AgentSkillsCN

Astria Api

在调用Astria的API以获取曲调、提示词、素材包或进行图像生成（Gemini/Seedream）时使用此功能。

SKILL.md
--- frontmatter
description: Use when making API calls to Astria for tunes, prompts, packs, or image generation (Gemini/Seedream)

Astria API Reference

API credentials are available as environment variables:

  • •$ASTRIA_AUTH_TOKEN — Bearer token
  • •$ASTRIA_BASE_URL — API base URL
  • •$WORKSPACE_ID — Workspace ID (if set)
  • •$GEMINI_TUNE_ID — Gemini tune ID
  • •$SEEDREAM_TUNE_ID — Seedream tune ID

Use these directly in curl commands. Always include both Authorization and X-Workspace-Id headers.

Tunes (Fine-tuned models)

List tunes

GET $ASTRIA_BASE_URL/tunes

Example: curl -s -H "Authorization: Bearer $ASTRIA_AUTH_TOKEN" -H "X-Workspace-Id: $WORKSPACE_ID" "$ASTRIA_BASE_URL/tunes" | jq '.[].id, .[].title, .[].name', .[].orig_images'

Get tune

GET $ASTRIA_BASE_URL/tunes/:id

Example: curl -s -H "Authorization: Bearer $ASTRIA_AUTH_TOKEN" -H "X-Workspace-Id: $WORKSPACE_ID" "$ASTRIA_BASE_URL/tunes/123" | jq '.[].id, .[].title, .[].name', .[].orig_images'

Create tune / reference

POST $ASTRIA_BASE_URL/tunes Content-Type: multipart/form-data

Parameters:

  • •tune[title] (required) - Display name
  • •tune[images][] - Images - multipart file upload
  • •tune[image_urls][] - Training images as URLs (alternative to file upload)
  • •tune[name] - Subject class name (e.g., man, woman, dog, style, shirt, pants, shoes, sandals, etc...)
  • •tune[base_tune_id] - $GEMINI_TUNE_ID
  • •tune[model_type] - "faceid"

Example - create tune with image URLs: curl -s -X POST "$ASTRIA_BASE_URL/tunes"
-H "Authorization: Bearer $ASTRIA_AUTH_TOKEN"
-H "X-Workspace-Id: $WORKSPACE_ID"
-F "tune[title]=Brown dress"
-F "tune[base_tune_id]=$GEMINI_TUNE_ID"
-F "tune[name]=dress"
-F "tune[characteristics]={"short_description":"satin brown dress"}"
-F "tune[image_urls][]=https://example.com/photo1.jpg"
-F "tune[image_urls][]=https://example.com/photo2.jpg" | jq '.[].id, .[].title, .[].name', .[].orig_images'

Update tune

PATCH $ASTRIA_BASE_URL/tunes/:id Parameters: tune[title], tune[name], tune[auto_extend], tune[characteristics]

Delete tune

DELETE $ASTRIA_BASE_URL/tunes/:id

Tune JSON response fields:

{ "id": 123, "title": "My Model", "name": "man", "token": "ohwx", "branch": "flux1", "model_type": "lora", "is_api": true, "steps": 100, "face_crop": false, "base_tune_id": null, "trained_at": "2024-01-01T00:00:00Z", "created_at": "2024-01-01T00:00:00Z", "cost_mc": 100, "url": "https://api.astria.ai/tunes/123.json", "orig_images": ["https://...", "https://..."] }

Prompts (Image generation)

Create prompt (generate images)

POST $ASTRIA_BASE_URL/tunes/:tune_id/prompts

When creating a prompt - make sure user is in the /prompts page or output [ASTRIA_NAV:/prompts] first to navigate the browser there first. No need to check prompt status as images will appear automatically in the UI when ready. Before creating a prompt, make sure to suggest the prompt to the user using [ASTRIA_PROMPT:...] - this will allow the user to edit the prompt in the prompt box. Explain to the user that he can click the COPY to edit the prompt before generating. Alternatively offer to create the prompt for him with different reasonable parameters of num_images, aspect_ratio, and resolution (for Gemini) based on his needs.

Parameters:

  • •prompt[text] (required) - Prompt text. Use <1> as subject token for fine-tuned models.
  • •prompt[num_images] - Number of images (1-8)
  • •prompt[aspect_ratio] - Aspect ratio string: "1:1","16:9,"9:16,"21:9,"9:21,"3:2","2:3","5:4","4:5","4:3"
  • •prompt[resolution] - "1K", "2K" (default), "4K" - only for @GEMINI_TUNE_ID

Example: curl -s -X POST "$ASTRIA_BASE_URL/tunes/@GEMINI_TUNE_ID/prompts"
-H "Authorization: Bearer $ASTRIA_AUTH_TOKEN"
-H "X-Workspace-Id: $WORKSPACE_ID"
-F "prompt[text]=180cm extremely thin and tall model faceid:4076525:1.0 woman \n Sleek, low, slicked-back bun hairdo\n faceid:4035726:1.0 earrings \n faceid:4076806:1.0 outfit woman pants longer til the end of t\n faceid:4043875:1.0 shoes \n front faceid:4042961:1.0 pose \n look book plain white background #fff"
-F "prompt[resolution]=2K"
-F "prompt[aspect_ratio]=3:4"
-F "prompt[num_images]=4" | jq '.[].id, .[].text, .[].num_images, .[].images[], .[].aspect_ratio, .[].resolution'

List prompts

GET $ASTRIA_BASE_URL/prompts Filters: text, base_pack_id, pack_id, user_id

List prompts for a tune

Can be used to find prompts tied to a specific reference just as a woman or clothing and get a sense of what the user has already generated with that reference. GET $ASTRIA_BASE_URL/tunes/:tune_id/prompts

Delete prompt

DELETE $ASTRIA_BASE_URL/tunes/:tune_id/prompts/:id

Prompt JSON response fields:

{ "id": 456, "tune_id": 123, "text": "portrait of <1> in a suit", "num_images": 4, "trained_at": "2024-01-01T00:00:00Z", "created_at": "2024-01-01T00:00:00Z", "images": ["https://cdn.astria.ai/...", "https://cdn.astria.ai/..."], "cost_mc": 50, "user_error": null, "url": "https://api.astria.ai/tunes/123/prompts/456.json" }

Seedream 4.5

POST $ASTRIA_BASE_URL/tunes/$SEEDREAM_TUNE_ID/prompts

Parameters:

  • •prompt[text] (required) - Describe the image. Do NOT use <1> token.
  • •prompt[num_images] - 1 to 4

Example: curl -s -X POST "$ASTRIA_BASE_URL/tunes/$SEEDREAM_TUNE_ID/prompts"
-H "Authorization: Bearer $ASTRIA_AUTH_TOKEN"
-H "X-Workspace-Id: $WORKSPACE_ID"
-F "prompt[text]=professional product photo of wireless headphones on marble surface"
-F "prompt[num_images]=2" | jq '.[].id, .[].text, .[].num_images, .[].images[]'

Packs (Prompt template collections)

List packs

GET $ASTRIA_BASE_URL/packs

Example: curl -s $ASTRIA_BASE_URL/packs
-H "Authorization: Bearer $ASTRIA_AUTH_TOKEN"
-H "X-Workspace-Id: $WORKSPACE_ID" | jq '.[].id, .[].title, .[].slug'

Error Handling

  • •422: Validation error (missing required fields, invalid params) - always present the error message from the "error" field in the JSON response to the user