot, a woman and a man having a conversation in the middle of a hallway building, neon lights
Before Silhouette
The output wasn’t what I wanted, and I was getting really frustrated. Then, I decided to try the silhouette approach. I noticed that images with a solid color background tend to influence the algorithm more effectively.
Using Silhouette as guidance along with the written prompt
The output:
As you can see the output is very close to what I wanted. It’s the perfect scene.
Additional points
- The written prompt is crucial, as it provides the necessary details and context for the scene. However, using a silhouette can further refine the output by visually guiding the AI toward the composition and action you’re describing.
- Experimenting with different silhouette images can lead to more precise and varied results. Each silhouette provides a unique visual cue that influences how the AI interprets the scene, helping to refine details such as posture, motion, and interaction.
- It’s not a perfect method — sometimes you’ll need to keep experimenting until you achieve the ideal output. Adjusting silhouettes, refining your prompt, and trying different background colors can all help fine-tune the results.
- Try experimenting with different background colors, as they can influence the AI’s interpretation of lighting, mood, and depth in the scene. A background that complements the overall atmosphere of your prompt can enhance the final output.
Extra
Another tip is to use a person (not just a silhouette) against a background that matches the scene’s lighting. This can create interesting effects and influence the algorithm in a unique way.
Example
The left image is an prompt image and the right side is the output.
Where to get silhouettes?
Well you can go to google and search for
“person performing action” + “silhouette”.
Or you can create them yourself by using photoshop , photopea, gimp, krita, etc…