EcoHash
Official EcoHash API nodes for public and dedicated models: image and video generation, image editing, LLM and vision tools, TTS and STT. One EcoHash API key, no local model weights required.
Nodes (14)
Say What Should Change, Not Where to Paint
Your Own Endpoint, Wired Into the Canvas
A Prompt Rewriter That Doesn't Eat Your VRAM
One Clip In, One Transcript Out, No Whisper Install
EcoHash Dedicated TTS Won't Run Until You Fill In the Voice
Keep the Job ID, It's Your Undo Button
Turn a Picture Back Into a Prompt
Edit an image with a sentence — no masks, no ControlNet, no inpainting
A text-to-image node with zero VRAM and zero model downloads
A prompt-enhancer LLM inside ComfyUI that never touches your VRAM
Speech to text in the graph — subtitles and roundtrips without a Whisper install
Give your silent renders a voice — hosted TTS with zero dependency pain
EcoHash Video Generate Is Text-Only, and the Model List Doesn't Advertise That
Turn any image back into a prompt — captioning in the graph, no local VLM
What is EcoHash
EcoHash runs an NVIDIA RTX PRO 6000 GPU cloud and, on top of it, an OpenAI-compatible inference API for open models: image generation and editing, LLM chat, vision-language description, video generation, text-to-speech, and speech-to-text. This package adds fourteen ComfyUI nodes that call that API directly from your workflows — no local GPU or model download required for these nodes.
Need a whole GPU instead of an API call? The same account rents RTX PRO 6000 Blackwell instances (96 GB VRAM) by the hour, so you can run a full ComfyUI stack with local checkpoints alongside these nodes. See the GPU cloud docs.
Sign up at ecohash.com to get an API key. New accounts receive a small free starter credit so you can try the nodes before adding a payment method. Full docs live at docs.ecohash.com.
Install
For complete setup instructions, all fourteen nodes, example workflows, video-task recovery, API compatibility and troubleshooting, read the User Guide.
Option A — Comfy Registry and ComfyUI-Manager
Find EcoHash, package ID comfyui-ecohash, on
Comfy Registry.
The page includes the publisher, published versions, repository link and install command.
In ComfyUI-Manager, search for EcoHash or comfyui-ecohash and install the package.
If you use Comfy CLI with your configured ComfyUI environment, run:
comfy node install comfyui-ecohash
Restart ComfyUI and refresh the browser after installation. Registry installations use published releases; the available nodes depend on the version installed.
Option B — git clone
cd ComfyUI/custom_nodes
git clone https://github.com/ecohash-ai/ComfyUI-EcoHash.git
cd ComfyUI-EcoHash
pip install -r requirements.txt
Restart ComfyUI after installing.
Dependencies must use ComfyUI's Python. For Windows Portable, run from the portable folder:
.\python_embeded\python.exe -m pip install -r .\ComfyUI\custom_nodes\ComfyUI-EcoHash\requirements.txt
For Desktop or a virtual-environment install, use that installation's Python environment. After updating the plugin, restart the ComfyUI server and refresh the browser.
Connection manager
Open the EcoHash cloud tab in the sidebar, click Connections (Chinese: 连接设置) on any EcoHash node, or use Extensions → EcoHash → EcoHash connections.
- Enter your key once in EcoHash API Key and click Save and verify API Key. Public models (A) and dedicated EcoHash instances (B) share this global credential.
- For dedicated models, click Add connection, enter a unique Connection name and API URL, then Save and connect. The list and nodes display that same name. EcoHash connections need no separate key. IDs are generated internally; timeouts and independent keys for third-party endpoints remain under Advanced connection settings.
- Saving checks the model list without running inference. When opened from a dedicated
node, the manager also applies the saved connection to that node. A service without
/modelsis reported as reachable, with authentication/inference still unverified. - Choose a discovered model in the node, or choose Enter model ID manually. A single discovered model fills an empty selection. Changing the connection clears the previous model; refreshing the catalog preserves an existing model ID for explicit remapping.
- Enter the prompt and generation settings, connect an output node and run. Use Refresh models after changing the deployment or platform catalog. No protocol selector is needed.
Generate test image is a separate action requiring a model ID and explicit confirmation before making a real, potentially billable generation request. Its preview appears in the manager. Test text inference separately verifies language-model completion with the selected connection; it also requires a model ID and explicit consent. Audio models can be tested in a workflow using Dedicated TTS/STT and Preview Audio or a downstream node.
