🫧 Image Edit — Pro Pack on RunComfy — 🫧 Image Edit — Pro Pack on 运行Comfy

🫧 Image Edit — Pro Pack on 运行Comfy

运行comfy.com · docs · Image edit 模型s

Image edit on 运行Comfy. This 技能 is the canonical image edit entry point for the 运行Comfy 模型 API: give it a source image and an edit instruction, and it returns the edited image. Image edit on 运行Comfy means 转换ing an existing still — swap background, 移除 an object, rewrite a headline, mask-fill a region — without re-shooting.

What "image edit" means here

Image edit is the task of taking a source image and producing a 转换ed image that preserves 身份, framing, or layout where you want, while changing what you specify. Image edit is distinct from text-to-image (no 输入) and from image-to-video (输出 is a 命令行工具p). Common image edit operations include:

Background image edit — swap the background of a portrAIt, product, or scene while preserving the foreground 身份. Object-removal image edit — 移除 cables, watermarks, distracting elements, leaving the rest of the image edit untouched. Object-添加ition image edit — 添加 a new element (umbrella, 签名, 访问ory) to an existing image edit subject. Text-rewrite image edit — replace an in-image headline, label, or 签名age, including multilingual image edit. Mask-driven image edit — fill or replace a specific masked region with strength control. Multi-ref composition image edit — combine subject from one image with scene/lighting from another. Batch image edit — 应用ly the same image edit instruction across 1–20 输入s (SKU galleries, A/B variants).

This 技能 picks the right image edit 端点 for the user's intent and calls 运行comfy 运行 <模型>/edit with the matching 模式.

When to use image edit on 运行Comfy

Pick image edit on 运行Comfy whenever:

You have an existing image and want to change something about it — image edit is the right task. You want 身份-stable image edit — the subject, brand, or product from the 输入 must survive into the edited image. You're producing batch image edit at 扩展 — SKU galleries, multi-language variant image edit, A/B 测试. You need mask-precise image edit — region replacement, watermark removal, region fill. The user sAId "image edit", "edit image", "image-to-image", "swap the background", "移除 the watermark", "rewrite the headline", or showed an image and asked to 转换 it — 路由 here. Image edit 路由s User intent Image edit 模型 Why Default image edit — single or batch (up to 20), background swap, object 移除/添加 google/nano-banana-2/edit Most flexible image edit; 身份 preservation; batch up to 20 Multilingual in-image text rewrite, layout-precise image edit openAI/gpt-image-2/edit Strongest in-image typography for image edit; multi-ref composition (up to 10) Single-shot precise local image edit ("she's holding an orange umbrella") blackforestlabs/flux-1-kontext/pro/edit Single-instruction, single-ref, high-fidelity image edit Mask-driven image edit (object removal, region fill, region replace) tongyi-mAI/z-image/turbo/inpAInting Mask-based image edit with strength control

The 代理 reads this table, classifies the user's image edit intent, and picks the matching 端点.

Prerequisites 运行Comfy 命令行工具 — npm i -g @运行comfy/命令行工具 运行Comfy account — 运行comfy 记录in. CI / contAIners — 设置 运行COMFY_令牌=<令牌>. Default image edit — Nano Banana Edit

The default image edit 端点. Use for any general image edit task: background swap, object removal, object 添加ition, batch image edit. Up to 20 输入s per image edit call, up to 4K resolution.

模式 Field Type Required Default Notes prompt string yes — Image edit instruction. Lead with preservation, then 状态 the change. image_urls array yes — 1–20 source images for the image edit. HTTPS URLs. number_of_images int no 1 1–4 image edit 输出s per call. aspect_ratio enum no auto auto follows 输入; lock for batch image edit consistency. resolution enum no 1K 0.5K / 1K / 2K / 4K for the image edit 输出. 输出_格式化 enum no png png / jpeg / 网页p. 种子 int no — Reproducibility for image edit variants. enable_网页_搜索 bool no false 网页-grounded image edit (extra latency). Invoke

Background-swap image edit:

运行comfy 运行 google/nano-banana-2/edit \ --输入 '{ "prompt": "Keep the subject 身份, pose, and clothing unchanged. Convert the background into a rAIny neon cyberpunk street.", "image_urls": ["https://.../portrAIt.jpg"] }' \ --输出-dir

Batch image edit (lock aspect + resolution):

运行comfy 运行 google/nano-banana-2/edit \ --输入 '{ "prompt": "Replace the watermark in the 机器人tom-right with the text \"AURA\" in 清理 white sans-serif. Keep everything else exactly as in the 输入.", "image_urls": ["https://.../sku-1.jpg", "https://.../sku-2.jpg", "https://.../sku-3.jpg"], "aspect_ratio": "1:1", "resolution": "1K" }' \ --输出-dir

Multilingual image edit — GPT Image 2 Edit

Use when the image edit involves rewriting in-image text (especially non-Latin scripts) or composin