MiniMax M2.7 Highspeed

import { createClient } from '@runware/sdk'

const client = await createClient({ apiKey: process.env.RUNWARE_API_KEY })
await client.connect()

const [result] = await client.run({
  model: 'minimax:m2.7@highspeed',
  messages: [
    {
      role: 'user',
      content: 'Turn our accessibility audit into an actionable remediation backlog for the checkout team. The audited journey is cart → shipping address → delivery method → payment → order confirmation.\n\nAudit findings:\n1. Cart quantity controls are icon-only buttons without accessible names.\n2. Removing an item updates the cart visually, but the change is not announced to screen readers.\n3. The shipping form uses placeholder text as the only visible label for apartment number and phone number.\n4. Validation errors appear only after submission, use color alone, and are not associated with their fields.\n5. After failed submission, keyboard focus remains on the Place order button instead of moving to the error summary or first invalid field.\n6. Delivery options are styled cards implemented as div elements with click handlers; they cannot be selected by keyboard.\n7. The selected delivery option is conveyed visually through a border but not programmatically.\n8. Card-number formatting inserts spaces while typing and causes the VoiceOver cursor to jump unpredictably.\n9. The payment iframe has the generic title \'secure field.\'\n10. The order total changes when delivery is selected, but the updated amount is not announced.\n11. Focus indicators on links and buttons fail contrast requirements against the dark green checkout background.\n12. The express-payment logos have redundant alt text that is read alongside visible provider names.\n13. The checkout progress indicator is read as an unlabeled list of numbers.\n14. The confirmation page uses an h1 followed by h4 headings.\n15. At 200% zoom, the order summary overlaps the Place order button at a 1280px viewport.\n16. The session-expiry warning cannot be dismissed with Escape, does not trap focus, and gives no advance warning to assistive technology.\n\nPlease prioritize by customer harm and release risk rather than simply listing findings in audit order.'
    },
    {
      role: 'assistant',
      content: 'I can structure this as an implementation backlog, but I need the frontend stack, supported assistive-technology baseline, release window, and whether the payment fields are owned by your team or a third-party provider. Should related findings be consolidated into larger tickets or kept as one ticket per defect?'
    },
    {
      role: 'user',
      content: 'Stack: React 18, TypeScript, Next.js 14, CSS Modules, React Hook Form, and Playwright. We use Storybook with axe-core checks in CI. Checkout state is managed with Zustand. Payment fields are hosted iframes from Adyen; we can configure field titles and event handling but cannot modify iframe internals.\n\nSupport baseline: current Chrome with NVDA, current Safari with VoiceOver on macOS and iOS, keyboard-only use, 200% browser zoom, and Windows High Contrast Mode. Target WCAG 2.2 AA.\n\nRelease window: three two-week sprints. Sprint 1 must address purchase-blocking and legally significant failures. Keep tickets independently deployable, but consolidate defects when they share one component and test surface. Assume one frontend squad with four engineers and one QA specialist.\n\nProduce the final backlog in Markdown. Start with a short prioritization rationale and a dependency map. Then provide 8–12 ordered tickets. For each ticket include: concise Jira-style title, priority, affected journey step, mapped audit findings, relevant WCAG 2.2 success criteria, user impact, implementation approach specific to our stack, testable acceptance criteria, automated test coverage, manual QA matrix, dependencies, rollout or regression risk, and a story-point estimate using 1/2/3/5/8. Distinguish changes we own from Adyen configuration or vendor escalation. End with a proposed allocation across the three sprints and a definition of done. Do not claim that axe-core or automated tests can prove full WCAG conformance.'
    }
  ]
})

import asyncio
import os

from runware import Runware


async def main():
    async with Runware(api_key=os.environ["RUNWARE_API_KEY"]) as client:
        results = await client.run({
            "model": "minimax:m2.7@highspeed",
            "messages": [
                {
                    "role": "user",
                    "content": "Turn our accessibility audit into an actionable remediation backlog for the checkout team. The audited journey is cart → shipping address → delivery method → payment → order confirmation.\n\nAudit findings:\n1. Cart quantity controls are icon-only buttons without accessible names.\n2. Removing an item updates the cart visually, but the change is not announced to screen readers.\n3. The shipping form uses placeholder text as the only visible label for apartment number and phone number.\n4. Validation errors appear only after submission, use color alone, and are not associated with their fields.\n5. After failed submission, keyboard focus remains on the Place order button instead of moving to the error summary or first invalid field.\n6. Delivery options are styled cards implemented as div elements with click handlers; they cannot be selected by keyboard.\n7. The selected delivery option is conveyed visually through a border but not programmatically.\n8. Card-number formatting inserts spaces while typing and causes the VoiceOver cursor to jump unpredictably.\n9. The payment iframe has the generic title 'secure field.'\n10. The order total changes when delivery is selected, but the updated amount is not announced.\n11. Focus indicators on links and buttons fail contrast requirements against the dark green checkout background.\n12. The express-payment logos have redundant alt text that is read alongside visible provider names.\n13. The checkout progress indicator is read as an unlabeled list of numbers.\n14. The confirmation page uses an h1 followed by h4 headings.\n15. At 200% zoom, the order summary overlaps the Place order button at a 1280px viewport.\n16. The session-expiry warning cannot be dismissed with Escape, does not trap focus, and gives no advance warning to assistive technology.\n\nPlease prioritize by customer harm and release risk rather than simply listing findings in audit order."
                },
                {
                    "role": "assistant",
                    "content": "I can structure this as an implementation backlog, but I need the frontend stack, supported assistive-technology baseline, release window, and whether the payment fields are owned by your team or a third-party provider. Should related findings be consolidated into larger tickets or kept as one ticket per defect?"
                },
                {
                    "role": "user",
                    "content": "Stack: React 18, TypeScript, Next.js 14, CSS Modules, React Hook Form, and Playwright. We use Storybook with axe-core checks in CI. Checkout state is managed with Zustand. Payment fields are hosted iframes from Adyen; we can configure field titles and event handling but cannot modify iframe internals.\n\nSupport baseline: current Chrome with NVDA, current Safari with VoiceOver on macOS and iOS, keyboard-only use, 200% browser zoom, and Windows High Contrast Mode. Target WCAG 2.2 AA.\n\nRelease window: three two-week sprints. Sprint 1 must address purchase-blocking and legally significant failures. Keep tickets independently deployable, but consolidate defects when they share one component and test surface. Assume one frontend squad with four engineers and one QA specialist.\n\nProduce the final backlog in Markdown. Start with a short prioritization rationale and a dependency map. Then provide 8–12 ordered tickets. For each ticket include: concise Jira-style title, priority, affected journey step, mapped audit findings, relevant WCAG 2.2 success criteria, user impact, implementation approach specific to our stack, testable acceptance criteria, automated test coverage, manual QA matrix, dependencies, rollout or regression risk, and a story-point estimate using 1/2/3/5/8. Distinguish changes we own from Adyen configuration or vendor escalation. End with a proposed allocation across the three sprints and a definition of done. Do not claim that axe-core or automated tests can prove full WCAG conformance."
                }
            ]
        })


asyncio.run(main())