Keys are saved only on the ComfyUI server in git-ignored connections.json (owner-readable
and writable on POSIX). The settings API never returns saved keys. Leave a key field blank to
keep its value. Per-field precedence: nonempty environment variable → saved settings →
config.ini → default. Environment-managed fields are marked and locked. The public API URL
and public timeout (180 seconds) are fixed; dedicated timeout defaults to 300 seconds.
The global key is used only for the exact https://api.ecohash.com origin (default HTTPS
port, including instance proxy paths). Other hosts, HTTP URLs and alternate ports require
their own endpoint key. Updating the global key applies to every EcoHash connection and
invalidates their model-discovery and node-execution caches. The key is referenced, not
copied into each connection record. ECOHASH_DEDICATED_API_KEY applies only to independent
endpoints; EcoHash connections always read ECOHASH_API_KEY / the global setting.
Each Dedicated node selects a real saved connection. No empty default dedicated endpoint is
listed. A new node selects a connection automatically only when it is unambiguous or was
explicitly preferred; otherwise it prompts for a selection. Multiple deployments can run
in one workflow. Workflows store internal connection IDs and model IDs; recipients select
their own local connection. Missing connections never silently fall back to another deployment.
Renaming a connection keeps its internal ID and updates its name in the list and nodes.
The public nodes use public.
The manager configures this ComfyUI server. Use it from the same trusted server environment where you manage your existing ComfyUI configuration.
Set your API key
This package deliberately has no API-key input widget on any node. Workflow JSON files are frequently shared, screenshotted, or committed to git, and a key typed into a node widget would be saved right there in plain text for anyone who opens the file. Instead, the key is read once at request time from outside the workflow graph. The connection manager is the easiest option; server deployments can also use:
-
Environment variable — set
ECOHASH_API_KEYin the environment ComfyUI runs in, e.g.:export ECOHASH_API_KEY="eco_..." -
config.ini— copyconfig.ini.example(in this folder) toconfig.iniand paste your key:[ecohash] api_key = eco_YOUR_KEY_HEREconfig.iniis git-ignored by this repo so it won't be committed by accident.
Get or manage your key in the API keys guide.
Never paste your API key into a node widget, a workflow JSON, or a screenshot. Any key value embedded in a saved workflow is exposed to everyone who receives that file.
Nodes
| Node | Inputs | Outputs | Purpose |
|---|---|---|---|
| EcoHash Image Generate | model, prompt, size (presets / custom), steps (0 = model default), seed (-1 = random), optional width/height | IMAGE | Generate an image from a text prompt. |
| EcoHash Image Edit | image (IMAGE), model (catalog models that support image edit), prompt (string), size (auto + preset sizes; auto keeps the source image's dimensions) | IMAGE | Edit/transform an existing image with a text instruction. |
| EcoHash Dedicated Image Generate | model (deployed ID), prompt, size, steps, seed, optional width/height and connection | IMAGE | Generate via the selected dedicated endpoint. |
| EcoHash Dedicated Image Edit | image, model (deployed ID), prompt, size, optional width/height and connection | IMAGE | Edit via the selected dedicated endpoint, if supported. |
| EcoHash LLM | model (catalog llm/llm_vision models), mode (chat / prompt_enhance / translate_to_english / custom), text (string), system_prompt (string, used only when mode = custom), temperature (float, 0–2), max_tokens (int) | STRING | Run a chat completion. prompt_enhance rewrites a rough idea into a detailed image-generation prompt; translate_to_english converts non-English text; useful chained before an Image Generate node. |
| EcoHash VLM Describe | image (IMAGE), model (catalog llm_vision models), prompt (string, defaults to a detailed-description instruction), max_tokens (int) | STRING | Describe an input image in detail with a vision-language model — handy for turning an existing image into a prompt for regeneration or variation. |
| EcoHash TTS | model (catalog speech_tts models), text (string), voice (string, e.g. af_bella), speed (float, 0.5–2.0) | AUDIO | Synthesize speech from text. |
| EcoHash STT | audio (AUDIO), model (catalog speech_stt models), language (string, ISO-639-1 code, empty = auto-detect) | STRING | Transcribe speech audio to text. |
