baidu
/

ERNIE-Image

@@ -1,97 +1,112 @@
 ---
 license: apache-2.0
 ---
-# baidu/ERNIE-Image Model Cards
-## Model Details
-### Model Description
-**ERNIE-Image** is a text-to-image generation model developed by the ERNIE team at Baidu.
-In terms of image quality, ERNIE-Image is on par with current state-of-the-art models. It demonstrates significant advantages in handling complex instructions, particularly in tasks that require **accurate text rendering** and **knowledge-intensive generation**.
-### Key Features
-* **Precise text rendering**: Especially strong in dense or complex text scenarios
-* **Excellent instruction following**: Accurately interprets and executes complex prompts
-* **High-quality portraits and stylized images**: Strong performance in both realism and artistic styles
-### Model Architecture
-ERNIE-Image consists of the following components:
-* An 8B-parameter Diffusion Transformer (DiT)
-* A 3B text encoder from Ministral
-* A VAE based on flux2.dev
-* A prompt enhancer fine-tuned using Ministral 3B
-### Deployment
-Thanks to its relatively compact model size, ERNIE-Image can be deployed on **consumer-grade GPUs (e.g., 24GB VRAM)**, making high-quality image generation more accessible and practical.
-## Evaluation
-### Benchmark
-### Showcase
-<style>
-.masonry {
-  column-count: 2;        /* 两列 */
-  column-gap: 12px;
-}
-.card {
-  break-inside: avoid;    /* 防止卡片被切断 */
-  margin-bottom: 12px;
-}
-.card img {
-  width: 100%;            /* 宽度统一 */
-  height: auto;           /* 高度自适应（关键） */
-  display: block;
-  border-radius: 8px;
-}
-</style>
-<section id="shape-3160722256177163633" class="tab-panel active">
-  <div class="masonry">
-    <article class="card">
-      <img src="https://cdn-uploads.huggingface.co/production/uploads/6358acfe0e4fef21982a929b/ukAjGbYZwG4jPRRs9UgJ0.jpeg">
-    </article>
-    <article class="card">
-      <img src="https://cdn-uploads.huggingface.co/production/uploads/6358acfe0e4fef21982a929b/nD1aI60oXzAuGnhBA0P8T.jpeg">
-    </article>
-    <article class="card">
-      <img src="https://cdn-uploads.huggingface.co/production/uploads/6358acfe0e4fef21982a929b/YveUuXuSIiBWDs-U_mYqI.jpeg">
-    </article>
-    <article class="card">
-      <img src="https://cdn-uploads.huggingface.co/production/uploads/6358acfe0e4fef21982a929b/_0RGqPUeIB0r5SigsPgjz.jpeg">
-    </article>
-    <article class="card">
-      <img src="https://cdn-uploads.huggingface.co/production/uploads/6358acfe0e4fef21982a929b/RLQonbN1cRg4cHyx3GnCS.jpeg">
-    </article>
-    <article class="card">
-      <img src="https://cdn-uploads.huggingface.co/production/uploads/6358acfe0e4fef21982a929b/C8NxWvTeC0-tqZuBmVPMT.jpeg">
-    </article>
-  </div>
-</section>
-## Uses
-### Installation & Download
-Install the latest version of diffusers:
-```
-pip install git+https://github.com/huggingface/diffusers
-```
-Download the model:
-```
-pip install -U huggingface_hub
-HF_XET_HIGH_PERFORMANCE=1 hf download baidu/ERNIE-Image
-```
 ### Recommended Parameters
 - Resolution:
     - 1024x1024
@@ -103,53 +118,58 @@ HF_XET_HIGH_PERFORMANCE=1 hf download baidu/ERNIE-Image
     - 1200x896
 - Guidance scale: 4.0
 - Inference steps: 50
-### Usage Example
-```
-import os
-os.environ["CUBLAS_WORKSPACE_CONFIG"] = ":4096:8"
-import random
-import numpy as np
 import torch
 from diffusers import ErnieImagePipeline
-seed = 42
-print(f"seed: {seed}")
-random.seed(seed)
-np.random.seed(seed)
-torch.manual_seed(seed)
-torch.cuda.manual_seed_all(seed)
-torch.backends.cudnn.deterministic = True
-torch.use_deterministic_algorithms(True)
-torch.backends.cudnn.benchmark = False
-# 加载 pipeline
 pipe = ErnieImagePipeline.from_pretrained(
-    "baidu/ERNIE-Image",
