EvoLink (Official)
Official EvoLink nodes — 56 nodes, one node per model variant: Nano Banana 2/Pro/Lite, Seedream 5.0, GPT Image 2, Qwen Image 3.0, Z-Image Turbo, Krea 2, Grok Imagine, Sora 2/Pro, Veo 3.1, Kling V3/Turbo/O3 (ref/edit/motion), Seedance 2.5/2.0 (ref/edit/extend), Wan 3.0, Hailuo 2.3, Gemini Omni Flash, Topaz Upscale. Chain generate→edit→upscale via result_urls. No local GPU, no third-party image hosting, async & cancellable, API key stays out of workflow JSON. 国内直连,中文支持。
Nodes (56)
Check your EvoLink balance before you run another video job
Edit a video you already generated, without leaving the graph
Turn one still into a 10-second clip on Google's cheap video tier
Keep a character consistent across clips with reference-to-video
The cheapest way to get a moving shot out of ComfyUI
GPT Image 2 Inside ComfyUI, No GPU Required
XAI's image model with the least restrictive filter, inside ComfyUI
One node, three video modes — text, first frame, or reference
Animate your image with xAI — and a mode knob that changes the vibe
XAI's text-to-video with a fun/spicy dial — and a 720p ceiling
The MiniMax motion model — strong acting, decent price, one knob to watch
The cheaper Hailuo — same motion, image-to-video only, one fewer frame
Every image model in the pack, behind one dropdown
Kling motion control transfers choreography
Reshoot a clip's jacket without reshooting the clip — Kling O3 video editing
4K image-to-video with a first and last frame
Reference video in, new video out — Kling O3 steals style and motion
4K text-to-video with sound on tap
The fast, cheap Kling — image-to-video, no frills, 1080p ceiling
Kling's cheap text-to-video tier, stripped to what matters
The full-feature workhorse with first/last frames
3 to 15 seconds of video from a text box
Krea 2 is open-source — so why call it through an API?
Nano Banana 2, inside your ComfyUI graph
Fixed 1K output, built for batches
The flagship Gemini image model, as a node
The model that can spell, now in your graph
The same great text rendering, one tier up
Change a photo by describing the change
Cheap motion from a frame you already like
The video draft tier that respects your balance
Turn one frame into up to 15 seconds
The cheapest way to animate a frame
The cheapest video in the EvoLink pack
ByteDance's video workhorse, up to 4K
The 'chain generation into editing' node
Keep the clip going past its last frame
First and last frame, 4 to 30 seconds of motion
Throw images, clips and audio at it
The current gen, with smart duration and free audio
ByteDance's image model, up to 15 shots a call
ByteDance's Seedream 4.5, In Your Graph — No GPU, No Downloads
The Volume Tier for When You Need Fifteen Images, Not One
One Great Image, Not Fifteen Decent Ones
OpenAI's Video Model, Via a Node That Isn't Secret
The 1080p Upgrade, and What It Costs You Per Second
The Post-Processing Step Your Cloud Pipeline Was Missing
The EvoLink Upload Node You Probably Don't Need (Yet)
First Frame In, Sound Included, Up to 4K Out
Keep the Subject, Not Just the First Frame
The Model That Brings Its Own Audio
The All-Models Escape Hatch, Beta Channels Included
First and Last Frame, With a 480p Budget Door
Images, Video, and Audio as Your Source Material
Alibaba's Newest Video, API-Only, 480p Budget Tier Included
The Local Darling, For When You Have No GPU
ComfyUI-EvoLink(EvoLink 官方节点)
English | 官方教程:https://evolink.ai/docs/cn/integration-guide/comfyui
在 ComfyUI 里直接调用 EvoLink 云端模型出图出视频,本地无需显卡。
