Atlas Cloud — Model Reference

CRITICAL: Fetch From the API — Do Not Fabricate

The tables in this file are a snapshot, not a source of truth. Model IDs, prices, parameter names, defaults, and enums all change. Any ID, parameter, or schema detail in code or user-facing output MUST come from a live API fetch performed in the current session.

Mandatory workflow — no exceptions:

1. Always fetch the model list first. No authentication required: GET https://api.atlascloud.ai/api/v1/models If the MCP server is installed, call atlas_list_models or atlas_search_docs instead — same live data.

2. Filter to display_console: true. Any model with display_console: false is internal and will fail for regular users. Do not surface those IDs.

3. Before writing a request body, fetch the schema. Never guess parameter names, enums, defaults, or required fields. For any target model: - MCP: atlas_get_model_info with the exact model ID - HTTP: GET the schema URL from the model entry (OpenAPI JSON) and read components.schemas.Input.properties

Build the payload strictly from the fields listed there. If a parameter you want isn't in the schema, it doesn't exist on that model.

4. Hardcoded IDs are only acceptable after verification. Either fetch and filter at runtime, or confirm the ID against a live GET /models response in the same session before embedding it.

Red flags — stop and fetch if you catch yourself doing any of these: - Copying a model ID from memory or from these tables without a same-session API check - Guessing a parameter name (e.g. assuming aspect_ratio when the model uses ratio, or image_url when it uses image) - Sending a parameter not listed in the schema - Reporting a price to the user that wasn't in the live API response

Image Models (priced per image)

Video Models (priced per second of output; figures are the 480p entry price — 720p/1080p cost more)

LLM Models (priced per million tokens)

Model Type → Endpoint Mapping

Type notes: 3D models (image/text-to-3D) are "Image" type; TTS, music, and speech-to-text (ASR) models are all "Audio" type; lipsync / talking-avatar models are "Video" type. See references/audio-gen.md for audio patterns.

Price Structure

The price field in the API response has this structure:

• Image/Video models: Use price.actual.base_price — this is the cost per generation
• LLM models: Use price.actual.input_price and price.actual.output_price — cost per million tokens
• Fallback chain: price.actual → price.sku.text → top-level inputPrice/basePrice
• price.discount: Discount percentage (e.g., "70" means 70% of original price)

Model Schema

Each model has a schema field pointing to an OpenAPI JSON file that describes all available parameters. Fetch it to understand what a specific model accepts:

import requests

# Get public model list
models = requests.get("https://api.atlascloud.ai/api/v1/models").json()["data"]
public_models = [m for m in models if m.get("display_console") == True]

# Find a specific model
model = next(m for m in public_models if m["model"] == "bytedance/seedream-v5.0-lite")

# Fetch its parameter schema
if model.get("schema"):
    schema = requests.get(model["schema"]).json()
    # schema["components"]["schemas"]["Input"]["properties"] contains all parameters