# HiDream-E1.1 Guide to Image Editing and Local Model Setup > Published 2026-04-04, updated 2026-09-05 ยท https://www.promptzone.com/riya_morales/hidream-e-11-ai-model-launches-3614 --- title: "HiDream-E1.1 Guide to Image Editing and Local Model Setup" published: true description: Run HiDream-E1.1 with the official editing script, prepare its local weights, and understand direct instructions and refinement. tags: ai,imagegeneration --- HiDream-E1.1 is HiDream-ai's instruction-based image editing model, built on the HiDream-I1 image-generation family. It takes an existing image and an editing instruction, with downloadable weights and official Python and Gradio examples. The public model repository uses the identifier `HiDream-ai/HiDream-E1-1`. [Model card][card] [Project repository][repo] ## What are the key facts about HiDream-E1.1? | Field | Verified information | | --- | --- | | Developer | HiDream-ai. [Model card][card] | | Released | July 16, 2025. [Project repository][repo] | | Type | Instruction-based image editing built on HiDream-I1. [Model card][card] | | Size or parameters | Hugging Face reports 17 billion parameters for the published weights. [Model card][card] | | License and access | MIT transformer weights; separate VAE and text-encoder terms apply. Download identifier: `HiDream-ai/HiDream-E1-1`. [Model card][card] | | Where it runs | Official CUDA Python inference script and local Gradio demo. [Project repository][repo] [Inference source][script] | The sibling [HiDream-I1 generator guide](/santiago_nguyen/hidream-fast-ai-image-generator-4l2j) covers creating an image from a written brief. E1.1 addresses a different starting point: you already have an image and want to direct a change while judging what the output retains. ## What image edits can HiDream-E1.1 perform? The developer documents direct editing instructions for E1.1. Its predecessor's workflow used a combined editing instruction and target-image description; E1.1 no longer requires that prompt-refinement stage. The practical implication is a simpler input format for an experiment with a source image. [Project repository][repo] The release adds dynamic resolution, documented at approximately one million pixels, and the project reports improved image quality and editing accuracy relative to E1-Full. Treat that as the developer's comparison: the repository publishes editing evaluations, but your own images still need task-specific review. [Project repository][repo] A useful first test is a single material or color change. For example, ask to make a ceramic vase matte blue while retaining its position and the background. Write down the feature that should change and the features that should remain recognizable. This is a suggested acceptance test, not a claimed demonstration result. Review the change and preservation separately. An output can satisfy the requested color while also altering an object's outline, so a single overall preference score may hide the failure that matters to your project. Keep the source and result side by side when making the decision. ## What are the limits of HiDream-E1.1 editing? The official script depends on HiDream-I1-Full and a Llama text encoder as well as the editing weights. The repository warns that downloading the Llama component requires accepting its license and authenticating the Hugging Face account. Access to the editor's own weights does not remove that dependency. [Project repository][repo] [Inference source][script] Dynamic resolution does not mean the model processes every uploaded image at its original pixel dimensions. The source code resizes and may crop the input for processing, then resizes the generated result back to the original dimensions. Inspect fine detail and framing after this path instead of equating the output file dimensions with native inference resolution. [Inference source][script] The `refine_strength` value controls the denoising-stage switch, and the repository documents `0.0` as disabling refinement. [Project repository][repo] The supplied loader and pipeline contain the actual weight-loading logic, so retain their revision alongside that value when comparing outputs. [Inference source][script] [Editing pipeline source][pipeline] A universal minimum VRAM requirement and guaranteed generation time are not published in the cited E1.1 setup instructions. The official example loads multiple components on CUDA. Plan an initial hardware test before committing to a batch deadline. [Project repository][repo] [Inference source][script] ## How do you install and run HiDream-E1.1 locally? Clone the official HiDream-E1 repository into a working directory. In a suitable Python and CUDA environment, install the repository requirements, Flash Attention, and the Hugging Face CLI. The project recommends CUDA 12.4 for manual installation; its requirements file also installs Diffusers from the upstream repository. [Project repository][repo] [Requirements file][requirements] [CLI documentation][cli] Before inference, accept the required Llama model terms using your Hugging Face account, then authenticate with `hf auth login`. The project uses an older CLI command name