token-plan-image

name	token-plan-image
description	Call the Qwen Token Plan text-to-image / image-edit models (qwen-image-2.0, qwen-image-2.0-pro, wan2.7-image, wan2.7-image-pro) to generate or edit images from a text description. Activates when the user asks to draw, render, generate, or edit an image. For Malay/SEA phenotype prompts, prefer minimax-cli (MiniMax image-01).
version	2.0.0
tags	["qwen","token-plan","image-generation","text-to-image","image-editing"]
metadata	{"hermes":{"category":"creative","related_skills":["minimax-cli","lightweight-image-generation"]},"forge_policy":"/root/A-FORGE/forge_work/2026-07-20/model-selection-policy.md"}

Call the Qwen Token Plan multimodal-generation API to generate (or edit) an image based on a description.

User request: $ARGUMENTS

Inputs to extract

Parse $ARGUMENTS and pull out (with defaults shown):

• prompt (required): the image description / edit instruction
• model (default qwen-image-2.0): one of qwen-image-2.0, qwen-image-2.0-pro, wan2.7-image, wan2.7-image-pro
• size (default 1024*1024): 1024*1024, 1280*1280, 720*1280, 1280*720, or any W*H accepted by the model
• n (default 1): number of images (1–4)

Step 1 — Call the API (use bash via terminal / shell tool)

The auth token is $QWEN_API_KEY (already in the shell environment, sourced from /root/.secrets/qwen.env or /root/.openclaw/.env).

curl -s -X POST "https://dashscope-intl.aliyuncs.com/api/v1/services/aigc/multimodal-generation/generation" \
  -H "Authorization: Bearer $QWEN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "<MODEL>",
    "input": {
      "messages": [
        { "role": "user", "content": [ { "text": "<PROMPT>" } ] }
      ]
    },
    "parameters": { "size": "<SIZE>", "n": <N> }
  }'

If the response is HTTP 400 with code: "Throttling.AllocationQuota" → token-plan quota exhausted. Tell the user to top up at https://home.qwencloud.com/billing/subscription/token-plan and stop.

If the response is HTTP 401 → token invalid. Tell the user to rotate QWEN_API_KEY in /root/.secrets/qwen.env and re-source.

If the response is HTTP 200 with code: "InvalidApiKey" → the API key has valid format but is expired/deactivated. This is distinct from 401 (bad format) — DashScope returns 200 with an error body. Tell the user the key needs renewal at https://home.qwencloud.com/billing/subscription/token-plan. Rotate key in /root/.secrets/qwen.env and re-source. Proven 2026-07-20: sk-sp-D.IPRH... returned {"code":"InvalidApiKey","message":"Invalid API-key provided."}.

Pre-flight: Before calling the API, verify the key is set and healthy: echo $QWEN_API_KEY should be non-empty. Expired keys fail with 200+InvalidApiKey, not 401 — easy to miss.

Step 2 — Extract image URLs

The response JSON shape:

{
  "output": {
    "choices": [
      {
        "message": {
          "content": [
            { "image": "https://<signed-url>" }
          ]
        }
      }
    ]
  }
}

Walk every output.choices[*].message.content[*].image URL.

Step 3 — Download the images

For each URL, download to current working directory:

curl -sL -o "generated_$(date +%Y%m%d_%H%M%S)_<i>.png" "<URL>"

If Content-Type from the response header is image/jpeg change the extension to .jpg; if image/webp to .webp. PNG is the default.

Step 4 — Report

Print to the user:

• The exact local file path(s) generated
• The model + size used
• A one-line description of the image (re-read the prompt)

📋 Notes

• Token Plan image generation is billed from Credits at the same rate as text — no extra surcharge.
• qwen-image-2.0-pro and wan2.7-image-pro are higher quality / slower; default qwen-image-2.0 is fastest.
• For image-editing (input image + edit prompt), use wan2.7-image-pro with input.messages[0].content = [{ "image": "<input_url>" }, { "text": "<edit prompt>" }].
• Quota: each generation consumes Credits; check usage block in response for cost.

---

🧭 Route

Canonical model-selection lives in minimax-cli. Policy: /root/A-FORGE/forge_work/2026-07-20/model-selection-policy.md

When to use Qwen over MiniMax: Text+image editing (wan2.7-image-pro — only model supporting image input). Generic prompts where phenotype doesn't matter. When MiniMax quota is exhausted.

Prompt decomposition for Malay slang: See minimax-cli 🧬 Phenotype. Add explicit "Southeast Asian Malay" tokens. Never rely on slang alone.

Default chain: MiniMax image-01 (SEA/realism) → Qwen image-2.0-pro (generic) → Pollinations (free).