curl https://api.runware.ai/v1 \
  -H "Authorization: Bearer $RUNWARE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '[
    {
      "taskType": "textInference",
      "taskUUID": "3c402cb4-5b46-44f1-a9d1-80f92ed601cc",
      "model": "minimax:m2.7@highspeed",
      "messages": [
        {
          "role": "user",
          "content": "Turn our accessibility audit into an actionable remediation backlog for the checkout team. The audited journey is cart → shipping address → delivery method → payment → order confirmation.\n\nAudit findings:\n1. Cart quantity controls are icon-only buttons without accessible names.\n2. Removing an item updates the cart visually, but the change is not announced to screen readers.\n3. The shipping form uses placeholder text as the only visible label for apartment number and phone number.\n4. Validation errors appear only after submission, use color alone, and are not associated with their fields.\n5. After failed submission, keyboard focus remains on the Place order button instead of moving to the error summary or first invalid field.\n6. Delivery options are styled cards implemented as div elements with click handlers; they cannot be selected by keyboard.\n7. The selected delivery option is conveyed visually through a border but not programmatically.\n8. Card-number formatting inserts spaces while typing and causes the VoiceOver cursor to jump unpredictably.\n9. The payment iframe has the generic title 'secure field.'\n10. The order total changes when delivery is selected, but the updated amount is not announced.\n11. Focus indicators on links and buttons fail contrast requirements against the dark green checkout background.\n12. The express-payment logos have redundant alt text that is read alongside visible provider names.\n13. The checkout progress indicator is read as an unlabeled list of numbers.\n14. The confirmation page uses an h1 followed by h4 headings.\n15. At 200% zoom, the order summary overlaps the Place order button at a 1280px viewport.\n16. The session-expiry warning cannot be dismissed with Escape, does not trap focus, and gives no advance warning to assistive technology.\n\nPlease prioritize by customer harm and release risk rather than simply listing findings in audit order."
        },
        {
          "role": "assistant",
          "content": "I can structure this as an implementation backlog, but I need the frontend stack, supported assistive-technology baseline, release window, and whether the payment fields are owned by your team or a third-party provider. Should related findings be consolidated into larger tickets or kept as one ticket per defect?"
        },
        {
          "role": "user",
          "content": "Stack: React 18, TypeScript, Next.js 14, CSS Modules, React Hook Form, and Playwright. We use Storybook with axe-core checks in CI. Checkout state is managed with Zustand. Payment fields are hosted iframes from Adyen; we can configure field titles and event handling but cannot modify iframe internals.\n\nSupport baseline: current Chrome with NVDA, current Safari with VoiceOver on macOS and iOS, keyboard-only use, 200% browser zoom, and Windows High Contrast Mode. Target WCAG 2.2 AA.\n\nRelease window: three two-week sprints. Sprint 1 must address purchase-blocking and legally significant failures. Keep tickets independently deployable, but consolidate defects when they share one component and test surface. Assume one frontend squad with four engineers and one QA specialist.\n\nProduce the final backlog in Markdown. Start with a short prioritization rationale and a dependency map. Then provide 8–12 ordered tickets. For each ticket include: concise Jira-style title, priority, affected journey step, mapped audit findings, relevant WCAG 2.2 success criteria, user impact, implementation approach specific to our stack, testable acceptance criteria, automated test coverage, manual QA matrix, dependencies, rollout or regression risk, and a story-point estimate using 1/2/3/5/8. Distinguish changes we own from Adyen configuration or vendor escalation. End with a proposed allocation across the three sprints and a definition of done. Do not claim that axe-core or automated tests can prove full WCAG conformance."
        }
      ]
    }
  ]'

runware run minimax:m2.7@highspeed \
  messages.0.role=user \
  messages.0.content="Turn our accessibility audit into an actionable remediation backlog for the checkout team. The audited journey is cart → shipping address → delivery method → payment → order confirmation.

Audit findings:
1. Cart quantity controls are icon-only buttons without accessible names.
2. Removing an item updates the cart visually, but the change is not announced to screen readers.
3. The shipping form uses placeholder text as the only visible label for apartment number and phone number.
4. Validation errors appear only after submission, use color alone, and are not associated with their fields.
5. After failed submission, keyboard focus remains on the Place order button instead of moving to the error summary or first invalid field.
6. Delivery options are styled cards implemented as div elements with click handlers; they cannot be selected by keyboard.
7. The selected delivery option is conveyed visually through a border but not programmatically.
8. Card-number formatting inserts spaces while typing and causes the VoiceOver cursor to jump unpredictably.
9. The payment iframe has the generic title 'secure field.'
10. The order total changes when delivery is selected, but the updated amount is not announced.
11. Focus indicators on links and buttons fail contrast requirements against the dark green checkout background.
12. The express-payment logos have redundant alt text that is read alongside visible provider names.
13. The checkout progress indicator is read as an unlabeled list of numbers.
14. The confirmation page uses an h1 followed by h4 headings.
15. At 200% zoom, the order summary overlaps the Place order button at a 1280px viewport.
16. The session-expiry warning cannot be dismissed with Escape, does not trap focus, and gives no advance warning to assistive technology.

Please prioritize by customer harm and release risk rather than simply listing findings in audit order." \
  messages.1.role=assistant \
  messages.1.content="I can structure this as an implementation backlog, but I need the frontend stack, supported assistive-technology baseline, release window, and whether the payment fields are owned by your team or a third-party provider. Should related findings be consolidated into larger tickets or kept as one ticket per defect?" \
  messages.2.role=user \
  messages.2.content="Stack: React 18, TypeScript, Next.js 14, CSS Modules, React Hook Form, and Playwright. We use Storybook with axe-core checks in CI. Checkout state is managed with Zustand. Payment fields are hosted iframes from Adyen; we can configure field titles and event handling but cannot modify iframe internals.

Support baseline: current Chrome with NVDA, current Safari with VoiceOver on macOS and iOS, keyboard-only use, 200% browser zoom, and Windows High Contrast Mode. Target WCAG 2.2 AA.

Release window: three two-week sprints. Sprint 1 must address purchase-blocking and legally significant failures. Keep tickets independently deployable, but consolidate defects when they share one component and test surface. Assume one frontend squad with four engineers and one QA specialist.

Produce the final backlog in Markdown. Start with a short prioritization rationale and a dependency map. Then provide 8–12 ordered tickets. For each ticket include: concise Jira-style title, priority, affected journey step, mapped audit findings, relevant WCAG 2.2 success criteria, user impact, implementation approach specific to our stack, testable acceptance criteria, automated test coverage, manual QA matrix, dependencies, rollout or regression risk, and a story-point estimate using 1/2/3/5/8. Distinguish changes we own from Adyen configuration or vendor escalation. End with a proposed allocation across the three sprints and a definition of done. Do not claim that axe-core or automated tests can prove full WCAG conformance."