| EcoHash Dedicated LLM | deployed model, mode, text, system_prompt, temperature, max_tokens, optional connection | STRING | Chat, enhance prompts or translate with a dedicated language model. |
| EcoHash Dedicated VLM Describe | image, deployed model, prompt, max_tokens, optional connection | STRING | Describe an image with a dedicated vision-language model. |
| EcoHash Dedicated TTS | deployed model, text, model-specific voice, speed, optional connection | AUDIO | Generate speech with a dedicated model. |
| EcoHash Dedicated STT | audio, deployed model, language, optional connection | STRING | Transcribe audio with a dedicated model. |
| EcoHash Video Generate | model, prompt, duration, aspect_ratio, resolution, generate_audio, optional job_id/wait_timeout | VIDEO, job-ID STRING | Generate a public text-to-video task, then connect VIDEO to Save Video. |
| EcoHash Dedicated Video Generate | deployed model, prompt, duration, aspect_ratio, resolution, optional custom width/height, job_id/wait_timeout, connection | VIDEO, job-ID STRING | Generate or resume video on the selected dedicated service. |
Public model menus initially use the cache or packaged snapshot without blocking node loading on the network. The frontend refreshes separately; Refresh models updates menus and reports live/cache/fallback status. Capability hints show editing support, default steps, supported sizes and voices when supplied by the service. Unknown private-model capabilities remain unspecified.
Image nodes offer landscape/portrait presets and custom dimensions (multiples of 16, 64–8192
per side). Width/height appear only in custom mode; the model must support the requested size.
LLM's system prompt appears only in custom mode. TTS lists voices when provided by the service;
otherwise enter a model-specific voice ID. Switching models selects a declared voice when
available, or clears the previous voice so it can be entered explicitly.
ComfyUI caches node outputs — re-queueing an identical graph returns cached results without a new API call; change any input (e.g. seed) to force a fresh call. Node status shows waiting, completed, cached, interrupted or failed execution.
Dedicated inference (B)
The node library has two groups under EcoHash: Public Models and Dedicated Models (Chinese: 公共模型 and 专有模型), each containing seven nodes. The seven dedicated nodes cover all supported model types: image generation/editing, LLM, vision-language description, video, TTS and STT. They call separately deployed models. The public nodes continue to use the public EcoHash API (A), so both can be used in the same workflow. Dedicated nodes discover deployed model IDs and also accept a free-text ID. They do not require registration in the public model catalog.
After configuring your global key, add the following to your local, git-ignored config.ini
(also see config.ini.example):
[dedicated]
base_url = https://api.ecohash.com/inference-instances/INSTANCE_ID/proxy/v1
timeout = 300
Alternatively, set these variables in the environment of the ComfyUI server process:
export ECOHASH_API_KEY="YOUR_ECOHASH_KEY"
export ECOHASH_DEDICATED_BASE_URL="https://api.ecohash.com/inference-instances/INSTANCE_ID/proxy/v1"
export ECOHASH_DEDICATED_TIMEOUT="300"
Nonempty environment values override saved UI settings, which override INI values. Timeout is in seconds
and defaults to 300. base_url must include the service's actual API prefix: /v1 is not
added automatically. A copied full generation endpoint is normalized to its API prefix. Trailing slashes are removed; a prefix such as /deployments/my-model/v1
is preserved. EcoHash connections share the global key. Other endpoints require their own
key and never inherit the EcoHash key. Neither the endpoint nor its key is stored in node
widgets or workflow JSON.
Additional connections can also be configured as INI sections. For a self-hosted or third-party endpoint, include its independent key (or use the manager's Advanced connection settings):
[connection:studio]
name = Studio image model
base_url = https://ANOTHER_HOST/v1
api_key = YOUR_DEDICATED_KEY_HERE
timeout = 300
In ComfyUI, add EcoHash Dedicated Image Generate, enter the exact model ID served by your
deployment, select its connection, then connect its output to Preview Image or Save Image. To edit, connect
an IMAGE to EcoHash Dedicated Image Edit and enter a model ID that supports editing.
An example workflow is in example_workflows/dedicated_text_to_image.json.
Dedicated nodes invalidate their cached result when their endpoint, key or timeout changes.
