Update model card with metadata, paper link, and sample usage

This PR improves the model card by:
- Adding the `text-to-image` pipeline tag.
- Adding the `diffusers` library name based on the provided code examples.
- Linking the model to the research paper: [Rethinking Global Text Conditioning in Diffusion Transformers](https://huggingface.co/papers/2602.09268).
- Adding a link to the official GitHub repository.
- Providing a sample usage code snippet for the FLUX.1-schnell implementation.

README.md

@@ -1,3 +1,76 @@
----
-license: apache-2.0
----
+---
+license: apache-2.0
+pipeline_tag: text-to-image
+library_name: diffusers
+---
+
+# Rethinking Global Text Conditioning in Diffusion Transformers
+
+This repository contains the research artifacts and implementation for the paper [Rethinking Global Text Conditioning in Diffusion Transformers](https://huggingface.co/papers/2602.09268).
+
+## Summary
+
+Diffusion transformers typically incorporate textual information via attention layers and a modulation mechanism using a pooled text embedding. This research addresses whether modulation-based text conditioning is necessary and reveals that while conventional usage of pooled embeddings contributes little to performance, they can provide significant gains when used as **guidance**. 
+
+This approach, called **Modulation Guidance**, enables controllable shifts toward desirable properties such as increased complexity, better aesthetics, and improved hand generation. It is:
+- **Training-free**: No additional training or fine-tuning is required.
+- **Efficient**: It incurs negligible runtime overhead.
+- **Versatile**: Applicable to various models including FLUX, SD3.5, HunyuanVideo, and COSMOS.
+
+For more information, please refer to the [official GitHub repository](https://github.com/quickjkee/modulation-guidance).
+
+## Sample Usage (FLUX.1-schnell)
+
+The following example demonstrates how to apply modulation guidance to enhance image complexity using the `diffusers` library. This requires the helper functions provided in the [official repository](https://github.com/quickjkee/modulation-guidance).
+
+```python
+import types
+import torch
+from functools import partial
+from diffusers import FluxPipeline
+# Assumes the 'models' directory from the official repo is in your python path
+from models.flux_schnell import encode_prompt, forward_modulation_guidance
+
+# 1. Load the model
+pipe = FluxPipeline.from_pretrained("black-forest-labs/FLUX.1-schnell", torch_dtype=torch.bfloat16).to('cuda')
+
+# 2. Define hyperparameters for complexity guidance
+prompt = 'A wolf on a plain background'
+prompt_positive = 'Extremely complex, the highest quality'
+prompt_negative = 'Very simple, no details at all'
+w = 3             # Guidance strength
+start_layer = 5   # Layer to start applying guidance
+
+# 3. Get pooled CLIP embeddings
+clip_positive = encode_prompt(pipe=pipe, prompt=prompt_positive)
+clip_negative = encode_prompt(pipe=pipe, prompt=prompt_negative)
+
+# 4. Override the transformer forward pass with modulation guidance
+forward_fn = partial(forward_modulation_guidance, 
+                     pooled_projections_1=clip_positive, 
+                     pooled_projections_0=clip_negative,
+                     w=w, start_layer=start_layer)
+pipe.transformer.forward = types.MethodType(forward_fn, pipe.transformer)
+
+# 5. Run generation
+seed = 0
+image = pipe([prompt],
+              guidance_scale=0.0,
+              num_inference_steps=4,
+              max_sequence_length=256,
+              generator=torch.Generator("cpu").manual_seed(seed),
+              output_type='pil').images[0]
+
+image.save("output.png")
+```
+
+## Citation
+
+```bibtex
+@article{starodubcev2025rethinking,
+  title={Rethinking Global Text Conditioning in Diffusion Transformers},
+  author={Starodubcev, Nikita and Pakhomov, Daniil and Wu, Zongze and Drobyshevskiy, Ilya and Liu, Yuchen and Wang, Zhonghao and Zhou, Yuqian and Lin, Zhe and Baranchuk, Dmitry},
+  journal={arXiv preprint arXiv:2602.09268},
+  year={2025}
+}
+```