{
  "taskType": "textInference",
  "taskUUID": "3c402cb4-5b46-44f1-a9d1-80f92ed601cc",
  "model": "minimax:m2.7@highspeed",
  "messages": [
    {
      "role": "user",
      "content": "Turn our accessibility audit into an actionable remediation backlog for the checkout team. The audited journey is cart → shipping address → delivery method → payment → order confirmation.\n\nAudit findings:\n1. Cart quantity controls are icon-only buttons without accessible names.\n2. Removing an item updates the cart visually, but the change is not announced to screen readers.\n3. The shipping form uses placeholder text as the only visible label for apartment number and phone number.\n4. Validation errors appear only after submission, use color alone, and are not associated with their fields.\n5. After failed submission, keyboard focus remains on the Place order button instead of moving to the error summary or first invalid field.\n6. Delivery options are styled cards implemented as div elements with click handlers; they cannot be selected by keyboard.\n7. The selected delivery option is conveyed visually through a border but not programmatically.\n8. Card-number formatting inserts spaces while typing and causes the VoiceOver cursor to jump unpredictably.\n9. The payment iframe has the generic title 'secure field.'\n10. The order total changes when delivery is selected, but the updated amount is not announced.\n11. Focus indicators on links and buttons fail contrast requirements against the dark green checkout background.\n12. The express-payment logos have redundant alt text that is read alongside visible provider names.\n13. The checkout progress indicator is read as an unlabeled list of numbers.\n14. The confirmation page uses an h1 followed by h4 headings.\n15. At 200% zoom, the order summary overlaps the Place order button at a 1280px viewport.\n16. The session-expiry warning cannot be dismissed with Escape, does not trap focus, and gives no advance warning to assistive technology.\n\nPlease prioritize by customer harm and release risk rather than simply listing findings in audit order."
    },
    {
      "role": "assistant",
      "content": "I can structure this as an implementation backlog, but I need the frontend stack, supported assistive-technology baseline, release window, and whether the payment fields are owned by your team or a third-party provider. Should related findings be consolidated into larger tickets or kept as one ticket per defect?"
    },
    {
      "role": "user",
      "content": "Stack: React 18, TypeScript, Next.js 14, CSS Modules, React Hook Form, and Playwright. We use Storybook with axe-core checks in CI. Checkout state is managed with Zustand. Payment fields are hosted iframes from Adyen; we can configure field titles and event handling but cannot modify iframe internals.\n\nSupport baseline: current Chrome with NVDA, current Safari with VoiceOver on macOS and iOS, keyboard-only use, 200% browser zoom, and Windows High Contrast Mode. Target WCAG 2.2 AA.\n\nRelease window: three two-week sprints. Sprint 1 must address purchase-blocking and legally significant failures. Keep tickets independently deployable, but consolidate defects when they share one component and test surface. Assume one frontend squad with four engineers and one QA specialist.\n\nProduce the final backlog in Markdown. Start with a short prioritization rationale and a dependency map. Then provide 8–12 ordered tickets. For each ticket include: concise Jira-style title, priority, affected journey step, mapped audit findings, relevant WCAG 2.2 success criteria, user impact, implementation approach specific to our stack, testable acceptance criteria, automated test coverage, manual QA matrix, dependencies, rollout or regression risk, and a story-point estimate using 1/2/3/5/8. Distinguish changes we own from Adyen configuration or vendor escalation. End with a proposed allocation across the three sprints and a definition of done. Do not claim that axe-core or automated tests can prove full WCAG conformance."
    }
  ]
}

import { createClient } from '@runware/sdk'

const client = await createClient({ apiKey: process.env.RUNWARE_API_KEY })
await client.connect()

const [result] = await client.run({
  model: 'minimax:m2.7@highspeed',
  settings: {
    systemPrompt: 'You are a support communications assistant for a B2B document-processing platform. Never invent live operational information. When a request depends on current platform status, call the relevant function and wait for the developer to execute it and return its result in a follow-up conversation turn. After receiving that result, write concise, factual customer-facing copy without exposing internal function names.',
    maxTokens: 32768
  },
  messages: [
    {
      role: 'user',
      content: 'A customer reports that three scheduled compliance-report exports have remained in the processing state for 20 minutes. We need a short reply suitable for their support ticket.'
    },
    {
      role: 'assistant',
      content: 'I can prepare the reply, but I should verify the live status of the export service before making any operational claims.'
    },
    {
      role: 'user',
      content: 'Check the live status of the file_exports component now. Make exactly one function call: use the platform-status function, not the UTC-time function. Do not draft the reply until the developer executes the returned call and sends its result in a follow-up turn. Once that result arrives, write a three-sentence customer update covering current status, what they should do, and when they can expect another update.'
    }
  ],
  tools: [
    {
      type: 'function',
      name: 'get_current_utc_time',
      description: 'Returns the current date and time in UTC.',
      schema: {}
    },
    {
      type: 'function',
      name: 'get_platform_status',
      description: 'Returns the current operational status of the platform.',
      schema: {}
    }
  ],
  toolChoice: {
    type: 'any'
  }
})

import asyncio
import os

from runware import Runware


async def main():
    async with Runware(api_key=os.environ["RUNWARE_API_KEY"]) as client:
        results = await client.run({
            "model": "minimax:m2.7@highspeed",
            "settings": {
                "systemPrompt": "You are a support communications assistant for a B2B document-processing platform. Never invent live operational information. When a request depends on current platform status, call the relevant function and wait for the developer to execute it and return its result in a follow-up conversation turn. After receiving that result, write concise, factual customer-facing copy without exposing internal function names.",
                "maxTokens": 32768
            },
            "messages": [
                {
                    "role": "user",
                    "content": "A customer reports that three scheduled compliance-report exports have remained in the processing state for 20 minutes. We need a short reply suitable for their support ticket."
                },
                {
                    "role": "assistant",
                    "content": "I can prepare the reply, but I should verify the live status of the export service before making any operational claims."
                },
                {
                    "role": "user",
                    "content": "Check the live status of the file_exports component now. Make exactly one function call: use the platform-status function, not the UTC-time function. Do not draft the reply until the developer executes the returned call and sends its result in a follow-up turn. Once that result arrives, write a three-sentence customer update covering current status, what they should do, and when they can expect another update."
                }
            ],
            "tools": [
                {
                    "type": "function",
                    "name": "get_current_utc_time",
                    "description": "Returns the current date and time in UTC.",
                    "schema": {}
                },
                {
                    "type": "function",
                    "name": "get_platform_status",
                    "description": "Returns the current operational status of the platform.",
                    "schema": {}
                }
            ],
            "toolChoice": {
                "type": "any"
            }
        })


asyncio.run(main())

curl https://api.runware.ai/v1 \
  -H "Authorization: Bearer $RUNWARE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '[
    {
      "taskType": "textInference",
      "taskUUID": "2c9f1a27-084c-4b14-8ed0-987565fd9db9",
      "model": "minimax:m2.7@highspeed",
      "settings": {
        "systemPrompt": "You are a support communications assistant for a B2B document-processing platform. Never invent live operational information. When a request depends on current platform status, call the relevant function and wait for the developer to execute it and return its result in a follow-up conversation turn. After receiving that result, write concise, factual customer-facing copy without exposing internal function names.",
        "maxTokens": 32768
      },
      "messages": [
        {
          "role": "user",
          "content": "A customer reports that three scheduled compliance-report exports have remained in the processing state for 20 minutes. We need a short reply suitable for their support ticket."
        },
        {
          "role": "assistant",
          "content": "I can prepare the reply, but I should verify the live status of the export service before making any operational claims."
        },
        {
          "role": "user",
          "content": "Check the live status of the file_exports component now. Make exactly one function call: use the platform-status function, not the UTC-time function. Do not draft the reply until the developer executes the returned call and sends its result in a follow-up turn. Once that result arrives, write a three-sentence customer update covering current status, what they should do, and when they can expect another update."
        }
      ],
      "tools": [
        {
          "type": "function",
          "name": "get_current_utc_time",
          "description": "Returns the current date and time in UTC.",
          "schema": {}
        },
        {
          "type": "function",
          "name": "get_platform_status",
          "description": "Returns the current operational status of the platform.",
          "schema": {}
        }
      ],
      "toolChoice": {
        "type": "any"
      }
    }
  ]'

runware run minimax:m2.7@highspeed \
  settings.systemPrompt="You are a support communications assistant for a B2B document-processing platform. Never invent live operational information. When a request depends on current platform status, call the relevant function and wait for the developer to execute it and return its result in a follow-up conversation turn. After receiving that result, write concise, factual customer-facing copy without exposing internal function names." \
  settings.maxTokens=32768 \
  messages.0.role=user \
  messages.0.content="A customer reports that three scheduled compliance-report exports have remained in the processing state for 20 minutes. We need a short reply suitable for their support ticket." \
  messages.1.role=assistant \
  messages.1.content="I can prepare the reply, but I should verify the live status of the export service before making any operational claims." \
  messages.2.role=user \
  messages.2.content="Check the live status of the file_exports component now. Make exactly one function call: use the platform-status function, not the UTC-time function. Do not draft the reply until the developer executes the returned call and sends its result in a follow-up turn. Once that result arrives, write a three-sentence customer update covering current status, what they should do, and when they can expect another update." \
  tools.0.type=function \
  tools.0.name=get_current_utc_time \
  tools.0.description="Returns the current date and time in UTC." \
  tools.1.type=function \
  tools.1.name=get_platform_status \
  tools.1.description="Returns the current operational status of the platform." \
  toolChoice.type=any

