flux2Dev / comfy / editImage

Flux 2 does something the FLUX.1 era could only dream about: generation and editing in one model. Civitai Image Gen (flux2Dev / comfy / editImage) is the editing half - you feed it an image, tell it what to change in plain language, and it rebuilds the picture to match. On the cloud, so the 56B of weights behind it never touch your machine.

What it is

The edit operation for the Flux 2 Dev ecosystem. Required input is just prompt - there's no separate instruction box, and notably no denoise_strength dial, unlike the pack's older createVariant nodes. Flux 2's edit interface is prompt-native: your text is the instruction, and the model figures out the rest. The source goes in via the optional images socket (IMAGE).

Inputs that matter

• prompt - the instruction. "Change the background to a rainy city street" or "make the subject wear a red jacket." This is the entire interface; the model's instruction-following is the reason people rate it the best local editor.
• images - optional IMAGE input for the source(s). The KB notes Flux 2 handles single- and multi-reference editing in one checkpoint, which is the part most people don't expect.
• width / height (512–2048, default 1024), steps (4–50, default 20), cfg_scale (1–20, default 1.0), sampler/scheduler, negative_prompt, loras, quantity (1–4), seed, output_format, image_metadata, api_config.

Outputs: images, errors, workflow_id, raw_json.

Why it matters

The KB's flux-2 essay has a striking stat: people who do serious editing still call Flux 2 Dev "the best local edit model by far," better than Qwen-Image-Edit and without Klein's anatomy mistakes - it even handles character sheets. The reason the community still said "too heavy to use" is the 53GB fp8 footprint. This node is the unlock: the editing quality without the hardware. Wire a LoadImage in, describe the change, and the edited image comes back as a native IMAGE ready for upscaling or compositing.

That also tells you the tradeoff: local editors like Klein's 4B or Qwen-Image-Edit cost nothing per call and run in seconds on a modest card. This node's pitch is specifically "I want the best, and I don't have a 48GB card." For occasional edits that's a fine trade; for daily iteration it's a metered habit.

Install & gotchas

Standard pack install - Manager (search Civitai Comfy Nodes), comfy node registry-install civitai-comfy-nodes, or git clone. Auth via env var / OAuth / Civitai Auth node.

The things worth knowing before you queue:

1. Your source images upload to Civitai's fleet. The edit input goes with the job. For licensed or sensitive reference material, that's the deciding factor, not the Buzz.
2. The prompt is the edit. There's no strength slider to rescue a vague instruction. Be specific about what changes and what must stay - that's the difference between a targeted edit and an unpredictable re-render.
3. quantity caps at 4 and each output is billed. Plan the batch before you queue, and check raw_json to see what a session actually cost.

The honest summary: if you want Flux 2's editing ability and you've never been able to run it, this is the node you've been looking for. Just budget like it's a service, because it is.