     torch_dtype=torch.bfloat16,
-)
-pipe = pipe.to("cuda")
-pipe.transformer.eval()
-pipe.vae.eval()
-pipe.text_encoder.eval()
-pipe.pe.eval()
-# 设置随机种子
-generator = torch.Generator(device="cuda").manual_seed(seed)
-# 生成图片
-output = pipe(
-    prompt=prompt,
     height=1024,
     width=1024,
     num_inference_steps=50,
-    guidance_scale=5.0,
-    generator=generator,
-    num_images_per_prompt=1,
-    use_pe=True
-)
-revised_prompt = output.revised_prompts
-images = output.images
-image.save(f"./hf_output_0.png")
-print(revised_prompt)
-```

 ---
 license: apache-2.0
+pipeline_tag: text-to-image
+library_name: diffusers
+tags:
+  - text-to-image
 ---
+# ERNIE-Image
+<p align="center">
+  <img src="mosaic.jpg" alt="ERNIE-Image Mosaic" width="60%">
+</p>
+<p align="center">
+  <a href="https://huggingface.co/Baidu/ERNIE-Image">🤗 ERNIE-Image</a> &nbsp;|&nbsp;
+  <a href="https://huggingface.co/Baidu/ERNIE-Image-Turbo">🤗 ERNIE-Image-Turbo</a> &nbsp;|&nbsp;
+  <a href="TODO">💻 GitHub</a> &nbsp;|&nbsp;
+  <a href="TODO">📖 Blog</a> &nbsp;|&nbsp;
+  <a href="TODO">🖼️ Gallery</a>
+</p>
+ERNIE-Image is an open text-to-image generation model developed by the ERNIE-Image team at Baidu. It is built on a single-stream Diffusion Transformer (DiT) and paired with a lightweight Prompt Enhancer that expands brief user inputs into richer structured descriptions. With only 8B DiT parameters, it reaches state-of-the-art performance among open-weight text-to-image models. The model is designed not only for strong visual quality, but also for controllability in practical generation scenarios where accurate content realization matters as much as aesthetics. In particular, ERNIE-Image performs strongly on complex instruction following, text rendering, and structured image generation, making it well suited for commercial posters, comics, multi-panel layouts, and other content creation tasks that require both visual quality and precise control. It also supports a broad range of visual styles, including realistic photography, design-oriented imagery, and more stylized aesthetic outputs.
+**Highlights:**
+- **Compact but strong**: Despite its compact 8B scale, ERNIE-Image remains highly competitive with substantially larger open-weight models across a range of benchmarks.
+- **Text rendering**: ERNIE-Image performs particularly well on dense, long-form, and layout-sensitive text, making it a strong choice for posters, infographics, UI-like images, and other text-heavy visual content.
+- **Instruction following**: The model is able to follow complex prompts involving multiple objects, detailed relationships, and knowledge-intensive descriptions with strong reliability.
+- **Structured generation**: ERNIE-Image is especially effective for structured visual tasks such as posters, comics, storyboards, and multi-panel compositions, where layout and organization are critical.
+- **Style coverage**: In addition to clean and readable design-oriented outputs, the model also supports realistic photography and distinctive stylized aesthetics, including softer and more cinematic visual tones.
+- **Practical deployment**: Thanks to its compact size, ERNIE-Image can run on consumer GPUs with 24G VRAM, which lowers the barrier for research, downstream use, and model adaptation.
+## Released Versions
+### [ERNIE-Image](https://huggingface.co/Baidu/ERNIE-Image): Our **SFT model**, delivers stronger general-purpose capability and instruction fidelity in typically **50 inference steps**.
+### [ERNIE-Image-Turbo](https://huggingface.co/Baidu/ERNIE-Image-Turbo): Our **Turbo model**, optimized by **DMD and RL**, achieves faster speed and higher aesthetics in only **8 inference steps**.