{
  "taskType": "textInference",
  "taskUUID": "2c9f1a27-084c-4b14-8ed0-987565fd9db9",
  "model": "minimax:m2.7@highspeed",
  "settings": {
    "systemPrompt": "You are a support communications assistant for a B2B document-processing platform. Never invent live operational information. When a request depends on current platform status, call the relevant function and wait for the developer to execute it and return its result in a follow-up conversation turn. After receiving that result, write concise, factual customer-facing copy without exposing internal function names.",
    "maxTokens": 32768
  },
  "messages": [
    {
      "role": "user",
      "content": "A customer reports that three scheduled compliance-report exports have remained in the processing state for 20 minutes. We need a short reply suitable for their support ticket."
    },
    {
      "role": "assistant",
      "content": "I can prepare the reply, but I should verify the live status of the export service before making any operational claims."
    },
    {
      "role": "user",
      "content": "Check the live status of the file_exports component now. Make exactly one function call: use the platform-status function, not the UTC-time function. Do not draft the reply until the developer executes the returned call and sends its result in a follow-up turn. Once that result arrives, write a three-sentence customer update covering current status, what they should do, and when they can expect another update."
    }
  ],
  "tools": [
    {
      "type": "function",
      "name": "get_current_utc_time",
      "description": "Returns the current date and time in UTC.",
      "schema": {}
    },
    {
      "type": "function",
      "name": "get_platform_status",
      "description": "Returns the current operational status of the platform.",
      "schema": {}
    }
  ],
  "toolChoice": {
    "type": "any"
  }
}

import { createClient } from '@runware/sdk'

const client = await createClient({ apiKey: process.env.RUNWARE_API_KEY })
await client.connect()

const [result] = await client.run({
  model: 'minimax:m2.7@highspeed',
  settings: {
    systemPrompt: 'You are the launch-control assistant for an online ticketing platform. Live operational facts must come from the available functions, never from assumptions. When the user\'s request requires live data, select exactly the one function that directly answers it, emit the function call, and stop. The developer will inspect the returned call, execute that function, and send its output in a new user message beginning with TOOL_RESULT. Only after receiving TOOL_RESULT should you provide a concise go/no-go recommendation grounded in that result.',
    maxTokens: 32768
  },
  messages: [
    {
      role: 'user',
      content: 'We are preparing tonight\'s public ticket release for the 42,000-seat Northbank Stadium concert. What live checks can you perform before we open sales?'
    },
    {
      role: 'assistant',
      content: 'I can retrieve the current UTC time or check the ticketing platform\'s live operational status. For a launch decision, tell me which single check you want performed.'
    },
    {
      role: 'user',
      content: 'Check the live platform status now for the release go/no-go decision. Do not retrieve the current UTC time. After the developer sends back the function result, summarize any affected services and recommend GO, HOLD, or ABORT.'
    }
  ],
  tools: [
    {
      type: 'function',
      name: 'get_current_utc_time',
      description: 'Returns the current date and time in UTC.',
      schema: {}
    },
    {
      type: 'function',
      name: 'get_platform_status',
      description: 'Returns the current operational status of the platform.',
      schema: {}
    }
  ],
  toolChoice: {
    type: 'any'
  }
})

import asyncio
import os

from runware import Runware


async def main():
    async with Runware(api_key=os.environ["RUNWARE_API_KEY"]) as client:
        results = await client.run({
            "model": "minimax:m2.7@highspeed",
            "settings": {
                "systemPrompt": "You are the launch-control assistant for an online ticketing platform. Live operational facts must come from the available functions, never from assumptions. When the user's request requires live data, select exactly the one function that directly answers it, emit the function call, and stop. The developer will inspect the returned call, execute that function, and send its output in a new user message beginning with TOOL_RESULT. Only after receiving TOOL_RESULT should you provide a concise go/no-go recommendation grounded in that result.",
                "maxTokens": 32768
            },
            "messages": [
                {
                    "role": "user",
                    "content": "We are preparing tonight's public ticket release for the 42,000-seat Northbank Stadium concert. What live checks can you perform before we open sales?"
                },
                {
                    "role": "assistant",
                    "content": "I can retrieve the current UTC time or check the ticketing platform's live operational status. For a launch decision, tell me which single check you want performed."
                },
                {
                    "role": "user",
                    "content": "Check the live platform status now for the release go/no-go decision. Do not retrieve the current UTC time. After the developer sends back the function result, summarize any affected services and recommend GO, HOLD, or ABORT."
                }
            ],
            "tools": [
                {
                    "type": "function",
                    "name": "get_current_utc_time",
                    "description": "Returns the current date and time in UTC.",
                    "schema": {}
                },
                {
                    "type": "function",
                    "name": "get_platform_status",
                    "description": "Returns the current operational status of the platform.",
                    "schema": {}
                }
            ],
            "toolChoice": {
                "type": "any"
            }
        })


asyncio.run(main())

curl https://api.runware.ai/v1 \
  -H "Authorization: Bearer $RUNWARE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '[
    {
      "taskType": "textInference",
      "taskUUID": "f40b6175-59b3-46d8-a2b5-9cee304b7178",
      "model": "minimax:m2.7@highspeed",
      "settings": {
        "systemPrompt": "You are the launch-control assistant for an online ticketing platform. Live operational facts must come from the available functions, never from assumptions. When the user's request requires live data, select exactly the one function that directly answers it, emit the function call, and stop. The developer will inspect the returned call, execute that function, and send its output in a new user message beginning with TOOL_RESULT. Only after receiving TOOL_RESULT should you provide a concise go/no-go recommendation grounded in that result.",
        "maxTokens": 32768
      },
      "messages": [
        {
          "role": "user",
          "content": "We are preparing tonight's public ticket release for the 42,000-seat Northbank Stadium concert. What live checks can you perform before we open sales?"
        },
        {
          "role": "assistant",
          "content": "I can retrieve the current UTC time or check the ticketing platform's live operational status. For a launch decision, tell me which single check you want performed."
        },
        {
          "role": "user",
          "content": "Check the live platform status now for the release go/no-go decision. Do not retrieve the current UTC time. After the developer sends back the function result, summarize any affected services and recommend GO, HOLD, or ABORT."
        }
      ],
      "tools": [
        {
          "type": "function",
          "name": "get_current_utc_time",
          "description": "Returns the current date and time in UTC.",
          "schema": {}
        },
        {
          "type": "function",
          "name": "get_platform_status",
          "description": "Returns the current operational status of the platform.",
          "schema": {}
        }
      ],
      "toolChoice": {
        "type": "any"
      }
    }
  ]'

runware run minimax:m2.7@highspeed \
  settings.systemPrompt="You are the launch-control assistant for an online ticketing platform. Live operational facts must come from the available functions, never from assumptions. When the user's request requires live data, select exactly the one function that directly answers it, emit the function call, and stop. The developer will inspect the returned call, execute that function, and send its output in a new user message beginning with TOOL_RESULT. Only after receiving TOOL_RESULT should you provide a concise go/no-go recommendation grounded in that result." \
  settings.maxTokens=32768 \
  messages.0.role=user \
  messages.0.content="We are preparing tonight's public ticket release for the 42,000-seat Northbank Stadium concert. What live checks can you perform before we open sales?" \
  messages.1.role=assistant \
  messages.1.content="I can retrieve the current UTC time or check the ticketing platform's live operational status. For a launch decision, tell me which single check you want performed." \
  messages.2.role=user \
  messages.2.content="Check the live platform status now for the release go/no-go decision. Do not retrieve the current UTC time. After the developer sends back the function result, summarize any affected services and recommend GO, HOLD, or ABORT." \
  tools.0.type=function \
  tools.0.name=get_current_utc_time \
  tools.0.description="Returns the current date and time in UTC." \
  tools.1.type=function \
  tools.1.name=get_platform_status \
  tools.1.description="Returns the current operational status of the platform." \
  toolChoice.type=any