For language models, use EcoHash Dedicated LLM with the same four modes as the public LLM. Its STRING output can feed either a public or dedicated image node. Use EcoHash Dedicated VLM Describe for a model that supports image inputs. For audio, use EcoHash Dedicated TTS/STT; enter the model's actual voice ID for TTS. Each node independently selects its connection, so language, image and audio deployments can use different URLs in one workflow. Official EcoHash connections share the global key; independent services use their own endpoint keys.
Image, chat and speech nodes use a synchronous, OpenAI-compatible API. Deployments implement the routes needed by their model type; they do not need to implement every route:
- Generation:
POST {base_url}/images/generations, JSON withmodel,prompt,size,response_format=b64_json, and optionalsteps/seed. - Editing:
POST {base_url}/images/edits, multipart form withmodel,prompt,sizeand animagePNG file.size=autosends the source dimensions explicitly. - Authentication:
Authorization: Bearer <global EcoHash key>for EcoHash URLs; an independent endpoint key elsewhere. - Result:
{"data": [{"b64_json": "..."}]}or{"data": [{"url": "https://..."}]}. Result URLs must be downloadable without the inference key; that key is never forwarded when fetching an image URL. Each call produces oneIMAGEfrom the first result. - Language/vision:
POST {base_url}/chat/completions, withmodel,messages,max_tokensand (for text LLM)temperature. Vision usestextandimage_urlparts with an inline PNG data URL. Return string text atchoices[0].message.content. Requests are non-streaming; a text-only model does not provide vision support. - TTS:
POST {base_url}/audio/speech, JSON withmodel,input,voice,speed,response_format=wav; return 16-bit PCM WAV bytes. - STT:
POST {base_url}/audio/transcriptions, multipart fieldsmodel, optionallanguage, and WAVfile; return{"text": "..."}JSON.
Image, chat and speech nodes report HTTP 202 jobs as unsupported. Video nodes implement the asynchronous lifecycle below. Other response schemas, authentication schemes, and deployment-routing headers also require an adapter. API requests do not follow redirects; configure the final API base URL directly. LoRA, ControlNet and other deployment-specific parameters are not exposed by these nodes yet.
Before testing a real service, verify it implements this protocol. The automated tests cover
configuration, credential isolation and real loopback HTTP image, language and audio calls against a test
service; they do not establish compatibility with any particular dedicated deployment.
See tests/DEDICATED_RESULTS.md for the verification record.
For images, after configuring your real endpoint, run the node-level smoke test in a Python environment with the ComfyUI dependencies available (this makes real, billable API calls):
python scripts/smoke_dedicated.py --model YOUR_DEPLOYED_MODEL_ID
# If the deployment also supports editing:
python scripts/smoke_dedicated.py --model YOUR_DEPLOYED_MODEL_ID --edit-model YOUR_EDIT_MODEL_ID
# Use a named connection:
python scripts/smoke_dedicated.py --connection studio --model YOUR_DEPLOYED_MODEL_ID
The script saves generated/edited PNGs in a temporary directory and prints their paths and dimensions. It exercises the node implementations directly; it does not start ComfyUI or validate browser interaction or workflow execution.
Video generation
Use EcoHash Video Generate under Public Models, or EcoHash Dedicated Video Generate
under Dedicated Models. Connect video (VIDEO) to ComfyUI's Save Video for playback and
permanent saving with its audio track. A current ComfyUI installation with native VIDEO support
is required. The node downloads an MP4 to ComfyUI's temporary directory; Save Video writes it
to the output directory. These nodes support text-to-video; models requiring image/audio/video
references need a separate reference-input adapter.
No node or connection exposes a protocol setting. The public node calls the EcoHash video
API (/video/generations and /video/jobs/{id}). The dedicated node uses the platform's
standard video contract (/videos, /videos/{id}, /videos/{id}/content). Wire formats are
internal implementation details, independent of model names.
By default dedicated generation sends model, prompt, seconds and size. The service owns
model-native parameters and default steps, frame rate, seeds, audio generation and provider
adaptation. Any audio in the returned MP4 is preserved. Public generation additionally offers
the gateway's audio option. Choose duration, aspect ratio and resolution; dedicated nodes
also support custom width/height. Duration limits come from model metadata where available.
Public resolution choices use /platform/video-pricing bands without reference-video input.
Those bands describe priced resolutions; the service still validates actual model support.
Missing capability metadata is shown as unknown. Common size presets are conveniences,
not assertions that a model supports them.