+## Benchmark
+### GENEval
+| Model | Single Object | Two Object | Counting | Colors | Position | Attribute Binding | Overall |
+|---|---:|---:|---:|---:|---:|---:|---:|
+| ERNIE-Image (w/o PE) | **1.0000** | 0.9596 | 0.7781 | 0.9282 | 0.8550 | **0.7925** | **0.8856** |
+| ERNIE-Image (w/ PE) | 0.9906 | 0.9596 | 0.8187 | 0.8830 | **0.8625** | 0.7225 | 0.8728 |
+| Qwen-Image | 0.9900 | 0.9200 | **0.8900** | 0.8800 | 0.7600 | 0.7700 | 0.8683 |
+| ERNIE-Image-Turbo (w/o PE) | **1.0000** | **0.9621** | 0.7906 | 0.9202 | 0.7975 | 0.7300 | 0.8667 |
+| ERNIE-Image-Turbo (w/ PE) | 0.9938 | 0.9419 | 0.8375 | 0.8351 | 0.7950 | 0.7025 | 0.8510 |
+| FLUX.2-klein-9B | 0.9313 | 0.9571 | 0.8281 | 0.9149 | 0.7175 | 0.7400 | 0.8481 |
+| Z-Image | **1.0000** | 0.9400 | 0.7800 | **0.9300** | 0.6200 | 0.7700 | 0.8400 |
+| Z-Image-Turbo | **1.0000** | 0.9500 | 0.7700 | 0.8900 | 0.6500 | 0.6800 | 0.8233 |
+### OneIG-EN
+| Model | Alignment | Text | Reasoning | Style | Diversity | Overall |
+|---|---:|---:|---:|---:|---:|---:|
+| Nano Banana 2.0 | 0.8880 | 0.9440 | 0.3340 | **0.4810** | **0.2450** | **0.5780** |
+| Seedream 4.5 | 0.8910 | **0.9980** | 0.3500 | 0.4340 | 0.2070 | 0.5760 |
+| ERNIE-Image (w/ PE) | 0.8678 | 0.9788 | **0.3566** | 0.4309 | 0.2411 | 0.5750 |
+| Seedream 4.0 | **0.8920** | 0.9830 | 0.3470 | 0.4530 | 0.1910 | 0.5730 |
+| ERNIE-Image-Turbo (w/ PE) | 0.8676 | 0.9666 | 0.3537 | 0.4191 | 0.2212 | 0.5656 |
+| ERNIE-Image (w/o PE) | 0.8909 | 0.9668 | 0.2950 | 0.4471 | 0.1687 | 0.5537 |
+| Z-Image | 0.8810 | 0.9870 | 0.2800 | 0.3870 | 0.1940 | 0.5460 |
+| Qwen-Image | 0.8820 | 0.8910 | 0.3060 | 0.4180 | 0.1970 | 0.5390 |
+| ERNIE-Image-Turbo (w/o PE) | 0.8795 | 0.9488 | 0.2913 | 0.4277 | 0.1232 | 0.5341 |
+| FLUX.2-klein-9B | 0.8871 | 0.8657 | 0.3117 | 0.4417 | 0.1560 | 0.5324 |
+| Qwen-Image-2512 | 0.8760 | 0.9900 | 0.2920 | 0.3380 | 0.1510 | 0.5300 |
+| GLM-Image | 0.8050 | 0.9690 | 0.2980 | 0.3530 | 0.2130 | 0.5280 |
+| Z-Image-Turbo | 0.8400 | 0.9940 | 0.2980 | 0.3680 | 0.1390 | 0.5280 |
+### OneIG-ZH
+| Model | Alignment | Text | Reasoning | Style | Diversity | Overall |
+|---|---:|---:|---:|---:|---:|---:|
+| Nano Banana 2.0 | **0.8430** | 0.9830 | **0.3110** | **0.4610** | 0.2360 | **0.5670** |
+| ERNIE-Image (w/ PE) | 0.8299 | 0.9539 | 0.3056 | 0.4342 | 0.2478 | 0.5543 |
+| Seedream 4.0 | 0.8360 | 0.9860 | 0.3040 | 0.4430 | 0.2000 | 0.5540 |
+| Seedream 4.5 | 0.8320 | 0.9860 | 0.3000 | 0.4260 | 0.2130 | 0.5510 |
+| Qwen-Image | 0.8250 | 0.9630 | 0.2670 | 0.4050 | **0.2790** | 0.5480 |
