Emu Edit: Edit images from written instructions
Emu Edit is a multi-task image editing model from Meta research that performs instruction-based edits across region-based, free-form and computer vision tasks.
What is Emu Edit?
Emu Edit is a multi-task image editing model presented by AI at Meta research. It is trained across a broad set of tasks, including region-based editing, free-form editing and computer vision work such as detection and segmentation, with all of them framed as generative tasks. Learned task embeddings steer the generation process toward the intended task, which the authors report improves how accurately the model follows an editing instruction. The project also covers few-shot adaptation through task inversion, where model weights stay frozen and only a task embedding is updated to fit a new task such as super-resolution or contour detection. Alongside the model, the team released a benchmark covering seven editing tasks and published Emu Edit's generations on that dataset for comparison.
Focus areas: Prompt Photo Editing.
Who is Emu Edit for?
- Researchers studying instruction-based image editing
- Machine learning engineers evaluating editing models
- Teams that need a benchmark for image editing tasks
Is Emu Edit worth it?
A research-grade image editing model with a public benchmark, but no published pricing or clear commercial product
Where Emu Edit is strong
- Trained across a wide range of editing tasks, including region-based, free-form and computer vision tasks
- Learned task embeddings help steer generation toward the correct task
- Public benchmark with seven editing tasks and released model generations
Where it falls short
- No pricing or commercial plan information is published on the site
- The site is a research page, so availability of a hosted editing product is unclear
How do you get started with Emu Edit?
- 01Open the Emu Edit research page
- 02Read the paper for architecture and training details
- 03Download the benchmark and Emu Edit's generations to compare results
Emu Edit: frequently asked questions
What kinds of edits can Emu Edit perform?
The model is trained on region-based editing, free-form editing and computer vision tasks such as detection and segmentation, all formulated as generative tasks.
How does Emu Edit follow an editing instruction?
It uses learned task embeddings that steer the generation process toward the correct generative task, which the authors say improves instruction accuracy.
Can Emu Edit adapt to tasks it was not trained on?
Yes. Through task inversion, the model weights stay frozen and only a task embedding is updated, which the authors show works for tasks such as super-resolution and contour detection.
Is there a benchmark for comparing image editing models?
The team released a benchmark with seven editing tasks, including background, global, style, remove, add, local and texture changes, plus Emu Edit's generations on that dataset.
Emu Edit alternatives
The closest tools in our directory, matched on category, focus areas and audience.
Relighting portraits and product shots with text or reference images
| Tool | Best for | Free plan | Starting price | Focus areas |
|---|---|---|---|---|
| Emu Edit | Researchers and engineers evaluating instruction-based image editing models | Unknown | — | Prompt Photo Editing |
| DraGAN | Controlling GANs through Point-based Manipulation. | Unknown | — | — |
| Flux1 Kontext | Targeted text-prompt edits that keep the rest of an image intact | No | $8.96/mo | Prompt Photo Editing |
| Nana Banana.Pro | Creators needing consistent characters across many images | No | $6.90/mo | Character Consistency, Prompt Photo Editing |
| TryNanoBanana.AI | Reusing one character across many generated images | No | — | Character Consistency, Prompt Photo Editing, Text to Image |
| Online Image Vectorizer | Online raster to vector image converter. | Unknown | — | — |
| Imagable | Prompt-based photo edits for product, interior and character images | Yes | $9.90/mo | Object Removal, Prompt Photo Editing, Style Transfer |
| IC Light AI | Relighting portraits and product shots with text or reference images | Yes | $7.50/mo | — |
| MimicBrush | Localized style and texture edits from reference images | Yes | $9.90/mo | — |
| Polarr Next | Photographers editing and culling large shoots consistently | Yes | $19.99/mo | — |
| Pristina.ai | 3D artists improving renders inside Blender or 3ds Max | Yes | $18/mo | — |