{
  "taskType": "textInference",
  "taskUUID": "f40b6175-59b3-46d8-a2b5-9cee304b7178",
  "model": "minimax:m2.7@highspeed",
  "settings": {
    "systemPrompt": "You are the launch-control assistant for an online ticketing platform. Live operational facts must come from the available functions, never from assumptions. When the user's request requires live data, select exactly the one function that directly answers it, emit the function call, and stop. The developer will inspect the returned call, execute that function, and send its output in a new user message beginning with TOOL_RESULT. Only after receiving TOOL_RESULT should you provide a concise go/no-go recommendation grounded in that result.",
    "maxTokens": 32768
  },
  "messages": [
    {
      "role": "user",
      "content": "We are preparing tonight's public ticket release for the 42,000-seat Northbank Stadium concert. What live checks can you perform before we open sales?"
    },
    {
      "role": "assistant",
      "content": "I can retrieve the current UTC time or check the ticketing platform's live operational status. For a launch decision, tell me which single check you want performed."
    },
    {
      "role": "user",
      "content": "Check the live platform status now for the release go/no-go decision. Do not retrieve the current UTC time. After the developer sends back the function result, summarize any affected services and recommend GO, HOLD, or ABORT."
    }
  ],
  "tools": [
    {
      "type": "function",
      "name": "get_current_utc_time",
      "description": "Returns the current date and time in UTC.",
      "schema": {}
    },
    {
      "type": "function",
      "name": "get_platform_status",
      "description": "Returns the current operational status of the platform.",
      "schema": {}
    }
  ],
  "toolChoice": {
    "type": "any"
  }
}

import { createClient } from '@runware/sdk'

const client = await createClient({ apiKey: process.env.RUNWARE_API_KEY })
await client.connect()

const [result] = await client.run({
  model: 'minimax:m2.7@highspeed',
  messages: [
    {
      role: 'user',
      content: 'Write the complete audio guide for “After the Last Train,” a museum exhibition about photographer Elena Varga’s nighttime documentation of regional rail stations. I’ll provide the approved curatorial facts. The script will be recorded by a professional narrator, so it must be polished and ready for the studio.'
    },
    {
      role: 'assistant',
      content: 'Please send the approved facts, required stops, target runtime, intended audience, interpretive tone, accessibility guidance, pronunciation notes, and any claims or language that must be avoided.'
    },
    {
      role: 'user',
      content: 'Use only the source packet below. Do not invent quotations, dates, locations, motives, biographical details, or photographic techniques.\n\nDELIVERABLE\nCreate a six-stop English-language audio guide totaling 1,050–1,200 words. Write for curious adult visitors without specialist art knowledge. The experience should feel intimate, observant, and quietly hopeful—not academic, sentimental, eerie, or promotional.\n\nReturn the finished script only, organized as:\n- Exhibition Introduction\n- Stop 1: Platform Clock, 23:17\n- Stop 2: Waiting Room with Fern\n- Stop 3: Signal Box Window\n- Stop 4: Last Bus Connection\n- Stop 5: First Light, Empty Track\n- Closing\n\nFor every section, include an estimated runtime in brackets beneath the heading. Write natural narration in complete paragraphs. Do not include production notes, music cues, stage directions, footnotes, bullet points, or citations in the script.\n\nSOURCE PACKET\nExhibition title: After the Last Train\nArtist: Elena Varga (pronounced EH-leh-nah VAR-gah)\nVenue: Northbank Museum of Photography\n\nThe exhibition contains 42 color photographs made from 2018 through 2023 at small, still-operating railway stations across Hungary, Slovakia, Croatia, and eastern Austria. Varga traveled only on scheduled public transportation and photographed between the final evening arrival and the first morning departure. She did not stage scenes or ask station staff to alter lighting.\n\nVarga began the series after noticing that stations change function late at night: they remain pieces of public infrastructure, but briefly become places of maintenance, waiting, handover, and rest. Her stated interest is in “the hours when a shared place belongs to very few people.” This is the only direct quotation approved for use.\n\nAcross the series, recurring details include clocks, ticket windows, potted plants, vending machines, fluorescent fixtures, handwritten notices, benches, reflective safety clothing, cleaning equipment, and tracks disappearing beyond the lit platform. People are present in 11 of the 42 photographs. No individual is identified by name.\n\nVarga used a tripod and long exposures where station rules allowed. The photographs retain mixed light sources rather than correcting them to a uniform white. As a result, fluorescent light may appear green, older lamps amber, and early daylight blue. Long exposure sometimes softens a moving figure, but the architecture and stationary objects remain precise.\n\nThe works are presented in chronological order from late evening toward dawn rather than by country. Labels omit exact station names at the artist’s request. She wanted visitors to attend first to repeated civic features rather than treat the series as a catalog of destinations.\n\nOBJECT NOTES\nIntroduction area: A wall-sized route diagram connects all photographed stations but does not associate individual pictures with named locations.\n\nStop 1, Platform Clock, 23:17, 2018: A round analog clock occupies the upper center of the frame. Below it, two empty benches face opposite directions. Rainwater reflects a strip of amber platform light. This is the earliest photograph in the series.\n\nStop 2, Waiting Room with Fern, 2019: A large fern sits between a radiator and a closed ticket window. A handwritten notice is visible but not legible. Green fluorescent light fills the room, while an adjoining corridor appears amber. No people are visible.\n\nStop 3, Signal Box Window, 2020: Photographed from a public platform toward an illuminated signal box. One railway worker is visible in profile behind glass, wearing reflective clothing and looking down at paperwork. The long exposure softens one raised hand. Avoid speculating about the worker’s identity, thoughts, emotions, or exact task.\n\nStop 4, Last Bus Connection, 2021: Three passengers stand beneath a bus shelter outside a station. One checks a paper timetable, one holds a folded bicycle, and one sits beside two shopping bags. Their faces are not clearly visible. The station clock reads 00:42. The photograph is about linked public systems and the practical choreography of waiting; do not describe the people as lonely or stranded.\n\nStop 5, First Light, Empty Track, 2023: Parallel tracks run toward a pale blue horizon. Platform lamps are still illuminated. A yellow cleaning cart stands near the frame’s edge. This is the final image in both the series and exhibition.\n\nINTERPRETIVE PRIORITIES\nHelp listeners look slowly at specific visual details before offering context. Connect the works to public infrastructure, shared responsibility, routine labor, and the transition from night to morning. Emphasize that an apparently empty station is still active within a larger system. Treat ordinary fixtures and maintenance objects with dignity without romanticizing them.\n\nACCESSIBILITY AND STYLE\nAssume some listeners have low vision. Establish the composition and spatial relationships clearly, but weave visual description into the interpretation rather than listing objects. Use short-to-medium sentences suited to speech. Avoid rhetorical questions, second-person commands such as “look” or “notice,” and phrases such as “as you can see.” Use “you” sparingly and only to orient the visitor within the exhibition. Define “long exposure” in plain language the first time it appears. Vary transitions so sections do not repeatedly begin with the artwork title or “Here.”\n\nEnd each object stop with one brief observational invitation that can also work for a listener with low vision—for example, inviting attention to contrasts, intervals, repeated sounds in the gallery, or the idea of waiting. Do not presume what a visitor can see. The Closing should connect the first clock to the final dawn image and end on a concrete, memorable sentence rather than a summary or call to action.\n\nBefore responding, silently verify the word count, factual accuracy, section order, pronunciation spelling, and all prohibited language. Output only the studio-ready script.'
    }
  ]
})

