이미지: X — 미디어·생성AI
Summary
- Higgsfield AI has unveiled 'Higgsfield Layers,' which automatically separates flat poster images into headline, graphic badge, and masked person layers
- The company said the layer-decomposition work that used to take hours can now be finished in seconds, and explained that the results are ready to use without further edits
- While generative image models compete on speed, Higgsfield is targeting the design workflow in the area of "decomposition and reconstruction" rather than "generation"
- 제품명
- Higgsfield Layers
- 핵심 기능
- 헤드라인·그래픽 배지·마스킹된 인물을 개별 레이어로 자동 분리
- 처리 시간
- 수 초 (기존 수동 작업은 수 시간 소요)
- 공개 시점
- 2026년 8월 12일 (X 게시 기준)
- 출력 상태
- 회사 측 표현으로 '프로덕션에 바로 쓸 수 있는' 레이어
- 공개 예시
- 'SUMMER DROP', 'OLD MASTERS' 등 텍스트가 담긴 시즌 포스터
A poster torn apart in seconds
Opening a single-image poster and pulling out the headline, logo badge, and person cutout as separate elements has long been a common chore for designers. Without the original source file, this meant redrawing text, cutting out figures from backgrounds, and recreating badge shapes from scratch. Higgsfield AI unveiled a tool that automates this work, called 'Higgsfield Layers,' on August 12. In a post on X (formerly Twitter), the company said, "Decomposing posters like this used to take hours."
What gets split, and how
The example image released showed a poster announcing a seasonal sale. It was a typical piece of graphic design containing phrases like 'SUMMER DROP,' '06.14–07.20 SUN,' 'OLD MASTERS,' and 'CANVAS TO CONCRETE,' along with a pink star-shaped 'NEW' badge and a year marking next to a barcode. Higgsfield Layers extracted the headline text, graphic badge, and masked person from this single image as separate layers. Comparing the input and output images side by side, the text content remained identical, suggesting no original information was lost during the layer separation process.
| Item | Manual work (existing) | Higgsfield Layers |
|---|---|---|
| Time required | Several hours | A few seconds |
| Output | Requires designer rework | Ready for production use (per company claim) |
| Target elements | Handled separately with individual tools | Headline, badge, and person separated simultaneously |
Why this was a difficult task
Posters finished for print or social media typically survive only as final images (JPG, PNG) without the original editable source file, such as a PSD. Re-separating text, graphics, and people from such images requires manual work — tracing outlines and masking out backgrounds by hand. Add in re-matching fonts or recreating badges as vectors, and it's not uncommon for practitioners to spend actual hours on the task. Higgsfield has focused on automating this repetitive labor.
A different direction amid the image-generation race
The image generation model industry has recently been locked in a fierce speed race. On August 9, Black Forest Labs unveiled FLUX.2 klein 9B, a 9-billion-parameter model, saying it supports sub-one-second inference via 4-step distillation. Around the same time, Runway added 'P-Image-Ideogram,' which takes at least 0.6 seconds to generate an image, to its platform. Zhipu AI also touted strong text-rendering capabilities with GLM-Image, which combines autoregression and diffusion. While all of these companies compete on the speed of "creating images from nothing," Higgsfield Layers has dug into the post-processing space of restoring already-finished images to an editable state. Each company appears to be staking out ground on a different axis — generation versus editing.
So what changes
Designers and marketers looking to repurpose a finished poster, social media card, or banner image for another campaign have often hit a wall: no original layers to work with. If Higgsfield Layers works as the company describes, the effort needed to turn already-published images back into editable form could shrink significantly at the working level. This aligns with a broader trend of generative AI expanding its use beyond "creating something new" to "reconstructing something that already exists." That said, this release is based on the company's own statements and a single example image, so whether the same level of separation quality holds up across diverse design styles and more complex layouts remains to be confirmed with additional cases.
