Fal API Flux Dev Image-to-Image

This is the stripped-down img2img node in the pack - no LoRA slots, no ControlNet, just an image, a prompt, and a strength dial, run against FLUX.1 [dev] on fal's servers. If you don't need a trained style or a control map and just want to push an existing image toward a new prompt, this is the node that does exactly that and nothing more.

The job is done by two inputs: image (required, your starting point) and strength (0–1, default 0.8), which sets how much room the model gets to depart from what you fed it. Push it toward 1 and you're basically letting the prompt take over, with your image acting more like a rough composition guide; pull it toward 0 and the output stays close to the source, with the model nudging details rather than repainting. 0.8 already leans toward "let the model do a lot of work" - if outputs feel unrecognizable from your input, that's the first thing to lower.

Generation controls are otherwise identical to plain text-to-image Dev: prompt (multiline - write it as a sentence, Dev's T5 text encoder is built to parse natural language rather than tag lists), width/height (1024 default, step 8 - this is the canvas fal renders to, independent of your source image's actual size, so mismatched aspect ratios get resized/cropped on fal's end), num_inference_steps (28 default, 1–100), guidance_scale (3.5 default, 0.1–40 - the value most often blamed for Flux's homogenized "same face" look, worth dropping toward 1.8–2.0 if that's biting you on portraits), num_images (1–4 per call, each billed separately), and enable_safety_checker. seed is optional and defaults to 0 for a random draw each run.

Output is a single IMAGE.

Installing it. No special steps beyond the rest of the pack: ComfyUI Manager, search "ComfyUI-Fal-API-Flux," install, restart - or clone https://github.com/yhayano-ponotech/ComfyUI-Fal-API-Flux.git manually into custom_nodes (the README's own manual-install command points at a placeholder repo, so don't copy that URL literally) and pip install -r requirements.txt. The step that actually gates everything: a config.ini file in the pack's folder, [falai] section, api_key = <your fal.ai key>. There's no in-app place to paste it instead - this is a plain text file the node reads on startup, so restart ComfyUI after creating or editing it.

Where this one typically goes sideways. Almost every complaint about img2img nodes in this family traces back to strength being set once and never revisited - if the result looks like a totally different image, it's too high; if it looks untouched, it's too low. Beyond that, the same short list applies as the rest of the pack: a missing or malformed config.ini, a fal.ai account without FLUX.1 access enabled, and checking the ComfyUI console rather than the node's own error text, since that's where fal's actual response gets logged.