import asyncio
import os

from runware import Runware


async def main():
    async with Runware(api_key=os.environ["RUNWARE_API_KEY"]) as client:
        results = await client.run({
            "model": "minimax:m2.7@highspeed",
            "messages": [
                {
                    "role": "user",
                    "content": "Write the complete audio guide for “After the Last Train,” a museum exhibition about photographer Elena Varga’s nighttime documentation of regional rail stations. I’ll provide the approved curatorial facts. The script will be recorded by a professional narrator, so it must be polished and ready for the studio."
                },
                {
                    "role": "assistant",
                    "content": "Please send the approved facts, required stops, target runtime, intended audience, interpretive tone, accessibility guidance, pronunciation notes, and any claims or language that must be avoided."
                },
                {
                    "role": "user",
                    "content": "Use only the source packet below. Do not invent quotations, dates, locations, motives, biographical details, or photographic techniques.\n\nDELIVERABLE\nCreate a six-stop English-language audio guide totaling 1,050–1,200 words. Write for curious adult visitors without specialist art knowledge. The experience should feel intimate, observant, and quietly hopeful—not academic, sentimental, eerie, or promotional.\n\nReturn the finished script only, organized as:\n- Exhibition Introduction\n- Stop 1: Platform Clock, 23:17\n- Stop 2: Waiting Room with Fern\n- Stop 3: Signal Box Window\n- Stop 4: Last Bus Connection\n- Stop 5: First Light, Empty Track\n- Closing\n\nFor every section, include an estimated runtime in brackets beneath the heading. Write natural narration in complete paragraphs. Do not include production notes, music cues, stage directions, footnotes, bullet points, or citations in the script.\n\nSOURCE PACKET\nExhibition title: After the Last Train\nArtist: Elena Varga (pronounced EH-leh-nah VAR-gah)\nVenue: Northbank Museum of Photography\n\nThe exhibition contains 42 color photographs made from 2018 through 2023 at small, still-operating railway stations across Hungary, Slovakia, Croatia, and eastern Austria. Varga traveled only on scheduled public transportation and photographed between the final evening arrival and the first morning departure. She did not stage scenes or ask station staff to alter lighting.\n\nVarga began the series after noticing that stations change function late at night: they remain pieces of public infrastructure, but briefly become places of maintenance, waiting, handover, and rest. Her stated interest is in “the hours when a shared place belongs to very few people.” This is the only direct quotation approved for use.\n\nAcross the series, recurring details include clocks, ticket windows, potted plants, vending machines, fluorescent fixtures, handwritten notices, benches, reflective safety clothing, cleaning equipment, and tracks disappearing beyond the lit platform. People are present in 11 of the 42 photographs. No individual is identified by name.\n\nVarga used a tripod and long exposures where station rules allowed. The photographs retain mixed light sources rather than correcting them to a uniform white. As a result, fluorescent light may appear green, older lamps amber, and early daylight blue. Long exposure sometimes softens a moving figure, but the architecture and stationary objects remain precise.\n\nThe works are presented in chronological order from late evening toward dawn rather than by country. Labels omit exact station names at the artist’s request. She wanted visitors to attend first to repeated civic features rather than treat the series as a catalog of destinations.\n\nOBJECT NOTES\nIntroduction area: A wall-sized route diagram connects all photographed stations but does not associate individual pictures with named locations.\n\nStop 1, Platform Clock, 23:17, 2018: A round analog clock occupies the upper center of the frame. Below it, two empty benches face opposite directions. Rainwater reflects a strip of amber platform light. This is the earliest photograph in the series.\n\nStop 2, Waiting Room with Fern, 2019: A large fern sits between a radiator and a closed ticket window. A handwritten notice is visible but not legible. Green fluorescent light fills the room, while an adjoining corridor appears amber. No people are visible.\n\nStop 3, Signal Box Window, 2020: Photographed from a public platform toward an illuminated signal box. One railway worker is visible in profile behind glass, wearing reflective clothing and looking down at paperwork. The long exposure softens one raised hand. Avoid speculating about the worker’s identity, thoughts, emotions, or exact task.\n\nStop 4, Last Bus Connection, 2021: Three passengers stand beneath a bus shelter outside a station. One checks a paper timetable, one holds a folded bicycle, and one sits beside two shopping bags. Their faces are not clearly visible. The station clock reads 00:42. The photograph is about linked public systems and the practical choreography of waiting; do not describe the people as lonely or stranded.\n\nStop 5, First Light, Empty Track, 2023: Parallel tracks run toward a pale blue horizon. Platform lamps are still illuminated. A yellow cleaning cart stands near the frame’s edge. This is the final image in both the series and exhibition.\n\nINTERPRETIVE PRIORITIES\nHelp listeners look slowly at specific visual details before offering context. Connect the works to public infrastructure, shared responsibility, routine labor, and the transition from night to morning. Emphasize that an apparently empty station is still active within a larger system. Treat ordinary fixtures and maintenance objects with dignity without romanticizing them.\n\nACCESSIBILITY AND STYLE\nAssume some listeners have low vision. Establish the composition and spatial relationships clearly, but weave visual description into the interpretation rather than listing objects. Use short-to-medium sentences suited to speech. Avoid rhetorical questions, second-person commands such as “look” or “notice,” and phrases such as “as you can see.” Use “you” sparingly and only to orient the visitor within the exhibition. Define “long exposure” in plain language the first time it appears. Vary transitions so sections do not repeatedly begin with the artwork title or “Here.”\n\nEnd each object stop with one brief observational invitation that can also work for a listener with low vision—for example, inviting attention to contrasts, intervals, repeated sounds in the gallery, or the idea of waiting. Do not presume what a visitor can see. The Closing should connect the first clock to the final dawn image and end on a concrete, memorable sentence rather than a summary or call to action.\n\nBefore responding, silently verify the word count, factual accuracy, section order, pronunciation spelling, and all prohibited language. Output only the studio-ready script."
                }
            ]
        })


asyncio.run(main())

