LaDe: Unified Multi-Layered Graphic Media Generation and Decomposition

Vlad-Constantin Lungu-Stan; Ionut Mironica; Mariana-Iuliana Georgescu

cs.CVMar 18, 2026

LaDe: Unified Multi-Layered Graphic Media Generation and Decomposition

Authors: Vlad-Constantin Lungu-Stan, Ionut Mironica, Mariana-Iuliana Georgescu

Organizations: Adobe Research, Romania

Abstract

Media design layer generation enables the creation of fully editable, layered design documents such as posters, flyers, and logos using only natural language prompts. Existing methods either restrict outputs to a fixed number of layers or require each layer to contain only spatially continuous regions, causing the layer count to scale linearly with design complexity. We propose LaDe (Layered Media Design), a latent diffusion framework that generates a flexible number of semantically meaningful layers. LaDe combines three components: an LLM-based prompt expander that transforms a short user intent into structured per-layer descriptions that guide the generation, a Latent Diffusion Transformer with a 4D RoPE positional encoding mechanism that jointly generates the full media design and its constituent RGBA layers, and an RGBA VAE that decodes each layer with full alpha-channel support. By conditioning on layer samples during training, our unified framework supports three tasks: text-to-image generation, text-to-layers media design generation, and media design decomposition. We compare LaDe to Qwen-Image-Layered on text-to-layers and image-to-layers tasks on the Crello test set. LaDe outperforms Qwen-Image-Layered in text-to-layers generation by improving text-to-layer alignment, as validated by two VLM-as-a-judge evaluators (GPT-4o mini and Qwen3-VL).

Figures & tables

Figure 2: Text-to-Image Generation with LaDe .

Figure 3: Text-to-Layers Generation with LaDe . The gray background is added to emphasize the while content.

Figure 4: Image-to-Layers Decomposition with LaDe . The samples are generated by two state-of-the-art proprietary GenAI frameworks called through their APIs.

Figure 5: Text-to-Layers generation pipeline (top). Given a short user prompt, LaDe expands it with additional information which is encoded by FlanT5 XXL [ 2 ] and given as additional input to the Diffusion Model. After denoising, the full media design along with RGBA layers are decoded through the RGBA decoder. Text-to-Image generation is obtained by setting the number of layers to 0, generating only the full media design. Image-to-Layers decomposition (bottom) starts from the original media design and goes through a captioning and layer splitting operation. The text information, along with the embedding of the input image is passed through the diffusion model. The rest of the pipeline is similar to the Text-to-Layers generation.

Figure 6: LaDe is trained with an enhanced LDM formula. The encoded images $x_{0}$ are noised into $x_{t}$ , split into tokens and packed in a linear format, then combined with the RoPE enhanced text embedding $emb_{p}$ to create the model input $z_{t}$ . The model learns to denoise the input into its original form $z_{0}$ , which can be re-assembled and decoded back in the original images

Metric	GPT-4o mini $\uparrow$				Qwen3-VL $\uparrow$
# of layers	2	3	4	5	2	3	4	5
Qwen-Image-T2I + Qwen-Image-Layered-I2L [ 28 ]	2.79	2.66	2.63	2.79	2.53	2.49	2.35	2.41
LaDe (ours)	3.58	3.64	3.94	3.92	3.20	3.37	4.07	4.01

Table 1: Media design Layer generation performance obtained with VLM-as-a-judge using GPT-4o mini [ 16 ] and Qwen3-VL-30B-A3B-Instruct [ 23 ] on the Crello [ 25 ] test set. The scores are between 1 and 5. LaDe is compared against Qwen-Image-T2I + Qwen-Image-Layered-I2L [ 28 ] . LaDe obtains the best performance regardless of the number of layers.

Metric	PSNR $\uparrow$				RGB L1 $\downarrow$				Qwen3-VL $\uparrow$
# of layers	2	3	4	5	2	3	4	5	2	3	4	5
Qwen-Image-Layered-I2L † [ 28 ]	31.59	30.99	30.14	29.49	4.22	4.40	5.02	5.12	3.56	3.40	3.23	3.07
LaDe (ours)	32.65	31.37	29.94	28.42	3.41	4.06	5.67	7.38	3.21	3.16	3.25	3.25

Table 2: Media design decomposition into RGBA layers performance on the Crello test set [ 25 ] in terms of PSNR, RGB L1 (distance between RGB channels weighted by the ground-truth alpha in the 0-255 interval) and VLM-as-a-judge with Qwen3-VL-30B-A3B-Instruct [ 23 ] (score between 1 and 5). LaDe performs similar to Qwen-Image-Layered-I2L [ 28 ] which was finetuned on the Crello training set. † : the model was finetuned on the Crello training set.

