🌈RH全能图片(多并发)@炮老师的小课堂

DapaoRHAllImageConcurrentNode is the batch-shipment version of DapaoRHAllImageNode in the RunningHub section of this pack. Same RunningHub API, same model family - but instead of "one prompt, one image," it takes 任务数量 (task count) and 🚀 并发数 (concurrency), fires a configurable number of parallel jobs, and keeps going when individual ones fail. If you're generating a hundred product shots for an e-commerce listing, this is the node you want; the single-node version is for when you need one image and you need it now.

What's different from the single node

The batch plumbing is all in the required inputs:

• 🔢 任务数量 - how many images total. Default 1, which makes this node behave like its single sibling.
• 🚀 并发数 - parallel jobs at once, default 4. This is your speed/cost/loudness knob: crank it and a 12-image batch finishes in a few waves; crank it too high and RunningHub may rate-limit you.
• 🧪 失败策略 - three options: skip-and-continue (default), and the others for stricter handling. This is the setting that makes unattended batch runs possible.
• 🛟 失败重试次数 - per-task retries, default 1. Note: unlike the dapaoAI-side nodes, this retry is on the submission/queue side, which is exactly where retries are safe.
• 🌊 流式接收 - stream results as they land, default true, so the first finished images show up before the batch finishes.
• 🏷️ 渠道 - note the default differs from the single node: this one defaults to 官方稳定版 (official-stable), where the single node defaults to 低价渠道版. The batch node's author-bias is "when you're paying for scale, pay for stability."

Everything else matches the single node: 🌐 API渠道 (国内版/国外版 - keys don't cross), 🔑 API密钥, 🤖 模型 (G-2 default / V2 / PRO), 🔀 模式 (文生图/图生图), 📝 提示词, ratio/resolution/quality, cache-only seed, up to 10 reference images, 📋 额外参数JSON, and the polling trio.

Outputs

• 🖼️ 图像批次 - a batched IMAGE (is_list false, so it arrives as one tensor batch - join it directly to a save node).
• 🔗 图片链接 - the URLs, as a string.
• 📋 响应信息 - per-task status/diagnostics.

The honest read

This node is the reason the RH section of the pack exists: RunningHub's whole pitch is cheap GPU-backed inference, and cheap + batch is where the savings compound. The concurrency and failure-policy fields are exactly what a real batch job needs, and the defaults (4 concurrent, skip failures, retry once) are sane for a first run.

Two realities to keep in view:

• Concurrency is money in motion. 任务数量 100 at 并发数 10 is 100 billed jobs; a runaway loop on a list input can burn a balance fast. The stream output helps you notice early.
• The cheap channel tempts, the stable channel delivers. The README's own framing is that the low-price channel trades quality consistency for cost. For anything you'll actually ship, run the stable channel - the price delta is usually worth not regenerating a quarter of a batch.

Gotchas

• 国内版/国外版 key mismatch - the eternal RH trap; keys don't transfer.
• Failure-policy "skip" can hide problems. If you get back 30 images from a 40-task run with no obvious error, check 响应信息 before re-running the whole batch.
• 图生图 still needs a reference image wired in - mode doesn't auto-switch.
• Chinese-only labels.

Install: ComfyUI Manager → "dapaoAPI", or git clone https://github.com/paolaoshi/ComfyUI-dapaoAPI.git → pip install -r requirements.txt → restart. Needs a RunningHub account + API key; the README's invite code (kol01-rh138) gives the author a bounty, which is how a free pack like this stays alive.