curl https://api.runware.ai/v1 \
  -H "Authorization: Bearer $RUNWARE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '[
    {
      "taskType": "textInference",
      "taskUUID": "6bbb3f76-580b-4094-8882-c58083da811d",
      "model": "minimax:m2.7@highspeed",
      "messages": [
        {
          "role": "user",
          "content": "Write the complete audio guide for “After the Last Train,” a museum exhibition about photographer Elena Varga’s nighttime documentation of regional rail stations. I’ll provide the approved curatorial facts. The script will be recorded by a professional narrator, so it must be polished and ready for the studio."
        },
        {
          "role": "assistant",
          "content": "Please send the approved facts, required stops, target runtime, intended audience, interpretive tone, accessibility guidance, pronunciation notes, and any claims or language that must be avoided."
        },
        {
          "role": "user",
          "content": "Use only the source packet below. Do not invent quotations, dates, locations, motives, biographical details, or photographic techniques.\n\nDELIVERABLE\nCreate a six-stop English-language audio guide totaling 1,050–1,200 words. Write for curious adult visitors without specialist art knowledge. The experience should feel intimate, observant, and quietly hopeful—not academic, sentimental, eerie, or promotional.\n\nReturn the finished script only, organized as:\n- Exhibition Introduction\n- Stop 1: Platform Clock, 23:17\n- Stop 2: Waiting Room with Fern\n- Stop 3: Signal Box Window\n- Stop 4: Last Bus Connection\n- Stop 5: First Light, Empty Track\n- Closing\n\nFor every section, include an estimated runtime in brackets beneath the heading. Write natural narration in complete paragraphs. Do not include production notes, music cues, stage directions, footnotes, bullet points, or citations in the script.\n\nSOURCE PACKET\nExhibition title: After the Last Train\nArtist: Elena Varga (pronounced EH-leh-nah VAR-gah)\nVenue: Northbank Museum of Photography\n\nThe exhibition contains 42 color photographs made from 2018 through 2023 at small, still-operating railway stations across Hungary, Slovakia, Croatia, and eastern Austria. Varga traveled only on scheduled public transportation and photographed between the final evening arrival and the first morning departure. She did not stage scenes or ask station staff to alter lighting.\n\nVarga began the series after noticing that stations change function late at night: they remain pieces of public infrastructure, but briefly become places of maintenance, waiting, handover, and rest. Her stated interest is in “the hours when a shared place belongs to very few people.” This is the only direct quotation approved for use.\n\nAcross the series, recurring details include clocks, ticket windows, potted plants, vending machines, fluorescent fixtures, handwritten notices, benches, reflective safety clothing, cleaning equipment, and tracks disappearing beyond the lit platform. People are present in 11 of the 42 photographs. No individual is identified by name.\n\nVarga used a tripod and long exposures where station rules allowed. The photographs retain mixed light sources rather than correcting them to a uniform white. As a result, fluorescent light may appear green, older lamps amber, and early daylight blue. Long exposure sometimes softens a moving figure, but the architecture and stationary objects remain precise.\n\nThe works are presented in chronological order from late evening toward dawn rather than by country. Labels omit exact station names at the artist’s request. She wanted visitors to attend first to repeated civic features rather than treat the series as a catalog of destinations.\n\nOBJECT NOTES\nIntroduction area: A wall-sized route diagram connects all photographed stations but does not associate individual pictures with named locations.\n\nStop 1, Platform Clock, 23:17, 2018: A round analog clock occupies the upper center of the frame. Below it, two empty benches face opposite directions. Rainwater reflects a strip of amber platform light. This is the earliest photograph in the series.\n\nStop 2, Waiting Room with Fern, 2019: A large fern sits between a radiator and a closed ticket window. A handwritten notice is visible but not legible. Green fluorescent light fills the room, while an adjoining corridor appears amber. No people are visible.\n\nStop 3, Signal Box Window, 2020: Photographed from a public platform toward an illuminated signal box. One railway worker is visible in profile behind glass, wearing reflective clothing and looking down at paperwork. The long exposure softens one raised hand. Avoid speculating about the worker’s identity, thoughts, emotions, or exact task.\n\nStop 4, Last Bus Connection, 2021: Three passengers stand beneath a bus shelter outside a station. One checks a paper timetable, one holds a folded bicycle, and one sits beside two shopping bags. Their faces are not clearly visible. The station clock reads 00:42. The photograph is about linked public systems and the practical choreography of waiting; do not describe the people as lonely or stranded.\n\nStop 5, First Light, Empty Track, 2023: Parallel tracks run toward a pale blue horizon. Platform lamps are still illuminated. A yellow cleaning cart stands near the frame’s edge. This is the final image in both the series and exhibition.\n\nINTERPRETIVE PRIORITIES\nHelp listeners look slowly at specific visual details before offering context. Connect the works to public infrastructure, shared responsibility, routine labor, and the transition from night to morning. Emphasize that an apparently empty station is still active within a larger system. Treat ordinary fixtures and maintenance objects with dignity without romanticizing them.\n\nACCESSIBILITY AND STYLE\nAssume some listeners have low vision. Establish the composition and spatial relationships clearly, but weave visual description into the interpretation rather than listing objects. Use short-to-medium sentences suited to speech. Avoid rhetorical questions, second-person commands such as “look” or “notice,” and phrases such as “as you can see.” Use “you” sparingly and only to orient the visitor within the exhibition. Define “long exposure” in plain language the first time it appears. Vary transitions so sections do not repeatedly begin with the artwork title or “Here.”\n\nEnd each object stop with one brief observational invitation that can also work for a listener with low vision—for example, inviting attention to contrasts, intervals, repeated sounds in the gallery, or the idea of waiting. Do not presume what a visitor can see. The Closing should connect the first clock to the final dawn image and end on a concrete, memorable sentence rather than a summary or call to action.\n\nBefore responding, silently verify the word count, factual accuracy, section order, pronunciation spelling, and all prohibited language. Output only the studio-ready script."
        }
      ]
    }
  ]'

runware run minimax:m2.7@highspeed \
  messages.0.role=user \
  messages.0.content="Write the complete audio guide for “After the Last Train,” a museum exhibition about photographer Elena Varga’s nighttime documentation of regional rail stations. I’ll provide the approved curatorial facts. The script will be recorded by a professional narrator, so it must be polished and ready for the studio." \
  messages.1.role=assistant \
  messages.1.content="Please send the approved facts, required stops, target runtime, intended audience, interpretive tone, accessibility guidance, pronunciation notes, and any claims or language that must be avoided." \
  messages.2.role=user \
  messages.2.content="Use only the source packet below. Do not invent quotations, dates, locations, motives, biographical details, or photographic techniques.

DELIVERABLE
Create a six-stop English-language audio guide totaling 1,050–1,200 words. Write for curious adult visitors without specialist art knowledge. The experience should feel intimate, observant, and quietly hopeful—not academic, sentimental, eerie, or promotional.

Return the finished script only, organized as:
- Exhibition Introduction
- Stop 1: Platform Clock, 23:17
- Stop 2: Waiting Room with Fern
- Stop 3: Signal Box Window
- Stop 4: Last Bus Connection
- Stop 5: First Light, Empty Track
- Closing

For every section, include an estimated runtime in brackets beneath the heading. Write natural narration in complete paragraphs. Do not include production notes, music cues, stage directions, footnotes, bullet points, or citations in the script.

SOURCE PACKET
Exhibition title: After the Last Train
Artist: Elena Varga (pronounced EH-leh-nah VAR-gah)
Venue: Northbank Museum of Photography

The exhibition contains 42 color photographs made from 2018 through 2023 at small, still-operating railway stations across Hungary, Slovakia, Croatia, and eastern Austria. Varga traveled only on scheduled public transportation and photographed between the final evening arrival and the first morning departure. She did not stage scenes or ask station staff to alter lighting.

Varga began the series after noticing that stations change function late at night: they remain pieces of public infrastructure, but briefly become places of maintenance, waiting, handover, and rest. Her stated interest is in “the hours when a shared place belongs to very few people.” This is the only direct quotation approved for use.

Across the series, recurring details include clocks, ticket windows, potted plants, vending machines, fluorescent fixtures, handwritten notices, benches, reflective safety clothing, cleaning equipment, and tracks disappearing beyond the lit platform. People are present in 11 of the 42 photographs. No individual is identified by name.

Varga used a tripod and long exposures where station rules allowed. The photographs retain mixed light sources rather than correcting them to a uniform white. As a result, fluorescent light may appear green, older lamps amber, and early daylight blue. Long exposure sometimes softens a moving figure, but the architecture and stationary objects remain precise.

The works are presented in chronological order from late evening toward dawn rather than by country. Labels omit exact station names at the artist’s request. She wanted visitors to attend first to repeated civic features rather than treat the series as a catalog of destinations.

OBJECT NOTES
Introduction area: A wall-sized route diagram connects all photographed stations but does not associate individual pictures with named locations.

Stop 1, Platform Clock, 23:17, 2018: A round analog clock occupies the upper center of the frame. Below it, two empty benches face opposite directions. Rainwater reflects a strip of amber platform light. This is the earliest photograph in the series.

Stop 2, Waiting Room with Fern, 2019: A large fern sits between a radiator and a closed ticket window. A handwritten notice is visible but not legible. Green fluorescent light fills the room, while an adjoining corridor appears amber. No people are visible.

Stop 3, Signal Box Window, 2020: Photographed from a public platform toward an illuminated signal box. One railway worker is visible in profile behind glass, wearing reflective clothing and looking down at paperwork. The long exposure softens one raised hand. Avoid speculating about the worker’s identity, thoughts, emotions, or exact task.