Model discovery uses /platform/models for public models and authenticated
/models?category=all for dedicated connections (EcoLink otherwise defaults to chat-only).
Filtering follows declared categories and input/output modalities, never model-name patterns.
Unclassified models remain selectable with an unverified capability hint. Model IDs and
configured API prefixes are sent unchanged; the plugin does not infer instance routing.
Deployment compatibility: an endpoint exposing a /videos route is not by itself proof
that its create payload implements the platform contract. A container that needs native
JSON parameters requires adaptation. For an existing nonstandard deployment, administrators
can install an explicit server-side request mapping bound to
its exact connection URL, served model ID and runtime metadata. This local compatibility
bridge does not fix the platform gateway or infer adapters from model names. No native
parameter or protocol selector is added to nodes or workflows. See the historical
verification record and current
configuration and EcoLink audit.
Old video workflows migrate field values and links; retired native
workflows display a compatibility reminder. Their old protocol/steps/seed fields are removed.
Video submission accepts HTTP 200 or 202 and never automatically retries a POST. The node
shows the job ID and status in Video task, also logs the ID on the server, and returns it
as an optional STRING output. Normal wiring is video → Save Video.video; job_id can stay
unconnected. To display or keep the completed job ID, connect job_id → Preview as Text.source
or job_id → Save Text.text. These downstream nodes receive the ID after generation completes.
While waiting, or after a failure, copy it from Video task or the server log instead.
Do not connect the output back to the same node's job_id input: that creates a graph cycle.
wait_timeout (default 1800 seconds) limits waiting for completion;
the connection timeout limits each request. To recover after a timeout or interruption, paste
the ID into job_id under Advanced settings, keeping the same connection. This resumes polling and download
without another generation. Clear job_id to generate a new video. Stopping ComfyUI only stops
local waiting; the remote job may continue. Signed result URLs and content redirects are fetched
without forwarding the inference key. Downloads are limited to 512 MiB per video.
Connection checks appear in the connection manager. Nodes show execution status; model capability information is available in selector tooltips. There is no separate model-information or connection-status row.
Connect EcoHash IMAGE, STRING, AUDIO and VIDEO outputs to compatible processing, preview and save nodes. See workflow composition for examples combining public and dedicated models.
Examples: text_to_video.json and
dedicated_text_to_video.json.
Example workflows
Ready-to-load example workflows (text-to-image, prompt-enhance-then-generate,
describe-and-regenerate, and TTS voiceover) live in example_workflows/.
Open Workflow → Browse Templates and find this plugin, or drag a JSON file into the canvas.
Node help is available in each node's information panel. The original public API smoke-test
record is in tests/SMOKE_RESULTS.md; dedicated verification is
recorded separately in tests/DEDICATED_RESULTS.md.
Pricing
EcoHash is metered, pay-as-you-go usage on top of your free starter credit. See current plans and per-model pricing at the billing docs.
Support
- Bugs and feature requests: GitHub Issues (this repo)
- Account or billing questions: the EcoLink console support chat — EcoHash support
Local verification
Using a ComfyUI Python environment (including PyTorch, Pillow, NumPy and aiohttp), install
pytest and run python -m pytest -q. Frontend checks use node --test tests/test_ui.mjs tests/test_manager.mjs;
no frontend build step is required. The tests isolate credentials and use temporary local HTTP
services. See tests/DEDICATED_RESULTS.md for the running-ComfyUI
verification and the remaining checks for a real dedicated deployment.
