Muapi brings instruction-based editing, masked inpainting, subject references, and image reframing into one API. Send an image URL, prompt, mask, or reference list depending on the model, then use the same submit-and-poll flow.
High-fidelity composited edits from multiple reference images, a prompt, and explicit resolution control.
Instruction editing from multiple reference images to combine, restyle, or merge elements.
Multi-image instruction editing with explicit width and height output control.
Uses mask_image_url to constrain an edit to a specific region.
Mask-driven object removal that fills the selected region without requiring a text prompt.
Keeps a subject consistent across a new prompt-driven scene from a reference image.
Character-consistency editing for recognizable subjects across generated scenes.
Outpaints an image to a target aspect ratio without changing the original content.
An AI image editing API accepts an existing image and changes it according to a prompt, mask, or reference image. Unlike a text-to-image endpoint, the source image anchors the composition, subject, or region that should remain unchanged.
Muapi exposes 70+ image-editing models through one request format and one credit balance. Choose instruction editing for natural-language changes, masked editing for precise local work, reference editing for subject consistency, or reframing for a new canvas shape.
Describe an object replacement, style change, lighting adjustment, or scene transformation in natural language.
Pass a mask_image_url when only a defined region should change and the rest of the composition must stay fixed.
Use images_list or model-specific reference fields to combine products, people, or styles from multiple inputs.
Reference models preserve a character or product identity while generating a new scene from a prompt.
Expand an image to a new aspect ratio without simply cropping the original subject.
There is no subscription or minimum commitment; each edit is charged according to its selected model.
| Model | Mode | Price | Best For |
|---|---|---|---|
| Nano Banana Pro Edit | Multi-Image | $0.12/generation | High-fidelity compositing |
| GPT-4o Edit | Masked | $0.04/generation | Targeted local edits |
| MiniMax Subject Reference | Reference | $0.01/generation | Consistent subjects |
| Ideogram v3 Reframe | Reframe | $0.15/generation | Aspect-ratio expansion |
The exact input field names differ by model. Use the Playground schema for the selected model before sending a production request.
curl -X POST https://api.muapi.ai/api/v1/gpt4o-edit \
-H "Content-Type: application/json" \
-H "x-api-key: YOUR_API_KEY" \
-d '{
"image_url": "https://example.com/input.jpg",
"mask_image_url": "https://example.com/mask.png",
"prompt": "replace the highlighted object with a red ceramic vase"
}'import requests
response = requests.post(
"https://api.muapi.ai/api/v1/gpt4o-edit",
headers={"x-api-key": "YOUR_API_KEY"},
json={
"image_url": "https://example.com/input.jpg",
"mask_image_url": "https://example.com/mask.png",
"prompt": "replace the highlighted object with a red ceramic vase",
},
)
request_id = response.json()["request_id"]
result = requests.get(
f"https://api.muapi.ai/api/v1/predictions/{request_id}/result",
headers={"x-api-key": "YOUR_API_KEY"},
)
print(result.json())Instruction editing describes the change in a prompt, while masked editing limits the change to a supplied region.
Yes, models such as Nano Banana Pro Edit, FLUX Kontext Pro, and Qwen Image Edit Plus accept multiple reference images.
No. Use a mask only when you need precise local control. Instruction and reference models can edit from images and prompts alone.
Yes. Use a subject-reference or character-consistency model and provide a clear reference image.
The listed models range from about $0.01 to $0.15 per generation; the exact price depends on the selected model and options.