Figure 7: A sample of our training set that contains the media design along with its composing layers. We employ InternVL3 [ 31 ] to obtain captions for the media design and the layers.

Figure 8: Text-to-Layers generation results comparing our one-step framework LaDe with Qwen-Image-T2I+Qwen-Image-Layered-I2L [ 28 ] . LaDe is able to create structured RGBA layers with homogeneous details included on different layers.

Figure 9: Qualitative results showing the decomposition obtained by LaDe (ours) versus Qwen-Image-Layered-I2L [ 28 ] and LayerD [ 22 ] . Qwen-Image-Layered-I2L [ 28 ] duplicates the text (left) in the second and third layers while LayerD [ 22 ] hallucinates the second layer (left). LaDe is able to decompose the media designs into meaningful layers, being able to reconstruct the occluded area.

Metric	GPT-4o mini $\uparrow$	Qwen3-VL $\uparrow$
Fixed layer count	3.01	3.78
Variable layer count	3.06	3.85
RGB VAE	3.70	3.50
RGBA VAE	3.53	3.68

Table 3: Ablation results obtained on 100 prompts evaluated with VLM-as-a-judge using GPT-4o mini [ 16 ] and Qwen3-VL-30B-A3B-Instruct [ 23 ] . The scores are between 1 and 5. Better performance is obtained when training with a variable number of layers .

Figure 10: The effect of Eq. 2 on the qualitative results. We noticed the content of the layers is duplicated, therefore, we applied Eq. 2 and each layer content follows its specific text information.

Image Generation

Image generation studies models that create or edit images from text, examples, layouts, and other structured controls.

See all 463 papers

UniWorld-Design: From Pixel Generation to Layer-Native Design
Aug 4, 2026Zongjian Li, Zhiyuan Yan, Chenxu Bai +9Image Generation Compositional Image Generation
UniWorld Team1,2∗
PixelART: Image-to-Layer Decomposition without Latents or Text-to-Image Pretraining
Sep 19, 2026Zelin Jia, Zhao Zhang, Zhicong Tang +2Pixel-Space Diffusion Models Diffusion Transformer
School of Software, BNRist, Tsinghua University · Canva Research
CreatiParser: Generative Image Parsing of Raster Graphic Designs into Editable Layers
Apr 21, 2026Weidong Chen, Dexiang Hong, Zhendong Mao +4
School of Information Science and Technology, University of Science and Technology of China, Hefei 230027, China · School of Information Science and Technology, University of Science and Technology of China, and are also with the Institute of Artificial Intelligence, Hefei Comprehensive National Science Center, Hefei 230027, China · Individual Researcher +1

Aug 4, 2026J/K move · Enter open · S save

UniWorld-Design: From Pixel Generation to Layer-Native Design

Zongjian Li, Zhiyuan Yan, Chenxu Bai, Chen Chen+8

UniWorld Team1,2∗

Image Generation Compositional Image Generation Controllable Image Generation

Open paper AskPDF arXiv

We introduce UniWorld-Design, a framework that redefines image generation from flat pixel synthesis to structured visual composition, with semantic RGBA layers as the atomic units of generation, understanding, and editing. Our key insight is that pixels define how an image is rendered, whereas layers define how an image is created, understood, and edited. Just as human designers create and manipulate visual content through layers rather than raw pixels, UniWorld-Design equips multimodal generative models with a layer-native design space. UniWorld-Design comprises two models. The Text-to-RGBA (T2RGBA) model generates standalone RGBA assets directly from text. The Image-to-Layer (I2L) model conditions on a finished image, a global instruction and per-layer prompts, and jointly produces ordered, complete semantic RGBA layers. Its instruction interface supports top-level decomposition, recursive decomposition and targeted extraction, making layering an instruction-addressable operation for agentic editing. Because I2L learns complete semantic objects rather than visible-pixel partitions, its layers stay usable when moved or removed. On the Crello benchmark, I2L reduces per-layer RGB L1 error by 37% and achieves a 34% relative improvement in Alpha Soft IoU over Qwen-Image-Layered. Separately, T2RGBA achieves the highest CLIP Score, outperforming LayerDiffuse and OmniAlpha.

Abstract

Image Generation

Explore similar work

UniWorld-Design: From Pixel Generation to Layer-Native Design

PixelART: Image-to-Layer Decomposition without Latents or Text-to-Image Pretraining

CreatiParser: Generative Image Parsing of Raster Graphic Designs into Editable Layers