+| ERNIE-Image-Turbo (w/ PE) | 0.8258 | 0.9386 | 0.3043 | 0.4208 | 0.2281 | 0.5435 |
+| Z-Image | 0.7930 | **0.9880** | 0.2660 | 0.3860 | 0.2430 | 0.5350 |
+| ERNIE-Image (w/o PE) | 0.8421 | 0.8979 | 0.2656 | 0.4212 | 0.1772 | 0.5208 |
+| Qwen-Image-2512 | 0.8230 | 0.9830 | 0.2720 | 0.3420 | 0.1570 | 0.5150 |
+| GLM-Image | 0.7380 | 0.9760 | 0.2840 | 0.3350 | 0.2210 | 0.5110 |
+| Z-Image-Turbo | 0.7820 | 0.9820 | 0.2760 | 0.3610 | 0.1340 | 0.5070 |
+| ERNIE-Image-Turbo (w/o PE) | 0.8326 | 0.9086 | 0.2580 | 0.4002 | 0.1316 | 0.5062 |
+| FLUX.2-klein-9B | 0.8201 | 0.4920 | 0.2599 | 0.4166 | 0.1625 | 0.4302 |
+### LongTextBench
+| Model | LongText-Bench-EN | LongText-Bench-ZH | Avg |
+|---|---:|---:|---:|
+| Seedream 4.5 | **0.9890** | **0.9873** | **0.9882** |
+| ERNIE-Image (w/ PE) | 0.9804 | 0.9661 | 0.9733 |
+| GLM-Image | 0.9524 | 0.9788 | 0.9656 |
+| ERNIE-Image-Turbo (w/ PE) | 0.9675 | 0.9636 | 0.9655 |
+| Nano Banana 2.0 | 0.9808 | 0.9491 | 0.9650 |
+| ERNIE-Image-Turbo (w/o PE) | 0.9602 | 0.9675 | 0.9639 |
+| ERNIE-Image (w/o PE) | 0.9679 | 0.9594 | 0.9636 |
+| Qwen-Image-2512 | 0.9561 | 0.9647 | 0.9604 |
+| Qwen-Image | 0.9430 | 0.9460 | 0.9445 |
+| Z-Image | 0.9350 | 0.9360 | 0.9355 |
+| Seedream 4.0 | 0.9214 | 0.9261 | 0.9238 |
+| Z-Image-Turbo | 0.9170 | 0.9260 | 0.9215 |
+| FLUX.2-klein-9B | 0.8642 | 0.2183 | 0.5413 |
+## Quick Start
 ### Recommended Parameters
 - Resolution:
     - 1024x1024
     - 1200x896
 - Guidance scale: 4.0
 - Inference steps: 50
+### Diffusers
+`pip install git+https://github.com/huggingface/diffusers`
+```python
 import torch
 from diffusers import ErnieImagePipeline
 pipe = ErnieImagePipeline.from_pretrained(
+    "Baidu/ERNIE-Image",
     torch_dtype=torch.bfloat16,
+).to("cuda")
+image = pipe(
+    prompt="A cinematic movie poster of a futuristic city at night with clear neon signage.",
     height=1024,
     width=1024,
     num_inference_steps=50,
+    guidance_scale=4.0,
+    use_pe=True # use prompt enhancer
+).images[0]
+image.save("output.png")
+```
+### SGLang
+Install the latest version of sglang:
+```
+git clone https://github.com/sgl-project/sglang.git
+```
+Start the server:
+```bash
+sglang serve --model-path baidu/ERNIE-Image
+```
+Send a generation request:
+```bash
+curl -X POST http://localhost:30000/generate \
+  -H "Content-Type: application/json" \
+  -d '{
+    "prompt": "一只黑白相间的中华田园犬",
+    "height": 1024,
+    "width": 1024,
+    "num_inference_steps": 50,
+    "guidance_scale": 4.0,
+    "use_pe": true
+  }' \
+  --output output.png
+```