flux1 / sdcpp / editImage

The edit form of the Flux.1 sdcpp node - and the one that finally makes Flux's image-editing abilities reachable without a 24GB model on your disk. Instead of re-rendering from an image with a strength slider (that's the createVariant node), this one takes one or more input images and rewrites them to match your instruction, using Flux's strong natural-language understanding to actually change the scene rather than just restyle it.

The honest use case: you have a picture and you want real edits - remove the background element, change the lighting, replace the subject - and you want them done by the model that understands instructions better than almost anything you can run locally. The cost is that this is a heavier job on the fleet (four-piece stack plus image processing), so it's the most expensive node in the Flux family to run, and it's billed in Buzz per attempt.

How it works

Same Orchestration API client, engine: sdcpp, ecosystem: flux1, operation: editImage. The input images are sent up (as data URLs) with the prompt, and the fleet's Flux stack generates the edited result. Output is a native ComfyUI IMAGE.

The inputs that matter

• prompt (required) - the edit instruction. Flux rewards specificity: "remove the red car from the background and replace it with a tree" beats "make it nicer."
• diffuser_model, vae_model, clip_lmodel, t5_xxlmodel (all required, CIVITAI_AIR) - the four-piece Flux stack via Civitai Model Selector nodes. All four must be populated or the job fails upstream.
• images (optional, IMAGE) - the source(s) to edit. This is optional in the schema because some recipes accept it as a later input, but an edit with no image is just... a generation. Wire in at least one.
• negative_prompt - optional, and Flux barely listens to it; keep it to a short list of artifacts if you use it at all.
• steps / cfg_scale - 28 / 3.5 defaults. Edits usually land fine at 20 steps; only go up if details are mushy.

Install

Pack install: ComfyUI Manager → search "Civitai Comfy Nodes" (publisher civitai) → Install → restart. Or clone https://github.com/civitai/civitai-comfy-nodes.git into custom_nodes, pip install -r civitai-comfy-nodes/requirements.txt (just requests), restart.

Common issues

Credentials, as ever: Civitai Auth node into api_config or the CIVITAI_API_TOKEN env var (from civitai.com/user/account).

The two practical stumbles: (1) forgetting to wire images - the node won't hard-fail, but the "edit" comes back as a fresh generation that ignores your source; and (2) expecting precision edits on low-res input. The edit re-emits the whole frame, and it's strongest when the source is reasonably clean. Feed it a 512px thumbnail and you'll get a 512px-flavored interpretation, not a surgical retouch.

Budget note: edits are the pricey side of this pack. Each attempt is a billable job and a failed instruction doesn't refund Buzz - get the prompt right before you queue quantity > 1.