Have you ever refined an AI-generated image only to lose the parts you liked? When you tweak a prompt—say, shift a subject left or remove an object—the entire scene often changes unpredictably. Backgrounds shift, lighting softens, and textures rewrite themselves, undoing hours of careful refinement.
This isn’t how traditional artists work. Photographers composite layers. Concept artists stack elements. Yet most AI image tools treat the final output as a single, unbreakable whole. What if you could break that image back into controllable parts?
That’s the idea behind a new workflow built on the Melius platform. Instead of regenerating an entire scene with each prompt adjustment, the system decomposes the image into reusable layers—backgrounds, midgrounds, foregrounds, and subjects—letting you move, scale, or swap elements without redrawing.
How the layer extraction pipeline works
The workflow begins with a Melius canvas, where an analyzer LLM dissects any input image. It generates a JSON blueprint that identifies up to seven candidate layers: one background, one midground, one foreground, and up to four distinct subjects. Each layer gets its own isolation prompt, extracted in parallel by seven specialized LLMs.
Next, seven NanoBanana Pro nodes regenerate each layer onto a chroma green background, preserving the original position, scale, and lighting. Background-removal nodes strip the green, leaving transparent layers. Finally, a unified NanoBanana Pro pass composites all seven layers into a cohesive, lighting-coherent image.
The result? You can drag the truck from the foreground to the midground, scale the background sky, or swap one subject for another—all without losing the original lighting, textures, or scene integrity.
{
"layers": [
{ "type": "background", "prompt": "snowy city street at dusk with soft streetlights" },
{ "type": "foreground", "prompt": "red delivery truck parked by a curb" },
{ "type": "subject", "prompt": "person in winter coat walking toward camera" }
]
}Why traditional AI image editing falls short
Most AI image tools rely on single-pass generation, treating the image as an indivisible unit. Tools like inpainting or reference conditioning offer partial fixes, but they still force the model to guess at occluded or missing regions. For example, moving a truck from one side of a street might require the model to invent the snow underneath from scratch.
This approach breaks the creative process. Artists rarely work this way. They layer elements, adjust compositions, and refine details incrementally. The new Melius workflow mirrors that process by never baking an image into a final state until the very last step. You treat the original as a brief, decompose it into stackable elements, and recompose only after arranging everything to your liking.
What’s included in the full workflow guide
The full breakdown, published on Scopeful Pro, details every node in the pipeline. It includes the system prompts used by the analyzer and extractor LLMs, the initial mistakes encountered during testing, and planned improvements for a future version.
For anyone curious about the technical implementation, the guide provides:
- Step-by-step node configurations for the Melius canvas
- The analyzer and extractor LLM prompts used to generate layer isolation instructions
- Common pitfalls and lessons learned during early testing
- Suggested adjustments for version 2 of the workflow
The workflow itself is available as a downloadable canvas, allowing users to adapt it for their own projects. A short video demo—showing layers "exploding" into separate elements—is also available on the author’s X feed.
A new standard for AI image editing?
This approach marks a shift from reactive AI image editing to proactive, layer-based composition. While still in its early stages, the workflow demonstrates how AI-generated images can adopt the flexibility of traditional digital art tools. The next step may involve refining the automation, improving layer accuracy, or expanding compatibility with more AI image models.
For artists, designers, and creators frustrated by the limitations of prompt-based refinements, this method offers a promising alternative. It’s not just about editing images—it’s about reclaiming control over the creative process.
AI summary
AI ile oluşturulan görüntüleri katmanlara ayırarak yeniden düzenleyin. Melius’un sunduğu bu yenilikçi yöntemle unsurları dilediğiniz gibi hareket ettirin ve nihai kompozisyonu oluşturun.