Stop 4, Last Bus Connection, 2021: Three passengers stand beneath a bus shelter outside a station. One checks a paper timetable, one holds a folded bicycle, and one sits beside two shopping bags. Their faces are not clearly visible. The station clock reads 00:42. The photograph is about linked public systems and the practical choreography of waiting; do not describe the people as lonely or stranded.

Stop 5, First Light, Empty Track, 2023: Parallel tracks run toward a pale blue horizon. Platform lamps are still illuminated. A yellow cleaning cart stands near the frame’s edge. This is the final image in both the series and exhibition.

INTERPRETIVE PRIORITIES
Help listeners look slowly at specific visual details before offering context. Connect the works to public infrastructure, shared responsibility, routine labor, and the transition from night to morning. Emphasize that an apparently empty station is still active within a larger system. Treat ordinary fixtures and maintenance objects with dignity without romanticizing them.

ACCESSIBILITY AND STYLE
Assume some listeners have low vision. Establish the composition and spatial relationships clearly, but weave visual description into the interpretation rather than listing objects. Use short-to-medium sentences suited to speech. Avoid rhetorical questions, second-person commands such as “look” or “notice,” and phrases such as “as you can see.” Use “you” sparingly and only to orient the visitor within the exhibition. Define “long exposure” in plain language the first time it appears. Vary transitions so sections do not repeatedly begin with the artwork title or “Here.”

End each object stop with one brief observational invitation that can also work for a listener with low vision—for example, inviting attention to contrasts, intervals, repeated sounds in the gallery, or the idea of waiting. Do not presume what a visitor can see. The Closing should connect the first clock to the final dawn image and end on a concrete, memorable sentence rather than a summary or call to action.

Before responding, silently verify the word count, factual accuracy, section order, pronunciation spelling, and all prohibited language. Output only the studio-ready script."

{
  "taskType": "textInference",
  "taskUUID": "6bbb3f76-580b-4094-8882-c58083da811d",
  "model": "minimax:m2.7@highspeed",
  "messages": [
    {
      "role": "user",
      "content": "Write the complete audio guide for “After the Last Train,” a museum exhibition about photographer Elena Varga’s nighttime documentation of regional rail stations. I’ll provide the approved curatorial facts. The script will be recorded by a professional narrator, so it must be polished and ready for the studio."
    },
    {
      "role": "assistant",
      "content": "Please send the approved facts, required stops, target runtime, intended audience, interpretive tone, accessibility guidance, pronunciation notes, and any claims or language that must be avoided."
    },
    {
      "role": "user",
      "content": "Use only the source packet below. Do not invent quotations, dates, locations, motives, biographical details, or photographic techniques.\n\nDELIVERABLE\nCreate a six-stop English-language audio guide totaling 1,050–1,200 words. Write for curious adult visitors without specialist art knowledge. The experience should feel intimate, observant, and quietly hopeful—not academic, sentimental, eerie, or promotional.\n\nReturn the finished script only, organized as:\n- Exhibition Introduction\n- Stop 1: Platform Clock, 23:17\n- Stop 2: Waiting Room with Fern\n- Stop 3: Signal Box Window\n- Stop 4: Last Bus Connection\n- Stop 5: First Light, Empty Track\n- Closing\n\nFor every section, include an estimated runtime in brackets beneath the heading. Write natural narration in complete paragraphs. Do not include production notes, music cues, stage directions, footnotes, bullet points, or citations in the script.\n\nSOURCE PACKET\nExhibition title: After the Last Train\nArtist: Elena Varga (pronounced EH-leh-nah VAR-gah)\nVenue: Northbank Museum of Photography\n\nThe exhibition contains 42 color photographs made from 2018 through 2023 at small, still-operating railway stations across Hungary, Slovakia, Croatia, and eastern Austria. Varga traveled only on scheduled public transportation and photographed between the final evening arrival and the first morning departure. She did not stage scenes or ask station staff to alter lighting.\n\nVarga began the series after noticing that stations change function late at night: they remain pieces of public infrastructure, but briefly become places of maintenance, waiting, handover, and rest. Her stated interest is in “the hours when a shared place belongs to very few people.” This is the only direct quotation approved for use.\n\nAcross the series, recurring details include clocks, ticket windows, potted plants, vending machines, fluorescent fixtures, handwritten notices, benches, reflective safety clothing, cleaning equipment, and tracks disappearing beyond the lit platform. People are present in 11 of the 42 photographs. No individual is identified by name.\n\nVarga used a tripod and long exposures where station rules allowed. The photographs retain mixed light sources rather than correcting them to a uniform white. As a result, fluorescent light may appear green, older lamps amber, and early daylight blue. Long exposure sometimes softens a moving figure, but the architecture and stationary objects remain precise.\n\nThe works are presented in chronological order from late evening toward dawn rather than by country. Labels omit exact station names at the artist’s request. She wanted visitors to attend first to repeated civic features rather than treat the series as a catalog of destinations.\n\nOBJECT NOTES\nIntroduction area: A wall-sized route diagram connects all photographed stations but does not associate individual pictures with named locations.\n\nStop 1, Platform Clock, 23:17, 2018: A round analog clock occupies the upper center of the frame. Below it, two empty benches face opposite directions. Rainwater reflects a strip of amber platform light. This is the earliest photograph in the series.\n\nStop 2, Waiting Room with Fern, 2019: A large fern sits between a radiator and a closed ticket window. A handwritten notice is visible but not legible. Green fluorescent light fills the room, while an adjoining corridor appears amber. No people are visible.\n\nStop 3, Signal Box Window, 2020: Photographed from a public platform toward an illuminated signal box. One railway worker is visible in profile behind glass, wearing reflective clothing and looking down at paperwork. The long exposure softens one raised hand. Avoid speculating about the worker’s identity, thoughts, emotions, or exact task.\n\nStop 4, Last Bus Connection, 2021: Three passengers stand beneath a bus shelter outside a station. One checks a paper timetable, one holds a folded bicycle, and one sits beside two shopping bags. Their faces are not clearly visible. The station clock reads 00:42. The photograph is about linked public systems and the practical choreography of waiting; do not describe the people as lonely or stranded.\n\nStop 5, First Light, Empty Track, 2023: Parallel tracks run toward a pale blue horizon. Platform lamps are still illuminated. A yellow cleaning cart stands near the frame’s edge. This is the final image in both the series and exhibition.\n\nINTERPRETIVE PRIORITIES\nHelp listeners look slowly at specific visual details before offering context. Connect the works to public infrastructure, shared responsibility, routine labor, and the transition from night to morning. Emphasize that an apparently empty station is still active within a larger system. Treat ordinary fixtures and maintenance objects with dignity without romanticizing them.\n\nACCESSIBILITY AND STYLE\nAssume some listeners have low vision. Establish the composition and spatial relationships clearly, but weave visual description into the interpretation rather than listing objects. Use short-to-medium sentences suited to speech. Avoid rhetorical questions, second-person commands such as “look” or “notice,” and phrases such as “as you can see.” Use “you” sparingly and only to orient the visitor within the exhibition. Define “long exposure” in plain language the first time it appears. Vary transitions so sections do not repeatedly begin with the artwork title or “Here.”\n\nEnd each object stop with one brief observational invitation that can also work for a listener with low vision—for example, inviting attention to contrasts, intervals, repeated sounds in the gallery, or the idea of waiting. Do not presume what a visitor can see. The Closing should connect the first clock to the final dawn image and end on a concrete, memorable sentence rather than a summary or call to action.\n\nBefore responding, silently verify the word count, factual accuracy, section order, pronunciation spelling, and all prohibited language. Output only the studio-ready script."
    }
  ]
}