Meta Muse Image API on WaveSpeedAI: Text-to-Image and Edit at $0.011

Meta Muse Image is available on WaveSpeedAI as two endpoints, meta/muse-image/text-to-image and meta/muse-image/edit, both live since August 27, 2026 and both priced at a flat $0.011 per request. That puts Meta’s image model between Z-Image Turbo ($0.005) and Seedream 5.0 Lite ($0.035) in the WaveSpeedAI catalog, with a small parameter surface: a prompt, an aspect ratio, an output format, and for edit, up to 10 input images.

This post covers what the endpoints accept, what they cost, and where they sit against the other low-cost image models on the platform. All prices and parameters below come from the live model detail API.

Muse Image versus Muse Spark

Two Meta product lines share the Muse name, and they are easy to confuse.

• Muse Image is the image model. Meta announced it on July 7, 2026 as the first image generation model from Meta Superintelligence Labs. Meta describes it as a model that “follows instructions faithfully, edits with precision, and composes from multiple references,” and says it can use tools such as code execution and web search during generation (Meta AI blog). At announcement, Meta reported it held the No. 2 spot on Arena for text-to-image, single-image editing, and multi-image editing. This is the model behind the WaveSpeedAI endpoints.
• Muse Spark is Meta’s multimodal language model line. Muse Spark 1.3 was released on September 2, 2026 (Meta AI Research); the lower-priced Contributor tier is listed on OpenRouter at $0.10 per million input tokens and $0.20 per million output tokens, with the note that prompts and outputs may be used to improve Meta’s products (OpenRouter). Muse Spark takes text, images, video, and files as input and returns text. It does not generate images.

Meta’s own announcement positions Muse Image as a consumer feature in the Meta AI app, meta.ai, Instagram Stories, and WhatsApp (Meta Newsroom) and does not describe a developer API. WaveSpeedAI exposes it as a standard REST endpoint with the same submit-and-poll pattern as every other model on the platform.

If you followed the earlier coverage on this blog, see What Is Muse Image? for the model background and Muse Image API Watch for the access timeline. The access question in that post is now answered by these endpoints.

Endpoints and pricing

Endpoint (model_uuid)	Type	Price	Formula
meta/muse-image/text-to-image	text-to-image	$0.011 per image	flat, no modifiers
meta/muse-image/edit	image-to-image	$0.011 per run	flat, no modifiers

Neither endpoint has a pricing formula attached, so aspect_ratio, output_format, and the number of image_urls do not change the cost. A 21:9 PNG edit that takes 10 reference images bills the same $0.011 as a 1:1 WebP generation from a single prompt.

Text-to-image parameters

meta/muse-image/text-to-image has three inputs. Only prompt is required.

Parameter	Type	Required	Default	Allowed values
prompt	string	yes	none	free text describing the image
aspect_ratio	string	no	1:1	21:9, 16:9, 4:3, 3:2, 1:1, 2:3, 3:4, 9:16, 9:21
output_format	string	no	webp	webp, png, jpeg

There is no size, seed, steps, or guidance control. Resolution is chosen by the model from the aspect ratio. Each request returns one image.

Edit parameters

meta/muse-image/edit adds a required image_urls array and makes aspect_ratio optional with no default.

Parameter	Type	Required	Default	Notes
prompt	string	yes	none	the edit instruction; describe what should change and what should stay
image_urls	array of strings	yes	none	1 to 10 image URLs
aspect_ratio	string	no	none	same nine values as text-to-image; leave empty to keep the closest supported layout
output_format	string	no	webp	webp, png, jpeg

The 10-image ceiling is what makes the edit endpoint useful beyond simple retouching. Meta’s announcement describes the model composing “elements from many input reference images in the prompt, including people, objects, clothing, styles, and environments,” and the WaveSpeedAI schema exposes that directly: pass a product shot, a background, and a style reference in one image_urls array and describe how to combine them in prompt. There is no mask input; edits are prompt-guided only.

Request pattern

Both endpoints follow the standard WaveSpeedAI flow: POST https://api.wavespeed.ai/api/v3/<model_uuid> with a JSON body, then poll GET /api/v3/predictions/<id>/result until the status is completed and read the output URL.

{
  "prompt": "Replace the studio backdrop with a sunlit rooftop at golden hour, keep the product and its shadow unchanged",
  "image_urls": [
    "https://example.com/product-front.jpg",
    "https://example.com/rooftop-reference.jpg"
  ],
  "aspect_ratio": "4:3",
  "output_format": "png"
}

Send that body to meta/muse-image/edit. For text-to-image, drop image_urls and keep the rest.

Where it sits among low-cost image models

The catalog has several image models priced under $0.05. The comparison below uses live per-image prices for a single default-size output.

Model (model_uuid)	Price per image	Edit endpoint	Notable controls
wavespeed-ai/z-image/turbo	$0.005	inpaint variant at $0.02	size, seed, image + strength for img2img
meta/muse-image/text-to-image	$0.011	meta/muse-image/edit at $0.011, up to 10 inputs	aspect_ratio only
bytedance/seedream-v5.0-lite	$0.035	bytedance/seedream-v5.0-lite/edit at $0.035	size up to 4096 x 4096
google/nano-banana-2-lite/text-to-image	$0.04	google/nano-banana-2-lite/edit at $0.04	14 aspect ratios including 8:1 and 1:8

Some practical reads of that table:

• Cheapest raw generation is still Z-Image Turbo at less than half the Muse Image price, with a seed for reproducibility and explicit width and height. If you need bulk thumbnails or deterministic regeneration, start there. The Z-Image Turbo API tutorial covers its parameters.
• Cheapest multi-reference editing is Muse Image. Z-Image Turbo takes a single image for img2img; Seedream 5.0 Lite Edit and Nano Banana 2 Lite Edit cost three to four times more per run. If your workload is mostly edits that combine several references, Muse Image at $0.011 is the lowest price in the catalog for that job.
• Resolution control favors Seedream 5.0 Lite, which accepts explicit size values up to 4096 x 4096. Muse Image does not expose a size parameter, so if you need a specific pixel dimension rather than an aspect ratio you will need to resize downstream or pick a different model.
• Text in images was a headline claim in Meta’s announcement (legible infographics and how-to guides). We have not benchmarked this independently; treat it as a vendor claim until you test it on your own prompts.

For a qualitative comparison of the same three families, see Muse Image vs Nano Banana 2 vs Seedream 5.0.

When to use which endpoint

Use meta/muse-image/text-to-image when you want a low-cost generation with a simple prompt and one of nine aspect ratios, and you do not need a seed or exact pixel dimensions. Use meta/muse-image/edit when you have one to ten source images and an instruction, especially when the task is combining references (identity plus product plus scene) rather than a masked local fix.

The pricing model is worth planning around. Because both endpoints are flat-rate, you can estimate a batch cost as request count multiplied by $0.011 with no per-parameter surprises. A 10,000-image campaign run is $110 on either endpoint.

Try it

• Meta Muse Image Text-to-Image
• Meta Muse Image Edit

Both model pages include the live schema, a playground, and the current price.