图像:GPT Image 2 | Nano Banana 2 / Pro / Lite | Seedream 5.0 Pro / Lite / 4.5 / 4.0 | Qwen Image 3.0 / Pro / Edit Plus | Z-Image Turbo | Krea 2 Turbo | Grok Imagine Image 2.0
视频:Sora 2 / Pro | Veo 3.1(文生 / 首尾帧 / 参考图)| Kling V3 / Turbo / O3(含参考视频 / 视频编辑 / 运动控制)| Seedance 2.5(含参考 / 编辑 / 延长)/ 2.0 / Fast / Mini | Wan 3.0(文生 / 首尾帧 / 参考)| Grok Imagine Video | Hailuo 2.3 | Gemini Omni Flash | Topaz 视频超分 —— 输出直连原生「保存视频」节点,任务可随时取消
通用:EvoLink Image / Video 动态节点覆盖全部模型(含 beta 低价渠道),新模型上线无需更新插件;Credits 节点一键查余额
编排:所有生成节点输出 result_urls / response_json,视频编辑/延长/超分节点直接吃上游的 result_urls,可串成「生成 → 编辑 → 超分」流水线
每个模型变体一个节点(文生视频与图生视频分开),参数即所见即所得,不会出现"选了某变体但参数不适用"的情况。图生视频节点的
first_frame为必接口。
- ✅ 图生图零门槛:「加载图像」连到
image1就行 —— 本地图自动经 EvoLink 官方通道上传(72 小时自动删除),不需要注册任何第三方图床 - ✅ Key 只留在本机:首次在节点框填写后自动保存,之后留空即可,不会跟着工作流 JSON 被分享出去
- ✅ 中文全程可用:prompt 支持中文,报错提示是中文
- ✅ 省钱:参数不变重复运行走 ComfyUI 缓存不重复扣费;失败/超时任务不扣费
旧版用户请注意:本插件替代已下架的旧版 Evolink-AI 官方节点,恢复了"加载图像直连图生图"的用法。 如果你在用社区版
torans/ComfyUI_EvoLink(图生图需要 imgbb 图床),两者可共存,迁移对照见下文。
安装
方式一:ComfyUI-Manager(推荐,上架后可用)
Manager → Custom Nodes Manager → 搜索 EvoLink → 安装 → 完全重启 ComfyUI。
方式二:git
cd ComfyUI/custom_nodes
git clone https://github.com/deeplearning-goethe/comfyui-evolink.git
方式三:下载 ZIP
Code → Download ZIP → 解压后把文件夹放进 ComfyUI/custom_nodes/。
⚠️ 检查一层:打开 custom_nodes/comfyui-evolink/ 应直接看到 __init__.py;如果里面还套了一层同名文件夹,把内层移出来。
依赖只有 requests(ComfyUI 环境一般已自带),装完后完全重启 ComfyUI(关掉控制台窗口重开,不是只刷新浏览器)。
快速开始
- 获取 API Key:https://evolink.ai/dashboard/keys (
sk-开头) - 画布空白处双击 → 搜
EvoLink→ 添加 EvoLink GPT Image 2 (Official) - 填
api_key(只需第一次)和prompt(支持中文) - 从
image输出口拖线 → 选「预览图像」或「保存图像」(必连,否则看不到图) - 点运行,约 35 秒出图(quality=low)
图生图(3 步):添加「加载图像」→ 连到 image1 → prompt 写修改要求 → 运行。
最多 16 张参考图:image1~4 口与 image_urls 文本框(每行一个公网 URL)可混用。
现成工作流见 example_workflows/:把 PNG 效果图直接拖进画布即可还原完整工作流(图片内嵌工作流元数据;另附同名 JSON 可用「加载」按钮导入)。
| 示例 | 效果图(拖入即用) |
|---|---|
| 01 文生图 |
|
| 02 图生图(吉卜力风重绘,姿势不变) |
|
| 03 多参考图(把主体放进新场景) |
|
参数说明
| 参数 | 说明 |
|---|---|
| prompt | 画面描述 / 修改要求,支持中文,最长 32000 字符 |
| model | gpt-image-2 / gpt-image-2-beta |
| size | auto(默认)/ 比例 1:1、16:9 等 / 像素 1024x1024 |
| resolution | 1K / 2K / 4K(size 为比例或 auto 时生效) |
| quality | low / medium / high,越高越贵,调试建议 low |
| n | 一次生成几张(1–10) |
| api_key | 首次填写自动存本机,之后留空;也支持环境变量 EVOLINK_API_KEY |
| image1~4 | 参考图口,接「加载图像」即图生图 |
| image_urls | 进阶:已有公网 URL 每行贴一个 |
| timeout_seconds | 最长等待(默认 300s),生成失败或审核拦截的任务不扣费 |
输出:image(结果图,已自动下载落地)、task_id(对账/报障凭证)、status_info(本次任务摘要)、result_urls(结果原始链接每行一个,24 小时过期)、response_json(平台任务接口的完整 JSON 响应,供下游节点解析编排)。所有生成节点输出一致;prompt 等参数框可转换为输入口接线(新版前端直接把上游 STRING 输出拖到参数框上,旧版右键节点 → 转换为输入)。
Key 安全
- Key 等同账户余额,不要发到聊天窗口、截图或公开仓库;
- 分享工作流前清空节点上的
api_key框(留空时 key 从本机evolink_config.json或环境变量读取,不会进入工作流 JSON); - 怀疑泄露立即到 dashboard/keys 删除重建。
计费
按量计费无订阅。每笔任务可在 任务日志 按 task_id 查询状态与扣费;失败/超时任务自动退回积分;参数不变重复运行是 ComfyUI 缓存复用,不发起新请求、不扣费。
从旧版/社区版迁移
| 你在用 | 迁移方法 |
|---|---|
| 旧版官方节点(已下架,带图像输入口) | 用法一致:加载图像连 image1 即可,直接换装本插件 |
| 社区版 torans/ComfyUI_EvoLink | 参数同名(prompt/model/size/resolution/quality/n/image_urls);Evolink Image Input + imgbb key → 不再需要,图直接连 image1;poll_interval/max_polls → 合并为 timeout_seconds。两插件可共存,无需先卸载 |
常见问题
- 搜不到节点:确认插件目录位置正确、没有嵌套两层文件夹、已完全重启;启动日志里搜
EvoLink看有无红色报错。 - 任务完成但看不到图:
image输出没连「预览图像/保存图像」,补连后再运行(走缓存不重复扣费)。 - 401:Key 无效或已删除;402:余额不足;429:请求过频稍后再试。
- 内容审核失败:调整提示词或参考图后重试,失败不扣费。
- 更多见官方教程:https://evolink.ai/docs/cn/integration-guide/comfyui
Roadmap
- [ ] Nano Banana(gemini 系列)/ Seedream 节点
- [ ] 余额查询节点(EvoLink Credits)
- [ ] mask 局部重绘(API 已支持
mask_url,实测后上线) - [ ] 视频模型节点(Veo / Sora / Kling)
License
MIT © 2026 EvoLink | 技术支持:[email protected]