中文说明
EcoHash 提供 NVIDIA RTX PRO 6000 GPU 云，以及构建其上的 OpenAI 兼容推理 API。ComfyUI-EcoHash 为 ComfyUI 提供 7 个公共 API 节点，调用图像生成/编辑、视频生成、LLM 对话、图像描述（VLM）、 文本转语音（TTS）和语音转文本（STT），以及覆盖相同能力的 7 个专有模型节点。 官方公共与专有连接共用一个 EcoHash API Key；这些 API 节点无需本地推理 GPU 或下载模型。
GPU 云：同一个账号也可按小时租用 RTX PRO 6000 Blackwell 实例（96 GB 显存）， 用于运行带本地模型的完整 ComfyUI 环境，详见 GPU 云文档。
注册：前往 ecohash.com 注册账号，新账号会获得少量免费额度用于试用；完整文档见 docs.ecohash.com。
安装：在 Comfy Registry 插件页面
查看发布版本和安装命令。可在 ComfyUI-Manager 中搜索 "EcoHash" / "comfyui-ecohash"，
或使用已配置好环境的 Comfy CLI 执行 comfy node install comfyui-ecohash。
也可将本仓库 git clone 到 ComfyUI/custom_nodes/，用 ComfyUI 自身的 Python
安装 requirements.txt。安装后重启 ComfyUI 并刷新浏览器。
界面配置：打开侧栏的 EcoHash 云图标，或节点上的 连接设置。
在顶部 全局 API Key 填写一次，点击 保存并验证 API Key，公共模型和 EcoHash 专有实例共用。
专有模型添加连接时只填写一个 连接名称 和 API 地址，点击 保存并连接。
连接列表和节点都显示同一个名称，禁止重名；内部 ID 自动生成，不在界面中另外命名。
未配置地址的默认专有连接不再出现在列表中。
从节点打开设置时会直接绑定该连接，再选择已发现的模型或手动填写 ID。
超时和第三方端点的独立密钥放在高级设置中；连接检查只读取模型列表。
全局密钥仅用于 https://api.ecohash.com 官方地址（默认 HTTPS 端口），更换后所有 EcoHash
连接立即生效，相关模型发现和节点缓存失效。第三方地址不会自动取得全局密钥。
语言、图像的实际推理测试需另行确认；TTS/STT 在工作流中连接音频节点验证。
专属推理（B）：在 config.ini 添加 [dedicated]，配置 base_url 和可选的
timeout（秒，默认 300），官方实例自动读取全局密钥；也可使用 ECOHASH_DEDICATED_BASE_URL
和 ECOHASH_DEDICATED_TIMEOUT 环境变量。第三方端点另外配置 api_key 或 ECOHASH_DEDICATED_API_KEY。
base_url 要包含服务实际的 API 前缀，插件不会自动添加
/v1。添加 EcoHash Dedicated Image Generate / Edit 节点，直接填写部署模型 ID 即可调用，
无需模型出现在公共目录中。专属节点的 connection 可分别选择不同部署，也可通过
[connection:连接ID] 配置多个端点。分享工作流时接收方配置同名连接，或在节点中重新选择。
专属语言模型使用 EcoHash Dedicated LLM，支持对话、提示词增强、翻译和自定义模式；
图片描述使用 EcoHash Dedicated VLM Describe，要求模型支持图片输入。语音模型使用
EcoHash Dedicated TTS / STT；TTS 的音色需按模型选择或填写。
每个节点都可选择不同专属实例，A、B 共用全局 EcoHash 密钥并可混用。当前支持同步的 OpenAI 兼容
图像、聊天和语音接口；视频节点支持异步提交、任务查询和 MP4 下载，详见上方 Video generation。
视频：新增 EcoHash 视频生成 / 专属视频生成，输出 VIDEO 接“保存视频”，保留音轨。
支持文生视频；需要参考图、参考音频或参考视频的模型尚未接入。
无需选择或配置接口协议。选择连接和模型后，填写提示词、时长和尺寸即可；模型能力、时长上限
和公共模型的分辨率档位从服务端读取，不按 Seedance、MiniMax、Wan 或 LTX 等名称分类。
步数、帧率等由平台管理。专有模型可选自定义宽高；已有任务 ID 和等待超时在高级设置中。
原生视频实例需要兼容接口或管理员安装的服务端请求映射，不能仅凭模型 ID 保证支持。
请求映射按当前连接和部署元数据绑定，不在节点中暴露原生参数，也不按模型名称套用规则。
节点不显示连接状态和模型信息字段；能力说明放在模型选择器的悬浮提示和连接设置中。
参数和帮助：size=custom 时显示宽高；LLM 的 custom 模式才显示系统提示词。
节点库的 EcoHash 下分为“公共模型”和“专有模型”两个分组，每组各包含 7 个节点。
音色按模型元数据显示可选项，未提供列表时支持手填。示例可从工作流模板浏览器加载；节点的信息面板
提供使用说明。Windows Portable 安装依赖时请使用内嵌 Python，命令见上方 Install。
价格：按量计费，详见计费文档。
支持：在本仓库的 GitHub Issues 提交问题，或通过 EcoLink 控制台的在线支持聊天获取帮助： 支持页面。