Muse Image

You're Claude Code, working in my repository.

Task
Pick the Muse Image recipe on this page that fits my project (basic generation, web-grounded product images, a consistent series, or editing a photo across turns), tell me which and why, then build its first two steps: a Responses call that generates or edits an image, and a chained follow-up with previous_response_id and only the change. Save each image and print each turn's usage.

Read this first
Meta's docs page "Muse Image" as Markdown: https://dev.meta.ai/docs/cookbook/muse-image.md
It's the source of truth for request and response shapes. The same page for people: https://dev.meta.ai/docs/cookbook/muse-image

Setup
- Base URL: https://api.meta.ai/v1
- Auth: send Authorization: Bearer $MODEL_API_KEY, reading MODEL_API_KEY from the environment. Never hard-code, print or commit the key.
- Model: muse-image-1.0
- Python: the OpenAI SDK (pip install openai) with base_url="https://api.meta.ai/v1" and api_key=os.environ["MODEL_API_KEY"] works, or plain requests.

Facts to get right
- POST /v1/responses with model muse-image-1.0. The image is the base64 result of the output item with type image_generation_call: find it by type, since reasoning and message items can come first.
- previous_response_id carries the prior image forward: send only the change, never re-attach a prior render.
- Reference images are input_image parts (a public URL or a data URL) in one {"role": "user", "content": [...]} message after the input_text; a bare list of parts gets HTTP 400.
- With store: false nothing stays on the server: replay the prior image_generation_call item first in the next input instead.
- $0.01 per image, built-in search included; a turn takes 10 to 60 s. The OpenAI SDK doesn't read MODEL_API_KEY on its own: pass it.

Starter code (Python)
import base64
import os
from pathlib import Path

from openai import OpenAI

client = OpenAI(base_url="https://api.meta.ai/v1", api_key=os.environ["MODEL_API_KEY"])


def turn(out: str, **body):
    reply = client.responses.create(model="muse-image-1.0", **body)
    item = next(i for i in reply.output if i.type == "image_generation_call")
    Path(out).write_bytes(base64.b64decode(item.result))
    return reply, item


fox, _ = turn("fox.webp", input="a watercolor painting of a red fox sitting in a snowy pine forest, soft golden morning light")
hat, _ = turn("fox_hat.webp", previous_response_id=fox.id, input="add a small red wool hat on the fox's head")
print(fox.usage, hat.usage, sep="\n")

Done when
- Turn 1 is status completed, and its image_generation_call result decodes to an image file.
- The follow-up sends previous_response_id and a one-line change, and keeps what it didn't touch.
- Each turn's usage prints: input, cached, output and reasoning tokens.
- The key comes from MODEL_API_KEY and never appears in code, logs or commits.
- API errors (4xx, 429) surface their message instead of failing silently.