Image Bench

Image Bench for ComfyUI

Image Bench is a generation workspace for ComfyUI, laid out so that the frequent steps of image generation take as few actions as possible. It runs in its own window and drives ComfyUI, which does the generation. Every image ComfyUI generates arrives in Image Bench with its parameters, and you keep the best ones in trays, where the common follow-up tasks are automated. For example, upscaling an image takes just one button click, with no need to look up its seed or other parameters, and an image goes to a graphics editor and back in one click, without copying and pasting. You do not need the file manager or an image browser at any point. See more at the extension site.

The full manual is built in: Library › Manual in the Image Bench window. If you are new to Image Bench, start with the tutorial, which walks through the common workflows with screenshots.

Note: This add-on is not designed to make ComfyUI less complicated. It automates repetitive tasks, not the creative process.

Prompting skill

Image Bench also has a Claude skill, diffusion-prompting. It writes, fixes and converts prompts and suggests sampler settings for the models described in the Library tab (SDXL, FLUX, Qwen-Image, Z-Image, WAN 2.2 and others), using the Image Bench model docs as its references. Download diffusion-prompting.zip and upload it to claude.ai (Settings › Capabilities › Skills) or unpack it into ~/.claude/skills/. It may work with other agents that support skills.

Installation

With ComfyUI-Manager:

1. Open Manager › Install via Git URL.
2. Enter https://github.com/GChristensen/comfyui-image-bench.
3. Restart ComfyUI.

Manager accepts a Git URL only with allow_git_url_install = true in the [default] section of its config.ini and ComfyUI listening on a loopback address (--listen 127.0.0.1, the default).

Or clone it by hand into ComfyUI's custom_nodes folder:

cd <ComfyUI>/custom_nodes
git clone https://github.com/GChristensen/comfyui-image-bench

Then restart ComfyUI: it loads extensions and the Image Bench Prompt node only at startup. Updates come from Manager (Update), or git pull in the extension folder. Updates keep your settings (config.json), the trays and presets in store/ and your node additions in store/node_config.json, which are not part of the repository.

Optional node packs:

| Package | Nodes | Needed for | |---------|-------|------------| | ComfyUI-Inpaint-CropAndStitch | Inpaint Crop / Inpaint Stitch (Improved) | The bundled *_inpaint workflows, and the mask blur, mask mode, inpaint area, padding and size options of the Generate tab's Inpaint group. Your own inpaint workflows work without it, but the group then hides these options | | comfyui_controlnet_aux | AIO Aux Preprocessor | ControlNet preprocessors (without it the preprocessor controls are hidden), and the ControlNet units of image_anima_txt2img and image_flux1_dev_txt2img | | ComfyUI-Advanced-ControlNet | Advanced ControlNet loader and apply nodes | ControlNet masks (without it Use Mask is hidden), and the ControlNet units of image_anima_txt2img | | ComfyUI-GGUF or ComfyUI-KJNodes | Unet Loader (GGUF), GGUF Loader KJ | .gguf models as the ib:cp loader; not used by the bundled workflows |

The Hires. fix checkbox of the bundled *_txt2img workflows (and video_ltx2_5_full) is a Switch node titled ib:hr. It is a ComfyUI core node, so it needs no node pack, but older ComfyUI versions do not have it: if it shows as missing, update ComfyUI. With a workflow open, Manager › Install Missing Custom Nodes installs the packs it needs.

Opening Image Bench

Image Bench could run in a standalone mode which loads ComfyUI in background, or be launched from ComfyUI and work along it.

• http://<comfyui address>/image-bench/app opens it without a ComfyUI tab: the window loads ComfyUI itself in a tab of its own.
• The 🔬 Image Bench button in the ComfyUI toolbar could open it as a new tab or a popup window depending on settings. By default, it offers to use the url above.

Preparing a workflow

Image Bench drives the nodes whose titles carry a marker (right-click the node › Title): ib:cp checkpoint loader, ib:pp / ib:np prompts, ib:smp sampler, and optionally ib:smp:1 / ib:hr (Hires. fix), ib:img / ib:mask (img2img, inpaint), ib:ref:N (references), ib:cn:N (ControlNet units), ib:lora (LoRA exploration). For the LoRA list of the Generate tab, use the Image Bench Prompt node as ib:pp. The workflow needs a Save Image or Preview Image node.

Example workflows with every node marked are copied into the Image Bench folder of the saved workflows. The manual's ComfyUI setup chapter describes the markers in full; NODE_CONFIG.md describes how to add custom nodes in store/node_config.json.

Development repository link

This is not a development repository! The full source code is here: https://github.com/GChristensen/image-bench