in its README; `hf auth login` is the current documented Hugging Face interface. [Project repository][repo] [CLI documentation][cli] The E1.1 inference source reads editing shards from a local directory rather than downloading those shards through `from_pretrained`. Download the repository into the path it expects, preserving the `transformer` subdirectory. Run these commands from your chosen parent directory. [Inference source][script] [CLI documentation][cli] ```bash git clone https://github.com/HiDream-ai/HiDream-E1 cd HiDream-E1 python -m pip install -r requirements.txt python -m pip install -U flash-attn --no-build-isolation python -m pip install -U huggingface_hub hf auth login hf download HiDream-ai/HiDream-E1-1 --local-dir HiDream-ai/HiDream-E1-1 python inference_e1_1.py ``` That final command runs the repository's example. To process your own image, edit `input_image_path`, `instruction`, and `output_path` inside the script's `main()` function before running it. These are source-code configuration fields; the supplied script does not define equivalent command-line flags. [Inference source][script] Start with the documented defaults and a single instruction. The script saves the edited image and a JSON sidecar containing settings and image information. Keep those together so a later review can distinguish a prompt change from a seed, guidance, or refinement change. [Inference source][script] For an interactive local interface, install Gradio with `python -m pip install gradio`, then run the project's `python gradio_demo_1_1.py`. [Gradio installation][gradioinstall] [Demo source][demo] Use that interface when repeated image selection is more convenient than changing the example script. Follow its model-loading requirements rather than assuming the browser interface makes inference remote. [Project repository][repo] After a successful run, compare the processed image with your acceptance notes. Check the intended change first, then the preserved subject, background, and framing. If you adjust refinement, retain the earlier output as a reference and keep the other inputs fixed. ## How does HiDream-E1.1 compare with Qwen-Image-Edit? | Model | Documented input and workflow | | --- | --- | | HiDream-E1.1 | Existing image plus direct instruction; official script includes optional refinement. [Project repository][repo] | | Qwen-Image-Edit | Existing image plus instruction through its separately documented editing pipeline. [Qwen editing card][qwenedit] | Compare both with the same source and requested change, then judge edit success and preservation separately. The [ComfyUI pillar](/tomas_novak/comfyui-2026-the-complete-guide-to-power-user-ai-image-generation-1g17) is useful background if you later move into visual workflows; the reproducible path established here is the developer's Python implementation. ## What should you know before editing with HiDream-E1.1? ### How many parameters does HiDream-E1.1 have? The official HiDream-E1.1 Hugging Face repository reports 17 billion parameters for its published model weights. Its download identifier is `HiDream-ai/HiDream-E1-1`. [Model card][card] ### Does HiDream-E1.1 need a separate target-image description? HiDream-E1.1 accepts direct editing instructions without requiring a separate target-image description. Begin with an explicit change and evaluate how well the result follows it. [Project repository][repo] ### Why can HiDream-E1.1 inference fail to find the editing weights? The supplied HiDream-E1.1 script reads editing shards from the local `HiDream-ai/HiDream-E1-1/transformer` directory. Download the repository into the expected location or update `HIDREAM_E1_PATH` to point to your download. [Inference source][script] ### Does HiDream-E1.1's MIT license cover every component? HiDream-E1.1's model card assigns MIT to the transformer weights and identifies separate terms for the VAE and text encoders. The Llama component also requires the account access described in the project's setup instructions. [Model card][card] [Project repository][repo] ## Sources - [Gradio installation guide][gradioinstall] - [Official local Gradio demo source][demo] - [HiDream-E1.1 model card][card] - [HiDream-E1 project documentation][repo] - [Official E1.1 inference source][script] - [Hugging Face CLI documentation][cli] - [Qwen-Image-Edit model card][qwenedit] - [HiDream-E1 requirements][requirements] - [HiDream editing pipeline source][pipeline] [card]: https://huggingface.co/HiDream-ai/HiDream-E1-1 [repo]: https://github.com/HiDream-ai/HiDream-E1 [script]: https://raw.githubusercontent.com/HiDream-ai/HiDream-E1/main/inference_e1_1.py [cli]: https://huggingface.co/docs/huggingface_hub/guides/cli [qwenedit]: https://huggingface.co/Qwen/Qwen-Image-Edit [requirements]: https://github.com/HiDream-ai/HiDream-E1/blob/main/requirements.txt [pipeline]: https://raw.githubusercontent.com/HiDream-ai/HiDream-E1/main/pipeline_hidream_image_editing.py [gradioinstall]: https://github.com/gradio-app/gradio [demo]: https://raw.githubusercontent.com/HiDream-ai/HiDream-E1/main/gradio_demo_1_1.py ## Related guides on PromptZone - [AI Model Releases Timeline](/ai-model-releases)